WEBVTT
00:00:00.050 --> 00:00:03.839
The work of an AI engineer is really to kind of build those systems.
00:00:04.330 --> 00:00:06.339
That effectively build systems.
00:00:06.794 --> 00:00:22.655
I guess if you're building agentic frameworks and then deploying those agents to effectively write code, and then the code is, for the purposes of building an application, then you're building the systems that effectively build the constituent systems that build the application.
00:00:22.664 --> 00:00:28.434
And in that regard, it's sort of more like industrial engineering than it is necessarily software engineering.
00:00:29.039 --> 00:00:34.950
13% of all pull requests are bot created today and they are creating a unique impact on your SDLC.
00:00:35.240 --> 00:00:56.880
LinearB his upcoming research exposes the effect bots are having on our teams developer experience and productivity, and engineering org who created a system for managing bot generated pRS are able to reduce their entire review load by over 6% while also making drastic improvements in their security and compliance posture you want to learn how your team can manage bot generated PRS and get early access to linear B's data report.
00:00:57.270 --> 00:01:01.649
Head to the show notes to register for our upcoming workshop on September 24th or 25th
00:01:01.649 --> 00:01:03.009
Welcome back to Dev Interrupted everyone.
00:01:03.070 --> 00:01:04.790
I'm your host, Conor Bronsdon.
00:01:04.790 --> 00:01:09.329
And today I'm joined by Amir Babani, a mathematician and AI ML expert.
00:01:09.680 --> 00:01:15.340
Amir is both a founder and the chief AI engineer at Memra, and he's had previous exits to both Google and Meta.
00:01:15.650 --> 00:01:17.239
Amir, thank you so much for joining me today.
00:01:17.700 --> 00:01:18.269
Thank you for having me.
00:01:18.641 --> 00:01:37.751
Yeah, I'm really excited for this conversation because we've talked a lot about how AI is impacting software engineering from the perspective of engineering leaders, but we haven't really had the opportunity to dive in depth on what's happening and, uh, as far as the forefront of it and kind of what we see moving forward with someone who's a deep expert in the research side.
00:01:37.900 --> 00:01:50.990
So in today's conversation, we're going to have that chance to zero into the future of AI and software engineering and hear from you about Kind of the cutting edge of research, uh, plus talk about agentic AI and how that may transform software workflows or other workflows in the world.
00:01:51.411 --> 00:01:53.700
But before we jump in, I do want to remind our listeners.
00:01:53.751 --> 00:01:56.361
I know we say it every time, but it really does matter.
00:01:56.400 --> 00:02:02.260
If you enjoy this episode, please just take one moment to rate and review the podcast on your X.
00:02:02.456 --> 00:02:10.536
App of choice, Stitcher, Google Podcasts, Spotify, whichever it is, it helps us bring more insightful conversations with leaders like Amir.
00:02:10.915 --> 00:02:15.265
And if you really love it, uh, tweet it or zeet it, uh, post on LinkedIn.
00:02:15.276 --> 00:02:16.235
We'd love to hear from you.
00:02:16.430 --> 00:02:19.569
but with no further ado, Amir, let's dive in.
00:02:19.659 --> 00:02:23.129
Agentic AI is rapidly redefining how we interact with technology.
00:02:23.495 --> 00:02:34.544
Right now, can you start by explaining for our audience what agentic AI means, just to make sure we're all on the same page and how it differs from prior concepts folks may have heard of such as RPA or robotic process automation.
00:02:35.375 --> 00:02:41.085
I think of agentic AI as Blocks of code that have reasoning capability.
00:02:41.555 --> 00:02:43.495
So they actually have access to an LLM.
00:02:43.544 --> 00:02:44.485
That's the first thing.
00:02:44.925 --> 00:02:47.585
Then they have access to long term memory.
00:02:47.645 --> 00:02:57.465
And you can think of that as akin to the RAG layer, so that would be the vector databases plus the graph databases plus other long term, forms of memory.
00:02:57.775 --> 00:03:02.638
Then they have access to adapt, uh, sort of short term memory that, increases their adaptability.
00:03:03.038 --> 00:03:06.698
And then they have access to tasks on which they're trained.
00:03:07.091 --> 00:03:10.010
And then finally they have access to some sort of integration layer.
00:03:10.050 --> 00:03:14.010
They integrate into enterprise applications or other workflows.
00:03:14.911 --> 00:03:23.121
And in so doing, unlike, let's say, RPA, because I often get the question, like, how does, how does this, technology contrast with, RPA?
00:03:24.127 --> 00:03:26.897
agents are interacting with knowledge.
00:03:27.229 --> 00:03:38.742
As context, as opposed to sort of data inputs, and they can sort of reason their way to a conclusion when they're interacting in so far as these workflows are concerned.
00:03:38.983 --> 00:03:45.872
So for example, you can say, write these documents to a database that begs a whole suite of questions.
00:03:46.323 --> 00:03:51.490
What fields in the database, how does the fields in the database map to this particular document?
00:03:51.520 --> 00:03:54.099
What do you want me to extract from this document and write to the database?
00:03:54.389 --> 00:03:55.199
Or database.
00:03:55.754 --> 00:03:57.634
What's the path of a base, etc.
00:03:58.055 --> 00:04:14.389
So the agents, have the ability, on the fly to reason through that imperative and get to an outcome that was so desired by the, by the user who's sort of, you know, Commanding them to, do that particular job.
00:04:14.429 --> 00:04:17.836
I
00:04:18.690 --> 00:04:21.430
deterministic code essentially where, Hey, we.
00:04:21.786 --> 00:04:24.096
We expect the output to remain the same.
00:04:24.396 --> 00:04:36.165
And this more non deterministic model of, of an agent or typically non deterministic, you get so much more capability as far as reasoning, but the array of results you may receive is more varied.
00:04:36.185 --> 00:04:47.586
And this becomes particularly interesting for folks in software engineering, because, uh, as we've seen at LinearB, we've just released a 2024 bot automation research study that found that more than 13 percent of pull requests.
00:04:47.685 --> 00:04:49.365
Today are already bot created.
00:04:49.795 --> 00:04:57.826
And we expect that number to increase with this continued innovation in AI, more happening on the agentic AI side of things.
00:04:58.055 --> 00:05:00.105
How rapidly do you expect to see.
00:05:00.531 --> 00:05:08.060
Fully AI, you know, non deterministic models, uh, come through and actually commit code to a code base.
00:05:08.706 --> 00:05:10.262
think it's happening as we speak.
00:05:10.262 --> 00:05:12.483
I've tried a myriad of these tools myself.
00:05:12.908 --> 00:05:14.798
And these, um, these other frameworks.
00:05:14.901 --> 00:05:18.271
I'm both a consumer of these things and a producer and I enjoy using them.
00:05:18.271 --> 00:05:19.670
They're, they're actually quite fun.
00:05:20.218 --> 00:05:38.404
I think in a lot of cases, there's a lot of work that is sort of needed for these things to fully automate, the development and the deployment of large scale applications, however, the pace at which that work and sort of that innovation is taking place.
00:05:38.773 --> 00:05:49.470
I don't see it too far in the distant future when, at least small scale, fully functioning applications are developed and deployed through a single prompt.
00:05:49.730 --> 00:05:58.819
With a bunch of the reasoning and sort of the Socraticisms that happen between the single prompt and the final outcome taking place.
00:05:58.894 --> 00:06:01.627
in between the command and the outcome.
00:06:01.985 --> 00:06:03.134
So I do see that taking place.
00:06:03.175 --> 00:06:07.564
And actually that, that speaks to, well, how do you maintain context?
00:06:08.485 --> 00:06:15.028
How do you actually break down a particular complex function into constituent sets of tasks?
00:06:15.350 --> 00:06:19.014
As you would maybe, MapReduce for a workflow.
00:06:19.254 --> 00:06:24.211
and for example, what is the right network topology to maintain context?
00:06:24.473 --> 00:06:25.733
And to get a job done.
00:06:25.843 --> 00:06:28.392
So, for example, there are different types of network topologies.
00:06:28.682 --> 00:06:31.523
A bucket brigade is a very simple network topology.
00:06:31.543 --> 00:06:36.543
There's a house burning here, there's a lake here, and you just move the buckets in a linear fashion.
00:06:37.233 --> 00:06:43.952
If you're disseminating orders in sort of a complex organization, you want that topology to be hierarchical.
00:06:44.923 --> 00:06:52.403
If access to information is necessary, sort of in near real time, you kind of want to have a flat or heterarchical topology.
00:06:52.702 --> 00:07:13.812
So identifying a given problem and mapping it to the right network topology so that you can break the tasks down into sort of a constituent set of subtasks and then have agents that are trained on specific subtasks allocated to those subtasks and then have let's say a master agent as a reducer summing the sums.
00:07:14.319 --> 00:07:35.009
Okay, and so if you kind of pull that together, and that's sort of like what's missing in a lot of these application or agentic frameworks that I've tried thus far as a consumer, if that can be pulled off, yes, I see that in a not so distant future, you can sort of command, development and deployment of at least small scale applications.
00:07:35.387 --> 00:07:43.377
So my kind of mental framework for this is that we're seeing this shift from something that I think a lot of us have done for a long time, which is.
00:07:43.682 --> 00:07:59.353
Hey, we're going to automate away repetitive tasks to not only are we going to automate away repetitive tasks, but we are going to essentially create agentic employees who are working on your behalf and where you have to really think about the inputs you're giving to those employees.
00:07:59.452 --> 00:08:03.343
And maybe there's less of the people management, but the information management becomes even more important.
00:08:03.932 --> 00:08:05.983
Is that an accurate framework to apply here?
00:08:05.983 --> 00:08:06.201
There's
00:08:06.350 --> 00:08:17.264
I think if, if you look at agents as they stand right now, without mentioning any specific agentic frameworks, but if you just, I mean, there's, there's some well known ones out there.
00:08:17.737 --> 00:08:22.497
what they do is they reference a function as a tool.
00:08:22.855 --> 00:08:23.954
And then that tool.
00:08:25.023 --> 00:08:31.096
with several other parameters is allocated to a particular block of code.
00:08:31.096 --> 00:08:44.495
So the block of code, as I said, has an LLM, it can have even an LLM router, it will have some sort of memory, and it will have access to this tool, and that tool is making a effectively making a function call, or it's having access to a function.
00:08:44.817 --> 00:08:56.477
Now, if you think of these functions as methods that belong to a class, And if you can sort of further abstract that the classes sort of fit into larger workflows.
00:08:56.784 --> 00:09:16.780
then you can, and this is actually what we're doing, so I'm kind of speaking from, from experience here, we're registering into memory of a master agent, not a single function to which it has access through a tool, but whole classes and methods that map to a particular workflow.
00:09:16.990 --> 00:09:23.071
So effectively, you've taken, a master agent and you've trained it on a set of workflows.
00:09:23.880 --> 00:09:38.197
Well, if you think about that in the context of workers, a human has a role, that role maps to a job, that job maps to a set of workflows, those workflows map to sets of constituent tasks.
00:09:38.909 --> 00:09:59.490
And right now, I think what we're saying is that, if we can take a workflow and break it down into constituent tasks and allocate agents to those tasks and complete those workflows, then effectively those jobs are being done by the agents, which then means these agents have a role to play within an organization.
00:09:59.846 --> 00:10:04.745
And in that case, I would say that these agents are effectively digital employees.
00:10:04.975 --> 00:10:15.543
And those digital employees paired with a human employee, say an individual contributor, suddenly that individual contributor has multiplicative returns to scale.
00:10:15.716 --> 00:10:17.706
As does, um, hiring a manager.
00:10:17.745 --> 00:10:22.505
If you hire a manager, perhaps, yes, indeed, perhaps, yeah.
00:10:22.596 --> 00:10:24.716
Even, and, and actually think of it from this perspective.
00:10:24.716 --> 00:10:30.576
If you have the employee right now using, let's say, ChatGPT, there's marginal improvement on their productivity.
00:10:31.066 --> 00:10:49.070
But if the human is taken out of the loop, because that human is maintaining context through that, through the dialogue with the GPT, sort of that's they're sitting around maintaining context, engaging, winnowing down on, a set of conversations and therein the outcomes of those conversations.
00:10:49.720 --> 00:10:58.080
But if the agent is doing that, then that's, that's a much higher return to scale if that human is sort of out of the loop.
00:10:58.139 --> 00:11:01.125
Now the human can be on top of the loop, so to speak.
00:11:01.625 --> 00:11:05.686
And on top of the loop means they're engaging in an imperative.
00:11:06.385 --> 00:11:07.456
Go do this.
00:11:07.875 --> 00:11:11.316
And they're not sitting in the middle, engaging in the interrogative.
00:11:12.503 --> 00:11:15.278
agents, are engaging in the interrogative.
00:11:15.477 --> 00:11:22.908
And that's what I mean by there's a Socraticism that happens between the command to go do something and the final output.
00:11:22.908 --> 00:11:25.493
At
00:11:25.878 --> 00:11:42.057
I mean, it still maintains some of the, I'd say, black box concerns folks have at the model level, but because of the way you're constructing these agents and how they are approaching things, it sounds like there's a much better auditing process for what's actually happening there.
00:11:42.457 --> 00:11:50.307
That then the human employees can focus on, okay, let's audit to make sure this is working and that we're actually delivering on these processes.
00:11:50.567 --> 00:12:03.248
Whereas I'll say currently, one of the big things we noticed in that, uh, bot report I mentioned was that 96 percent of all basic bot created PRs that are being sent out right now are not linked to product management tools.
00:12:03.557 --> 00:12:17.192
Well, if I have an agent doing that, I presume we can simply make sure the agent is using a project management tool to track this and kind of improve the observability of a lot of this code that's coming into the code base and a lot of the approaches we're taking here.
00:12:17.544 --> 00:12:32.341
present moment, we're able to Um, the steps that an agent takes to complete tasks, and in fact, um, we're able to cache and, and sort of have a, an auditable record of the reasoning process.
00:12:32.754 --> 00:12:36.034
And so it's not at the workflow level.
00:12:36.034 --> 00:12:37.955
It's not fully black box.
00:12:38.024 --> 00:12:45.764
Now, one could argue that again, these agents have access to LLMs and the LLMs have sort of deep learning models.
00:12:46.240 --> 00:12:52.879
And maybe there's some black box models at the root of sort of the language models.
00:12:52.909 --> 00:13:06.932
That's, that's definitely, I mean, that's probably a case, I mean, when I think about older, sort of like the last decade where we're building, machine learning models, some of those non parametric models, they're black box.
00:13:08.251 --> 00:13:09.741
It's really hard to interpret.
00:13:10.110 --> 00:13:22.384
what's actually going on and, and that's different from the parametric models where you have sort of, you know, a prior assumption as to the distributions, of the fields that comprise the training set.
00:13:22.631 --> 00:13:31.192
and I'm not sure in all cases, the black box nature of those models was necessarily an issue.
00:13:31.201 --> 00:13:31.211
Uh
00:13:31.355 --> 00:13:32.035
I hear you.
00:13:32.186 --> 00:13:45.895
It's, it's interesting though, because I think a lot of, you know, survey data will, will show that the concerns that you, when you talk to a, you know, CISO or just a general software engineer, is they're saying, Hey, I don't get what's happening here.
00:13:46.296 --> 00:13:49.615
Uh, I mean, we can also talk about the concerns of, is this going to replace me?
00:13:49.645 --> 00:14:02.956
I think we will get to that in the labor economics side of this, but what would you say to those folks who are just worried about implementing this and having, not being able to unpack when something goes wrong or, uh, other concerns on that kind of black box piece?
00:14:03.192 --> 00:14:03.451
Do
00:14:03.672 --> 00:14:22.479
Well, I think if you're, if we're talking about agents, then, and agentic frameworks, then a lot of the reasoning is, There's layers of reasoning that are abstracted above and beyond the deep learning models that are sort of the base of these LLMs.
00:14:22.998 --> 00:14:33.009
that set of reasoning processes, you're able to cache that, you're able to trace it, and you're able to audit it.
00:14:33.759 --> 00:14:55.206
So at that level, I think, um, and generally as I, as I think about this, the worker, maybe the, the, the software engineer or whatnot, they would want access to that layer of reasoning, I'm not sure they necessarily need access to, um, reasoning that's happening, so to speak, or the inferencing that's happening, so to speak, at the LLM layer.
00:14:55.860 --> 00:15:07.636
you think software engineering is going to be one of the disciplines that particularly benefits from this increasing use of agentic AI and its ability to autonomously handle tasks?
00:15:08.024 --> 00:15:08.543
I think so.
00:15:08.543 --> 00:15:20.739
I think that software engineering is going to benefit as other disciplines, knowledge work disciplines, I think it's going to benefit from sort of an An industrial engineering paradigm, right?
00:15:20.739 --> 00:15:35.182
So if you go back in time, you have, let's say factories with seamstresses were designing and fabricating sewing effectively clothing, but the industrial revolution came along and said, okay, we're going to design.
00:15:35.741 --> 00:15:43.001
The factories and the systems that output those articles of clothing and software engineers.
00:15:43.011 --> 00:15:51.629
Similarly, we're, in fact, arguably in some regard, the work of an AI engineer is really to kind of build those systems.
00:15:52.119 --> 00:15:54.129
That effectively build systems.
00:15:55.134 --> 00:16:12.294
I guess if you're building agentic frameworks and then deploying those agents to effectively write code, and then the code is, um, for the purposes of building an application, then you're building the systems that effectively build the, the constituent systems that build the application.
00:16:12.304 --> 00:16:18.075
And in that regard, it's sort of more like industrial engineering than it is necessarily software engineering.
00:16:18.389 --> 00:16:23.690
Or I would almost describe it as more of an architect mindset versus a carpenter mindset.
00:16:23.750 --> 00:16:30.203
Um, and we've already talked about that kind of, mean, plenty of people have an architect title already, and I think we're going to see a push more towards that.
00:16:30.374 --> 00:16:37.634
what are, what are the other industries that you expect to see a lot of changes, uh, coming due to agentic AI improvements?
00:16:37.996 --> 00:16:47.403
Well, I think in terms of roles within the knowledge work, the arena of knowledge work, I, I envision these agents climbing the career ladder.
00:16:47.549 --> 00:16:51.504
So initially they can do some data entry, then they can book meetings.
00:16:52.063 --> 00:17:00.604
You know, then they can send emails out and sort of monitor, maybe drip campaign and then eventually they kind of just go up this career ladder.
00:17:00.634 --> 00:17:04.503
They progress, um, maybe they write some basic applications.
00:17:04.804 --> 00:17:09.503
I do think it's a, we're, we're far from the agents necessarily solving for product market fit.
00:17:09.686 --> 00:17:12.576
Or doing sort of more, you know, or maybe designing
00:17:12.676 --> 00:17:12.916
Good.
00:17:12.916 --> 00:17:13.866
I need to keep my job.
00:17:13.866 --> 00:17:14.116
So,
00:17:14.920 --> 00:17:18.227
I think sort of the more, you know, the complex and creative.
00:17:18.653 --> 00:17:27.339
job functions, as you say, for example, architecture, it's not necessarily, that's a layer abstracted away from doing specific tasks.
00:17:27.579 --> 00:17:36.054
So in that regard, I see, a whole myriad of especially task based work being disrupted.
00:17:36.054 --> 00:17:50.155
I Something I'm realizing when I interact with companies is that there seems to be that of three or four stages prior to even engaging, but each of those stages, the customers are willing to pay for.
00:17:50.516 --> 00:17:59.199
And the first is enterprises they want people to explain to them how AI fits into everything.
00:17:59.991 --> 00:18:07.058
They don't have, let's say, necessarily, an immediate answer to the use case.
00:18:07.737 --> 00:18:10.616
Within the organization that can benefit from AI.
00:18:11.156 --> 00:18:16.247
And I actually think that type of reasoning is this design of experiment reasoning.
00:18:16.896 --> 00:18:17.126
Hmm.
00:18:17.402 --> 00:18:21.721
And you learn about that in human factors engineering, you learn about it in research methods.
00:18:22.321 --> 00:18:28.751
Certainly if you're doing machine learning work, you have to sort of say, okay, what is the use case?
00:18:28.751 --> 00:18:34.071
But you're reducing that use case down to a single variable called the target variable or the dependent variable.
00:18:34.645 --> 00:18:39.746
And you're aligning a bunch of data that you have to sort of bear results for that target variable.
00:18:39.836 --> 00:18:43.695
And that's a, that's a way of thinking about the world.
00:18:44.118 --> 00:18:47.169
And that is sort of like an experimental form of reasoning.
00:18:47.361 --> 00:18:58.527
And that lends itself to kind of going into an organization and figuring out how things are functioning and then figuring out how AI can benefit those systems and processes.
00:18:59.250 --> 00:19:02.090
And right now I notice that companies just want help with that.
00:19:02.513 --> 00:19:11.874
And some subset of those companies, I think, as a sort of an emerging AI startup, those are your pilots.
00:19:12.406 --> 00:19:15.176
And then eventually those are your MSAs.
00:19:16.568 --> 00:19:20.048
I think Bain Capital put out a report recently that I thought was interesting.
00:19:20.251 --> 00:19:25.538
showed some hurdles that have been overcome in so far as AI adoption.
00:19:25.568 --> 00:19:27.838
And some of those hurdles were InfoSec related,
00:19:28.171 --> 00:19:28.471
Hmm.
00:19:28.549 --> 00:19:29.296
It
00:19:29.445 --> 00:19:36.278
data private, but the other hurdles that have yet to be overcome, sort of like, what's the use case?
00:19:36.499 --> 00:19:38.108
It's like kind of a classic thing.
00:19:38.118 --> 00:19:39.098
Like what's the use case?
00:19:39.098 --> 00:19:40.269
How do we apply this?
00:19:40.868 --> 00:19:42.348
How do we wrap our minds around this?
00:19:43.084 --> 00:20:03.570
And it seems like the role of the AI engineer right now is really that, and then eventually that translates into ingesting data, ingesting content, building systems, analyzing, sort of the output, based on the output, where's the refinement, and then based on all of that, can you get agents to kind of automate processes.
00:20:04.675 --> 00:20:08.726
um, there's a lot of work right now to be done just informationally.
00:20:08.726 --> 00:20:10.435
The work doesn't stop there, by the way.
00:20:10.945 --> 00:20:13.046
Um, and, and it really shouldn't just stop there.
00:20:13.066 --> 00:20:18.746
That's not, that's not the end all be all, but it's, um, it's certainly a, it's certainly a need right now,
00:20:19.041 --> 00:20:31.971
feels like part of what's happening is we know there's so much potential to apply AI, uh, and there's so many opportunities that zeroing in on where to start and kind of the steps to take along the path is a challenge for a lot of folks right now.
00:20:32.352 --> 00:20:32.872
precisely.
00:20:32.872 --> 00:20:36.362
One way I would think about this design experiment reasoning is that.
00:20:36.886 --> 00:20:42.186
You've identified the why and you can distill that through the what into the how.
00:20:42.621 --> 00:20:49.221
there's a lot of, um, content out there that, okay, the why is getting to the point of being self explanatory.
00:20:49.431 --> 00:20:51.250
How do you distill that down to the how?
00:20:51.419 --> 00:21:03.909
In the short run, you've noted that AI will encourage engineers to think more Socratically, to ask more questions, to really elicit more, what does that mean for day to day development now?
00:21:03.949 --> 00:21:07.229
And then I'm curious where you think it's going to go in the future.
00:21:15.597 --> 00:21:18.438
by engineers specifically.
00:21:18.478 --> 00:21:20.034
Um, I don't know.
00:21:20.074 --> 00:21:26.621
I mean, I, I think, more specifically, the agents themselves will take on that Socratic reasoning.
00:21:27.017 --> 00:21:27.196
And
00:21:27.281 --> 00:21:29.821
you think the shift is already happening to a more.
00:21:30.112 --> 00:21:35.172
Managerial mindset for engineers where it's like, Hey, I'm going to manage these agents.
00:21:35.392 --> 00:21:40.362
I need to make sure they're, they're functioning well versus let me elicit info from them.
00:21:40.825 --> 00:21:46.525
I think that the shift is happening where the human being commands the agents.
00:21:46.904 --> 00:21:47.795
That's the imperative.
00:21:48.633 --> 00:21:57.546
And then the reasoning that needs to happen between the imperative and the completed task, no pun intended, it begs a lot of questions.
00:21:58.214 --> 00:22:03.355
And it's, it's that interrogative process that I sort of think of as like the Socratic dialogue.
00:22:03.789 --> 00:22:04.079
Got it.
00:22:04.089 --> 00:22:07.970
And then obviously on the more, I'll call it individual level.
00:22:08.009 --> 00:22:13.250
We're all engaging in that Socratic dialogue when we simply use chat2BT to like rewrite an email for
00:22:13.309 --> 00:22:14.200
Precisely.
00:22:14.442 --> 00:22:19.303
And when we're engaging in that Socratic dialogue, we're maintaining context in our minds.
00:22:19.873 --> 00:22:40.382
To sort of guide the conversation towards an outcome, whether it's winnowing down the reasoning process towards something that we want, uh, with increased specificity, that process again depends on these, the access to long term memory, the access to short term memory for adaptability, training these agents on particular tasks.
00:22:40.731 --> 00:22:42.500
Again, there's a difference between knowledge, and knowledge.
00:22:42.741 --> 00:22:43.682
And expertise.
00:22:43.776 --> 00:22:49.185
So there's a knowledge layer here with these agents and there's sort of like a, an expertise layer.
00:22:49.726 --> 00:23:03.791
And for example, you see that, for example, if you want to have a contract written there, you can leverage chat GPT to edit various clauses and you'd probably want someone with expertise to just look it over, see if it, is cogent.
00:23:04.000 --> 00:23:08.557
that is sort of where a lot of this modeling context comes into play.
00:23:08.557 --> 00:23:17.637
And all of that modeling context, as I see it, is a bus between sort of the foundational layer and the, and the application layer.
00:23:17.647 --> 00:23:18.678
And it's, it's really.
00:23:19.548 --> 00:23:31.127
All of us right now with regard to these agentic frameworks, we're sort of just developing the bus that sits between that, the, the CPU analogy and the application layer now.
00:23:32.096 --> 00:23:38.027
Like, I think many folks listening, I am only at this point using kind of the Socratic approach.
00:23:38.027 --> 00:23:39.967
I'm leveraging maybe Copilot.
00:23:39.977 --> 00:23:43.686
Maybe I'm just leveraging ChatGPT to help me with certain test writing tasks.
00:23:43.936 --> 00:23:51.297
I'm not yet leveraging agents, but I see that it's going that way and I want to start leveraging agents within my software engineering team.
00:23:51.946 --> 00:23:54.457
What would be your advice to those folks who are listening and saying?
00:23:55.027 --> 00:23:55.307
Damn.
00:23:55.307 --> 00:23:55.666
Okay.
00:23:55.856 --> 00:23:56.557
Here's the next level.
00:23:56.557 --> 00:23:57.436
I need to start moving there.
00:23:58.020 --> 00:24:06.359
one advice is to say, with regard to the labor market transaction, there's two sides and there's a sell side of the labor market transaction.
00:24:06.359 --> 00:24:07.380
There's the buy side.
00:24:07.837 --> 00:24:14.137
it will be, I think, increasingly difficult to compete with agents On the sell side of that labor market transaction.
00:24:14.425 --> 00:24:23.983
So if you're, if you're on the buy side of that transaction, then these agents are great because they've reduced your operational expenditure, et cetera.
00:24:24.541 --> 00:24:30.086
so the way I see it that, we as humans will probably leverage these agents.
00:24:30.606 --> 00:24:44.578
To do work and if we can own relational structure, we own the product of the labor and we kind of own the relational structure, then the humans directly benefit from the agents.
00:24:45.398 --> 00:24:54.365
But if you don't have ownership over that relational structure, And you don't have ownership over that work product, then effectively you're competing with the agents.
00:24:55.132 --> 00:24:55.561
Yeah.
00:24:55.612 --> 00:25:04.491
And I think this is where a lot of the fear currently comes in for folks where they hear that and they go, wait, does that mean I need to run my own company?
00:25:04.511 --> 00:25:07.501
Does that mean I need to go be an entrepreneur?
00:25:07.741 --> 00:25:13.251
Does that mean I'm going to Not be as strong within the labor market in a couple of years as I am today.
00:25:13.261 --> 00:25:14.582
Like, what do I need to do to change that?
00:25:14.852 --> 00:25:20.521
Um, and obviously AI has extremely broad implications on what's happening with the labor economy.
00:25:21.001 --> 00:25:26.172
How do you see AI agents impacting the broader structure of work and labor contracts in the coming year?
00:25:26.703 --> 00:25:30.013
Well, I think the roles based economy is too rigid.
00:25:30.118 --> 00:25:43.809
So I, I see that companies have roles, and again, like I said, those roles map to jobs, to workflows, to tasks, et cetera, and roles offer a degree of economies of scale for these companies.
00:25:44.049 --> 00:25:47.809
Agents actually, by the way, offer economies of scope.
00:25:48.376 --> 00:25:50.096
Cause the way to think about an agent is.
00:25:50.537 --> 00:25:56.012
You're designing a tooling line that can effectively, manufacture various products.
00:25:56.596 --> 00:26:01.576
To your comparison earlier, I'm building the factory versus I'm building the digital piece on the line.
00:26:01.941 --> 00:26:02.641
Right, right.
00:26:03.170 --> 00:26:17.119
I generally think that if you're building agents into enterprise, into the enterprise, and that inclusive of, of workflows, but, but also sort of, um, culturally and structurally.
00:26:17.970 --> 00:26:22.730
Then you may want to rethink this sort of concept of a role.
00:26:24.400 --> 00:26:40.800
when you're starting your own company, when you're functioning as an entrepreneur, you're not engaging day in, day out at the task level, or at like a single workflow level, or you don't just have this one job that maps to this one role.
00:26:40.810 --> 00:26:45.105
I mean, that's, that's, that's, that's, that's, that's, In my opinion, that's very junior.
00:26:45.444 --> 00:26:51.242
And another way of thinking about it is the roles can be assigned to these agents.
00:26:51.637 --> 00:26:58.407
So if the roles are being imparted to agents, how does that redefine the roles based labor economy?
00:26:58.664 --> 00:27:08.193
And if, if again, you're going back and saying these, these effective workers who can sort of think Socratically, some, in many cases, they're polymaths.
00:27:08.970 --> 00:27:19.164
And if they can go in between, Various functions and the agents are sort of doing the work and why should they sit and just do one role all day long?
00:27:19.454 --> 00:27:20.924
This doesn't make any sense whatsoever.
00:27:21.059 --> 00:27:34.227
So all of that, I think if you're thinking about updating the frameworks of the labor economy, I think you slowly want to start thinking about the rigidity of the roles based labor market.
00:27:34.227 --> 00:27:39.297
I mean, we're not, we're not all sort of sitting on a factory line and sort of.
00:27:39.950 --> 00:27:42.890
at that level in, in sort of the knowledge work economy.
00:27:43.319 --> 00:27:48.825
And yet a lot of how we've built work is based off of that factory model, to your point.
00:27:49.085 --> 00:27:51.434
Uh, which is very early 20th century.
00:27:51.663 --> 00:27:52.151
fascinating.
00:27:52.227 --> 00:28:07.926
I wonder if this is going to exacerbate or kind of remove the trend of like freelance work of gig work that we've seen increasingly over the last, you know, 10 years, do you think that agents are really going to replace freelancers?
00:28:07.926 --> 00:28:14.787
Or are we going to see, uh, this kind of, you know, freelancer gig work approach where, Hey, I'm running agents on behalf of a company now.
00:28:15.211 --> 00:28:36.196
Um, because I, I know, and you've talked about this before that many companies, many large enterprises are taking this insourcing approach and saying, well, if I can build an agent to solve these tasks for me, I'd rather be able to have broader observability and potentially cost savings compared to, say, passing this basic data task off to another country where it's cheaper.
00:28:36.675 --> 00:28:42.113
I think there's a, even, let me go one, one, step more, more broadly.
00:28:42.272 --> 00:28:44.932
I think there's a case to be made for vertical integration.
00:28:45.286 --> 00:28:49.268
I think that these agentic frameworks inherently have a multi layer stack.
00:28:49.897 --> 00:29:21.132
You have the foundational layer, you have the long term memory layer, You have the short term memory for adaptability, you have these agents that are trained on tasks, you have this integration layer, and there are feedback loops that, in fact, there's a two fold feedback loop that just comes, you know, comes to mind right now, and that is to say, if you're allocating these agents to specific workflows within an organization, and they're improving processes, and they're improving processes.
00:29:21.560 --> 00:29:27.330
Okay, there's an improvement in so far as the returns to scale and so far as those process improvements.
00:29:28.016 --> 00:29:40.960
But then also the, the, the outcome of those process improvements and the knowledge derived therein can go back and train the models and effectively train the agents to kind of further that process.
00:29:41.480 --> 00:29:48.650
And it's very, very difficult to have those feedback processes without vertical integration.
00:29:48.936 --> 00:29:50.426
So that's one, that's one point.
00:29:50.457 --> 00:30:07.883
There's a second point here, which is that if you think of this multi layered stack, if you will, Again, and go back, let's say 10 years and think about an analogy to AI, which a very good analogy to AI is sort of like machine learning, set of companies in the early 2010s.
00:30:08.282 --> 00:30:13.813
And those companies were, the ML models were like an afterthought to the use case.
00:30:14.307 --> 00:30:15.337
And the application.
00:30:15.817 --> 00:30:30.106
And so you're sort of building at the application layer and then, yeah, you're, you may have an ML model somewhere and, you know, and that, that ML model is benefiting the application, but you're really thinking at a top down, uh, from a top down perspective.
00:30:30.836 --> 00:30:34.833
With these AI, systems, everything's really happening bottoms up.
00:30:34.942 --> 00:30:38.742
The foundational layer comes about first, then the rag layer.
00:30:39.315 --> 00:30:43.694
Then the short term memory layer, then these agentic frameworks.
00:30:44.194 --> 00:30:51.171
And frankly speaking, the application layer, the AI application layer is sort of maybe still around the corner.
00:30:51.459 --> 00:30:58.818
So if all of that stuff is sort of happening bottoms up, well then that begs two interesting questions.
00:30:58.878 --> 00:31:14.586
One is, Where are these AI first companies with vertical go to markets, and then how is that different from an existing incumbent, and that's redundant, but an incumbent at the present moment?
00:31:15.645 --> 00:31:26.174
Acquiring sort of the AI capabilities because they have already distribution and they have sort of monopoly pricing power over those distribution channels.
00:31:26.855 --> 00:31:32.484
And even a vertical go to market AI first company, they have to solve for distribution.
00:31:33.224 --> 00:31:37.099
And arguably they have to solve for distribution faster than the incumbent solves for innovation.
00:31:37.481 --> 00:31:38.991
So there's like a difference of race problem.
00:31:39.521 --> 00:31:44.402
So maybe in the short run you're going into these large enterprises and agentifying them.
00:31:44.832 --> 00:31:52.467
But you have to do that vertically, or at least it would seem that there's a case to be made for vertical integrations that you get those feedback loops.
00:31:52.856 --> 00:31:53.696
That's one thought.
00:31:53.787 --> 00:31:59.729
And then the other thought is those vertical AI first go to market companies.
00:32:00.121 --> 00:32:02.810
This whole analogy is very similar to like Blockbuster.
00:32:03.580 --> 00:32:19.685
So there's an argument to be made with a sort of a smarter company going to Blockbuster and saying, Hey, we're going to, bring something analogous to Netflix, but have, have that inside of Blockbuster and sort of bring Blockbuster up to code for the current decade.
00:32:20.816 --> 00:32:28.756
And alternatively, there's just the, the paradigm where Netflix disrupts Blockbuster,
00:32:29.502 --> 00:32:36.942
So it sounds like what you're saying is if I'm at the startup level, I need to kind of focus on the innovation side.
00:32:37.002 --> 00:32:39.502
Uh, I don't have that vertical integration opportunity quite yet.
00:32:39.502 --> 00:32:44.492
I can start to move towards it, but I don't have the vast stores of data that, you know, the belt and resources.
00:32:44.731 --> 00:32:51.071
and what I can do is I can, like many startups have in the past, out innovate these enterprises if they're not careful.
00:32:51.386 --> 00:32:56.067
Whereas the enterprise layer needs to look at it and say, Hey, I need to insource.
00:32:56.096 --> 00:33:05.816
I need to, really vertically integrate so that I can take advantage of these, you know, monopolistic prices, uh, these data advantages, these scale economies that I have already.
00:33:06.115 --> 00:33:09.226
Um, and that's kind of how you would think about it from, from those perspectives.
00:33:09.567 --> 00:33:17.209
Yeah, when I think about a startup in general, I think of the stock of a startup having an inherent option premium.
00:33:17.509 --> 00:33:23.239
It's not an explicit options contract, but there is no profit that's distilling down to shareholder value.
00:33:23.308 --> 00:33:25.519
So what is the, what is the equity buyer, i.
00:33:25.538 --> 00:33:25.548
e.
00:33:25.558 --> 00:33:30.019
the investor, Buying the generally buying the option premium.
00:33:30.392 --> 00:33:43.632
And as you solve for product market fit, IE, you solve for distribution as a startup, you're exercising that option premium and converting that option premium to enterprise value or effectively equity value.
00:33:43.662 --> 00:33:50.616
And, and sort of that, that analogy is sort of like potential about potential energy being transferred, transferred into kinetic energy.
00:33:50.944 --> 00:33:58.285
And I don't think, you know, generally, a lot of startups are going to fail because they can't innovate.
00:33:58.295 --> 00:34:01.444
They fail because they can't solve for distribution.
00:34:01.824 --> 00:34:06.473
And meanwhile, the vertical integration strategy says the following.
00:34:07.213 --> 00:34:10.403
The acquiring entity has distribution.
00:34:10.664 --> 00:34:36.900
Now, interestingly, there's a There's something called a reverse acquihire now, and that's an interesting form of vertical integration where the larger company, let's say the incumbent, is not really buying your stock, they're actually employing the founders and signing a non exclusive licensing deal for the underlying AI technology, and exclusive share.
00:34:37.351 --> 00:34:39.510
Licensing deal in U.
00:34:39.510 --> 00:34:39.650
S.
00:34:39.670 --> 00:34:51.139
copyright law is a transfer of IP, but a non exclusive licensing deal allows your product still to exist, and you can kind of have non exclusive licensing with many different, uh, vendors.
00:34:51.340 --> 00:35:01.585
But here, there's an interesting hybrid, because the founders are sort of being employed to Work at the incumbent and sort of build that AI sort of from the inside.
00:35:02.378 --> 00:35:12.028
And that's a form of vertical integration that, all of this is really kind of interesting in the short run until those AI first vertical go to market companies come about.
00:35:12.759 --> 00:35:12.858
Yeah.
00:35:12.869 --> 00:35:14.429
And that's certainly coming.
00:35:14.777 --> 00:35:27.387
So we've been talking about this from kind of the, the company perspective, the company's strategy side, but we've also spoken a little bit about the individual side and trying to make that shift to, Hey, I'm managing these agents.
00:35:27.398 --> 00:35:31.878
Um, I'm helping them, uh, versus, Hey, I'm individually executing tasks.
00:35:32.657 --> 00:35:42.427
How do you think folks who aren't on that founder level, uh, but are maybe You know, an engineering manager listening to this, who, who listened to the show thinking, Hey, I'm trying to improve in my career.
00:35:42.427 --> 00:35:43.338
I'm trying to grow.
00:35:43.722 --> 00:35:55.472
Or even like an individual engineer right now, maybe a junior or senior engineer who has been thinking, Hey, I want to move into engineering management, but I am, you know, maybe I'm worried now that, you know, is my role going to go away?
00:35:55.472 --> 00:35:56.592
Am I going to have this opportunity?
00:35:56.762 --> 00:36:00.242
How should they be thinking about their approach to the labor market?
00:36:00.503 --> 00:36:05.617
You know, now in the next couple of years, as this AI transformation wave continues to rock through us,
00:36:05.699 --> 00:36:10.927
Um, there are two ways that just, I think, in the short run, They come to mind.
00:36:11.168 --> 00:36:22.818
One is that you leverage the AI agents or just the AI frameworks for marginal improvement over your current productivity.
00:36:23.054 --> 00:36:28.014
So you can get maybe 20, 30 percent improvement over otherwise.
00:36:28.130 --> 00:36:33.980
which I think where a lot of us are today where it's like, Hey, I'm, I'm leveraging Claude or cursor or whatever else to help
00:36:34.034 --> 00:36:35.094
That that's right.
00:36:35.445 --> 00:36:46.567
The other way, again, in the short run, that comes to mind is effectively designing these AI systems that effectively that which you're tasked to work on.
00:36:46.666 --> 00:36:52.262
So it's building the meta models as opposed to building the models, building the, the, the systems that build systems.
00:36:53.413 --> 00:37:01.413
And that is at least right now, not something with which, these AI agents.
00:37:02.242 --> 00:37:03.833
Are, um, best suited.
00:37:04.043 --> 00:37:06.193
Maybe that'll change in short order.
00:37:06.193 --> 00:37:12.864
I mean, things are moving very, very fast right now, but those are the two ways I would say, to the, the person whom you mentioned.
00:37:13.253 --> 00:37:14.623
It is certainly moving really fast.
00:37:14.634 --> 00:37:16.804
We've got to make sure we get this episode out quickly so that not too
00:37:16.853 --> 00:37:16.873
yeah,
00:37:16.994 --> 00:37:18.884
changes before we go live.
00:37:19.454 --> 00:37:23.463
I know another thing that I've heard you talk about is this idea of a synthetic marketplace.
00:37:23.659 --> 00:37:28.798
I'd love for you to expand on that concept as I'm starting to increasingly see that as kind of where we're going.
00:37:28.869 --> 00:37:32.969
Well, you think of two sided marketplaces, so Uber.
00:37:33.688 --> 00:37:35.699
Airbnb, etc.
00:37:35.871 --> 00:37:37.561
What is a two sided marketplace?
00:37:37.572 --> 00:37:40.326
It's a, it's a, there are two demand curves.
00:37:40.639 --> 00:37:47.018
So in the case of Uber, there's a demand function for passengers and there's a demand function for the drivers.
00:37:47.295 --> 00:37:59.425
And there's an interesting, recursive cross product price elasticity of demand that defines how those two demand curves interact with one another, or I'm sorry, interact with each other.
00:38:00.434 --> 00:38:04.440
An example of this would be Adobe Acrobat, Reader and Writer.
00:38:04.646 --> 00:38:06.456
Once Upon a Time, the Reader had a price.
00:38:07.141 --> 00:38:08.771
And the Writer had a price.
00:38:08.951 --> 00:38:17.460
Now, if we're not familiar with the Writer, Adobe Acrobat Writer would create the PDFs, and it was sort of the de facto way that you would create a PDF.
00:38:17.840 --> 00:38:20.610
And the Reader is, you just read the PDF.
00:38:21.246 --> 00:38:27.766
And you would price these two products in accordance with the point on demand curve where there's unitary elasticity.
00:38:27.960 --> 00:38:41.702
But what Adobe figures out is, okay, if we give the Reader a way for free, and therein maximize consumer surplus, what's for each unit downloaded of the reader, you're pushing out the demand curve for the writer and the demand curve for the writer is fairly inelastic.
00:38:42.481 --> 00:38:51.911
So price times quantity on the writer side, that revenue differential more than offsets the fact that you completely subsidize the reader.
00:38:52.516 --> 00:38:58.766
And that's that cross product sort of price elasticity of demand, that's sort of recursive, if you will.
00:38:59.286 --> 00:39:01.237
And that defines a platform.
00:39:01.757 --> 00:39:06.054
And so a platform, or maybe a marketplace, has those two demand curves.
00:39:06.331 --> 00:39:08.481
And they're sort of feeding off of each other.
00:39:08.902 --> 00:39:14.012
Now, if you think of Uber versus Waymo, Uber has two demand curves.
00:39:14.362 --> 00:39:17.612
It doesn't seem like Waymo has two demand curves because there's no driver in the car.
00:39:17.889 --> 00:39:20.869
But they, but it functions like a marketplace.
00:39:21.251 --> 00:39:26.702
And the user, from the user's perspective, they're calling a car and the car comes and picks them up, as does an Uber.
00:39:27.112 --> 00:39:30.681
So that's what I'm thinking of in terms of a synthetic marketplace.
00:39:31.007 --> 00:39:37.367
You're not, you don't have that inherent chicken and egg problem that you would when you're trying to see a marketplace.
00:39:38.012 --> 00:39:56.672
with human beings and with regard to these agents, if these agents have the ability to automate entire jobs because they've been trained on workflows, then you can effectively build a synthetic marketplace where humans describe the work that is to be done.
00:39:57.376 --> 00:40:15.072
That's very different than posting a job for a role, but if you could describe the work that is to be done and then sort of the synthetic marketplace itself allocates agents to doing that type of work, that to me is sort of a Waymo meets Upwork.
00:40:15.842 --> 00:40:21.824
And that goes back to your earlier point about how does this stuff disrupt the freelancing space.
00:40:22.434 --> 00:40:27.224
Yeah, it's interesting too, because it also, mean, obviously Waymo has other challenges, right?
00:40:27.224 --> 00:40:38.610
They have capital challenges, they have resource needs, but those are needs that companies Can scale more effectively without some of the complications that come with this double sided marketplace.
00:40:38.940 --> 00:40:41.860
And, uh, frankly, companies already have that challenge.
00:40:41.860 --> 00:40:42.590
Do I have the cash?
00:40:42.590 --> 00:40:43.949
Do I have the resource I need?
00:40:43.949 --> 00:40:48.757
So it does simplify it for the entity and gives them opportunities.
00:40:49.047 --> 00:40:51.751
Uh, you know, I love the Adobe example you use.
00:40:51.751 --> 00:40:53.172
I think that's a great way to explain it.
00:40:53.222 --> 00:40:57.382
So, uh, this has been a fantastic conversation on the labor economics side.
00:40:57.661 --> 00:40:59.461
I want to close this conversation.
00:40:59.461 --> 00:41:07.041
I wish we had another hour, but it is what it is, uh, to kind of dive into a bit more about where you see the future of AI and agentic AI going.
00:41:07.766 --> 00:41:14.306
How do you see AI impacting, particularly code development and software engineering in the kind of mid to long term?
00:41:14.586 --> 00:41:22.815
think in the mid to long term, humans are in the loop and they're, they are contributing to the code base and they are doing so quite meaningfully.
00:41:23.041 --> 00:41:28.204
And they're benefiting from AI and the two work as, as counterparts.
00:41:29.284 --> 00:41:32.784
And that will be probably, a flow.
00:41:33.204 --> 00:41:42.197
And then there's another flow that's probably happening at the same time where humans are effectively building these frameworks to automate the other flow.
00:41:42.420 --> 00:41:48.041
And then the question then becomes, you know, in the not so distant future, how does that bear out?
00:41:48.657 --> 00:41:58.121
And then what flows are actually allocated to the agents, the agents and the agentic frameworks and the humans that are effectively designing those frameworks.
00:41:58.463 --> 00:42:05.471
And then what flows are going to sort of this, with a co pilot versus an organization with an autopilot.
00:42:05.701 --> 00:42:17.704
this seems like it aligns really well to what you're doing with Memra, uh, your company, where you're helping enterprises to leverage gigantic AI, and kind of scale out operations internally versus outsourcing.
00:42:18.224 --> 00:42:18.813
Correct.
00:42:18.987 --> 00:42:26.331
I think tactically there's some work that has been going, maybe offshore or to various outsourcing vendors.
00:42:26.454 --> 00:42:36.534
That I see as low hanging fruit that can kind of come back into the enterprise, but, brought back to the enterprise for AI to, for AI to take on.
00:42:36.728 --> 00:42:37.307
Fantastic.
00:42:37.307 --> 00:42:39.527
I really appreciate all these insights, Amir.
00:42:39.597 --> 00:42:41.197
It's been a wonderful conversation.
00:42:41.802 --> 00:42:54.514
As we wrap up, maybe if you could share with me any closing thoughts you have for You know, software engineers or AI practitioners listening who are looking to stay ahead and kind of get a jump on leveraging AI agents.
00:42:54.525 --> 00:42:55.364
What should they be doing?
00:42:55.532 --> 00:42:58.222
I think the, the answer to that is do.
00:42:58.697 --> 00:43:07.081
I mean, I think with a lot of things, maybe there are thinkers and doers, and they're not the same, like the thinkers are allocating tasks.
00:43:07.262 --> 00:43:10.422
They're contemplating and then they're allocating tasks to the doers.
00:43:11.027 --> 00:43:27.516
But here, specifically, I think the doers are the thinkers, and specifically there's a lot of thinking that you can do as an engineer as to how to sort of shape the future of agentic frameworks or the future of work just by doing.
00:43:27.824 --> 00:43:34.025
I think a lot of the very successful AI companies, the startups, they're developer led right now.
00:43:34.391 --> 00:43:45.284
I think this model of, delegating work, I think it doesn't necessarily work, it doesn't necessarily translate, with AI as well as it does with other areas.
00:43:45.494 --> 00:43:47.135
And that's okay in the short run.
00:43:47.144 --> 00:44:16.867
I think in the long run, yeah, you may want to, uh, benefit from, um, the returns to scale that come from delegation, but in the short run, there's so much to learn by interacting with these models, by developing the infrastructure to augment the models, by learning about the processes the organizations have, and sort of improving on those processes and linking them to the agentic frameworks and the effectively the LLMs, there's so much learning that can take place with the LLMs.
00:44:17.079 --> 00:44:23.079
and in fact, some of that learning can fuel these vertical go to market AI first companies.
00:44:23.579 --> 00:44:28.469
And then some of that learning, um, maybe speaks to the other point about vertical integration.
00:44:28.663 --> 00:44:30.472
And there's a lot of work to be done here.
00:44:30.853 --> 00:44:32.432
It's super interesting work.
00:44:32.432 --> 00:44:38.672
It's really kind of fun to be honest with you as well, both to produce and to consume this stuff is super exciting.
00:44:39.014 --> 00:44:39.534
Perfect.
00:44:39.664 --> 00:44:41.735
Well, thank you so much for coming on the show, Amir.
00:44:41.744 --> 00:44:43.894
It's been wonderful having this conversation with you.
00:44:43.905 --> 00:44:44.925
I know I've learned a lot.
00:44:44.925 --> 00:44:46.485
I hope our audience does as well.
00:44:46.864 --> 00:44:52.244
And for those who want to learn more, want to follow Amir's work on these topics, you can find him on LinkedIn.
00:44:52.255 --> 00:44:53.764
He shares a lot of incredible stuff there.
00:44:53.965 --> 00:44:55.454
And you can also check out Memra at memra.
00:44:55.485 --> 00:44:58.228
co, to learn more about AI and agentic systems.
00:44:58.543 --> 00:45:05.222
As always, thanks everyone for tuning in to Dev Interrupted, uh, and be sure to describe to our newsletter on Substack for deeper dives into the topics we discussed today.
00:45:05.632 --> 00:45:06.842
Amir, thank you so much for coming on.
00:45:07.152 --> 00:45:07.842
Thank you very much.
00:00:00.050 --> 00:00:03.839
The work of an AI engineer is really to kind of build those systems.
00:00:04.330 --> 00:00:06.339
That effectively build systems.
00:00:06.794 --> 00:00:22.655
I guess if you're building agentic frameworks and then deploying those agents to effectively write code, and then the code is, for the purposes of building an application, then you're building the systems that effectively build the constituent systems that build the application.
00:00:22.664 --> 00:00:28.434
And in that regard, it's sort of more like industrial engineering than it is necessarily software engineering.
00:00:29.039 --> 00:00:34.950
13% of all pull requests are bot created today and they are creating a unique impact on your SDLC.
00:00:35.240 --> 00:00:56.880
LinearB his upcoming research exposes the effect bots are having on our teams developer experience and productivity, and engineering org who created a system for managing bot generated pRS are able to reduce their entire review load by over 6% while also making drastic improvements in their security and compliance posture you want to learn how your team can manage bot generated PRS and get early access to linear B's data report.
00:00:57.270 --> 00:01:01.649
Head to the show notes to register for our upcoming workshop on September 24th or 25th
00:01:01.649 --> 00:01:03.009
Welcome back to Dev Interrupted everyone.
00:01:03.070 --> 00:01:04.790
I'm your host, Conor Bronsdon.
00:01:04.790 --> 00:01:09.329
And today I'm joined by Amir Babani, a mathematician and AI ML expert.
00:01:09.680 --> 00:01:15.340
Amir is both a founder and the chief AI engineer at Memra, and he's had previous exits to both Google and Meta.
00:01:15.650 --> 00:01:17.239
Amir, thank you so much for joining me today.
00:01:17.700 --> 00:01:18.269
Thank you for having me.
00:01:18.641 --> 00:01:37.751
Yeah, I'm really excited for this conversation because we've talked a lot about how AI is impacting software engineering from the perspective of engineering leaders, but we haven't really had the opportunity to dive in depth on what's happening and, uh, as far as the forefront of it and kind of what we see moving forward with someone who's a deep expert in the research side.
00:01:37.900 --> 00:01:50.990
So in today's conversation, we're going to have that chance to zero into the future of AI and software engineering and hear from you about Kind of the cutting edge of research, uh, plus talk about agentic AI and how that may transform software workflows or other workflows in the world.
00:01:51.411 --> 00:01:53.700
But before we jump in, I do want to remind our listeners.
00:01:53.751 --> 00:01:56.361
I know we say it every time, but it really does matter.
00:01:56.400 --> 00:02:02.260
If you enjoy this episode, please just take one moment to rate and review the podcast on your X.
00:02:02.456 --> 00:02:10.536
App of choice, Stitcher, Google Podcasts, Spotify, whichever it is, it helps us bring more insightful conversations with leaders like Amir.
00:02:10.915 --> 00:02:15.265
And if you really love it, uh, tweet it or zeet it, uh, post on LinkedIn.
00:02:15.276 --> 00:02:16.235
We'd love to hear from you.
00:02:16.430 --> 00:02:19.569
but with no further ado, Amir, let's dive in.
00:02:19.659 --> 00:02:23.129
Agentic AI is rapidly redefining how we interact with technology.
00:02:23.495 --> 00:02:34.544
Right now, can you start by explaining for our audience what agentic AI means, just to make sure we're all on the same page and how it differs from prior concepts folks may have heard of such as RPA or robotic process automation.
00:02:35.375 --> 00:02:41.085
I think of agentic AI as Blocks of code that have reasoning capability.
00:02:41.555 --> 00:02:43.495
So they actually have access to an LLM.
00:02:43.544 --> 00:02:44.485
That's the first thing.
00:02:44.925 --> 00:02:47.585
Then they have access to long term memory.
00:02:47.645 --> 00:02:57.465
And you can think of that as akin to the RAG layer, so that would be the vector databases plus the graph databases plus other long term, forms of memory.
00:02:57.775 --> 00:03:02.638
Then they have access to adapt, uh, sort of short term memory that, increases their adaptability.
00:03:03.038 --> 00:03:06.698
And then they have access to tasks on which they're trained.
00:03:07.091 --> 00:03:10.010
And then finally they have access to some sort of integration layer.
00:03:10.050 --> 00:03:14.010
They integrate into enterprise applications or other workflows.
00:03:14.911 --> 00:03:23.121
And in so doing, unlike, let's say, RPA, because I often get the question, like, how does, how does this, technology contrast with, RPA?
00:03:24.127 --> 00:03:26.897
agents are interacting with knowledge.
00:03:27.229 --> 00:03:38.742
As context, as opposed to sort of data inputs, and they can sort of reason their way to a conclusion when they're interacting in so far as these workflows are concerned.
00:03:38.983 --> 00:03:45.872
So for example, you can say, write these documents to a database that begs a whole suite of questions.
00:03:46.323 --> 00:03:51.490
What fields in the database, how does the fields in the database map to this particular document?
00:03:51.520 --> 00:03:54.099
What do you want me to extract from this document and write to the database?
00:03:54.389 --> 00:03:55.199
Or database.
00:03:55.754 --> 00:03:57.634
What's the path of a base, etc.
00:03:58.055 --> 00:04:14.389
So the agents, have the ability, on the fly to reason through that imperative and get to an outcome that was so desired by the, by the user who's sort of, you know, Commanding them to, do that particular job.
00:04:14.429 --> 00:04:17.836
I
00:04:18.690 --> 00:04:21.430
deterministic code essentially where, Hey, we.
00:04:21.786 --> 00:04:24.096
We expect the output to remain the same.
00:04:24.396 --> 00:04:36.165
And this more non deterministic model of, of an agent or typically non deterministic, you get so much more capability as far as reasoning, but the array of results you may receive is more varied.
00:04:36.185 --> 00:04:47.586
And this becomes particularly interesting for folks in software engineering, because, uh, as we've seen at LinearB, we've just released a 2024 bot automation research study that found that more than 13 percent of pull requests.
00:04:47.685 --> 00:04:49.365
Today are already bot created.
00:04:49.795 --> 00:04:57.826
And we expect that number to increase with this continued innovation in AI, more happening on the agentic AI side of things.
00:04:58.055 --> 00:05:00.105
How rapidly do you expect to see.
00:05:00.531 --> 00:05:08.060
Fully AI, you know, non deterministic models, uh, come through and actually commit code to a code base.
00:05:08.706 --> 00:05:10.262
think it's happening as we speak.
00:05:10.262 --> 00:05:12.483
I've tried a myriad of these tools myself.
00:05:12.908 --> 00:05:14.798
And these, um, these other frameworks.
00:05:14.901 --> 00:05:18.271
I'm both a consumer of these things and a producer and I enjoy using them.
00:05:18.271 --> 00:05:19.670
They're, they're actually quite fun.
00:05:20.218 --> 00:05:38.404
I think in a lot of cases, there's a lot of work that is sort of needed for these things to fully automate, the development and the deployment of large scale applications, however, the pace at which that work and sort of that innovation is taking place.
00:05:38.773 --> 00:05:49.470
I don't see it too far in the distant future when, at least small scale, fully functioning applications are developed and deployed through a single prompt.
00:05:49.730 --> 00:05:58.819
With a bunch of the reasoning and sort of the Socraticisms that happen between the single prompt and the final outcome taking place.
00:05:58.894 --> 00:06:01.627
in between the command and the outcome.
00:06:01.985 --> 00:06:03.134
So I do see that taking place.
00:06:03.175 --> 00:06:07.564
And actually that, that speaks to, well, how do you maintain context?
00:06:08.485 --> 00:06:15.028
How do you actually break down a particular complex function into constituent sets of tasks?
00:06:15.350 --> 00:06:19.014
As you would maybe, MapReduce for a workflow.
00:06:19.254 --> 00:06:24.211
and for example, what is the right network topology to maintain context?
00:06:24.473 --> 00:06:25.733
And to get a job done.
00:06:25.843 --> 00:06:28.392
So, for example, there are different types of network topologies.
00:06:28.682 --> 00:06:31.523
A bucket brigade is a very simple network topology.
00:06:31.543 --> 00:06:36.543
There's a house burning here, there's a lake here, and you just move the buckets in a linear fashion.
00:06:37.233 --> 00:06:43.952
If you're disseminating orders in sort of a complex organization, you want that topology to be hierarchical.
00:06:44.923 --> 00:06:52.403
If access to information is necessary, sort of in near real time, you kind of want to have a flat or heterarchical topology.
00:06:52.702 --> 00:07:13.812
So identifying a given problem and mapping it to the right network topology so that you can break the tasks down into sort of a constituent set of subtasks and then have agents that are trained on specific subtasks allocated to those subtasks and then have let's say a master agent as a reducer summing the sums.
00:07:14.319 --> 00:07:35.009
Okay, and so if you kind of pull that together, and that's sort of like what's missing in a lot of these application or agentic frameworks that I've tried thus far as a consumer, if that can be pulled off, yes, I see that in a not so distant future, you can sort of command, development and deployment of at least small scale applications.
00:07:35.387 --> 00:07:43.377
So my kind of mental framework for this is that we're seeing this shift from something that I think a lot of us have done for a long time, which is.
00:07:43.682 --> 00:07:59.353
Hey, we're going to automate away repetitive tasks to not only are we going to automate away repetitive tasks, but we are going to essentially create agentic employees who are working on your behalf and where you have to really think about the inputs you're giving to those employees.
00:07:59.452 --> 00:08:03.343
And maybe there's less of the people management, but the information management becomes even more important.
00:08:03.932 --> 00:08:05.983
Is that an accurate framework to apply here?
00:08:05.983 --> 00:08:06.201
There's
00:08:06.350 --> 00:08:17.264
I think if, if you look at agents as they stand right now, without mentioning any specific agentic frameworks, but if you just, I mean, there's, there's some well known ones out there.
00:08:17.737 --> 00:08:22.497
what they do is they reference a function as a tool.
00:08:22.855 --> 00:08:23.954
And then that tool.
00:08:25.023 --> 00:08:31.096
with several other parameters is allocated to a particular block of code.
00:08:31.096 --> 00:08:44.495
So the block of code, as I said, has an LLM, it can have even an LLM router, it will have some sort of memory, and it will have access to this tool, and that tool is making a effectively making a function call, or it's having access to a function.
00:08:44.817 --> 00:08:56.477
Now, if you think of these functions as methods that belong to a class, And if you can sort of further abstract that the classes sort of fit into larger workflows.
00:08:56.784 --> 00:09:16.780
then you can, and this is actually what we're doing, so I'm kind of speaking from, from experience here, we're registering into memory of a master agent, not a single function to which it has access through a tool, but whole classes and methods that map to a particular workflow.
00:09:16.990 --> 00:09:23.071
So effectively, you've taken, a master agent and you've trained it on a set of workflows.
00:09:23.880 --> 00:09:38.197
Well, if you think about that in the context of workers, a human has a role, that role maps to a job, that job maps to a set of workflows, those workflows map to sets of constituent tasks.
00:09:38.909 --> 00:09:59.490
And right now, I think what we're saying is that, if we can take a workflow and break it down into constituent tasks and allocate agents to those tasks and complete those workflows, then effectively those jobs are being done by the agents, which then means these agents have a role to play within an organization.
00:09:59.846 --> 00:10:04.745
And in that case, I would say that these agents are effectively digital employees.
00:10:04.975 --> 00:10:15.543
And those digital employees paired with a human employee, say an individual contributor, suddenly that individual contributor has multiplicative returns to scale.
00:10:15.716 --> 00:10:17.706
As does, um, hiring a manager.
00:10:17.745 --> 00:10:22.505
If you hire a manager, perhaps, yes, indeed, perhaps, yeah.
00:10:22.596 --> 00:10:24.716
Even, and, and actually think of it from this perspective.
00:10:24.716 --> 00:10:30.576
If you have the employee right now using, let's say, ChatGPT, there's marginal improvement on their productivity.
00:10:31.066 --> 00:10:49.070
But if the human is taken out of the loop, because that human is maintaining context through that, through the dialogue with the GPT, sort of that's they're sitting around maintaining context, engaging, winnowing down on, a set of conversations and therein the outcomes of those conversations.
00:10:49.720 --> 00:10:58.080
But if the agent is doing that, then that's, that's a much higher return to scale if that human is sort of out of the loop.
00:10:58.139 --> 00:11:01.125
Now the human can be on top of the loop, so to speak.
00:11:01.625 --> 00:11:05.686
And on top of the loop means they're engaging in an imperative.
00:11:06.385 --> 00:11:07.456
Go do this.
00:11:07.875 --> 00:11:11.316
And they're not sitting in the middle, engaging in the interrogative.
00:11:12.503 --> 00:11:15.278
agents, are engaging in the interrogative.
00:11:15.477 --> 00:11:22.908
And that's what I mean by there's a Socraticism that happens between the command to go do something and the final output.
00:11:22.908 --> 00:11:25.493
At
00:11:25.878 --> 00:11:42.057
I mean, it still maintains some of the, I'd say, black box concerns folks have at the model level, but because of the way you're constructing these agents and how they are approaching things, it sounds like there's a much better auditing process for what's actually happening there.
00:11:42.457 --> 00:11:50.307
That then the human employees can focus on, okay, let's audit to make sure this is working and that we're actually delivering on these processes.
00:11:50.567 --> 00:12:03.248
Whereas I'll say currently, one of the big things we noticed in that, uh, bot report I mentioned was that 96 percent of all basic bot created PRs that are being sent out right now are not linked to product management tools.
00:12:03.557 --> 00:12:17.192
Well, if I have an agent doing that, I presume we can simply make sure the agent is using a project management tool to track this and kind of improve the observability of a lot of this code that's coming into the code base and a lot of the approaches we're taking here.
00:12:17.544 --> 00:12:32.341
present moment, we're able to Um, the steps that an agent takes to complete tasks, and in fact, um, we're able to cache and, and sort of have a, an auditable record of the reasoning process.
00:12:32.754 --> 00:12:36.034
And so it's not at the workflow level.
00:12:36.034 --> 00:12:37.955
It's not fully black box.
00:12:38.024 --> 00:12:45.764
Now, one could argue that again, these agents have access to LLMs and the LLMs have sort of deep learning models.
00:12:46.240 --> 00:12:52.879
And maybe there's some black box models at the root of sort of the language models.
00:12:52.909 --> 00:13:06.932
That's, that's definitely, I mean, that's probably a case, I mean, when I think about older, sort of like the last decade where we're building, machine learning models, some of those non parametric models, they're black box.
00:13:08.251 --> 00:13:09.741
It's really hard to interpret.
00:13:10.110 --> 00:13:22.384
what's actually going on and, and that's different from the parametric models where you have sort of, you know, a prior assumption as to the distributions, of the fields that comprise the training set.
00:13:22.631 --> 00:13:31.192
and I'm not sure in all cases, the black box nature of those models was necessarily an issue.
00:13:31.201 --> 00:13:31.211
Uh
00:13:31.355 --> 00:13:32.035
I hear you.
00:13:32.186 --> 00:13:45.895
It's, it's interesting though, because I think a lot of, you know, survey data will, will show that the concerns that you, when you talk to a, you know, CISO or just a general software engineer, is they're saying, Hey, I don't get what's happening here.
00:13:46.296 --> 00:13:49.615
Uh, I mean, we can also talk about the concerns of, is this going to replace me?
00:13:49.645 --> 00:14:02.956
I think we will get to that in the labor economics side of this, but what would you say to those folks who are just worried about implementing this and having, not being able to unpack when something goes wrong or, uh, other concerns on that kind of black box piece?
00:14:03.192 --> 00:14:03.451
Do
00:14:03.672 --> 00:14:22.479
Well, I think if you're, if we're talking about agents, then, and agentic frameworks, then a lot of the reasoning is, There's layers of reasoning that are abstracted above and beyond the deep learning models that are sort of the base of these LLMs.
00:14:22.998 --> 00:14:33.009
that set of reasoning processes, you're able to cache that, you're able to trace it, and you're able to audit it.
00:14:33.759 --> 00:14:55.206
So at that level, I think, um, and generally as I, as I think about this, the worker, maybe the, the, the software engineer or whatnot, they would want access to that layer of reasoning, I'm not sure they necessarily need access to, um, reasoning that's happening, so to speak, or the inferencing that's happening, so to speak, at the LLM layer.
00:14:55.860 --> 00:15:07.636
you think software engineering is going to be one of the disciplines that particularly benefits from this increasing use of agentic AI and its ability to autonomously handle tasks?
00:15:08.024 --> 00:15:08.543
I think so.
00:15:08.543 --> 00:15:20.739
I think that software engineering is going to benefit as other disciplines, knowledge work disciplines, I think it's going to benefit from sort of an An industrial engineering paradigm, right?
00:15:20.739 --> 00:15:35.182
So if you go back in time, you have, let's say factories with seamstresses were designing and fabricating sewing effectively clothing, but the industrial revolution came along and said, okay, we're going to design.
00:15:35.741 --> 00:15:43.001
The factories and the systems that output those articles of clothing and software engineers.
00:15:43.011 --> 00:15:51.629
Similarly, we're, in fact, arguably in some regard, the work of an AI engineer is really to kind of build those systems.
00:15:52.119 --> 00:15:54.129
That effectively build systems.
00:15:55.134 --> 00:16:12.294
I guess if you're building agentic frameworks and then deploying those agents to effectively write code, and then the code is, um, for the purposes of building an application, then you're building the systems that effectively build the, the constituent systems that build the application.
00:16:12.304 --> 00:16:18.075
And in that regard, it's sort of more like industrial engineering than it is necessarily software engineering.
00:16:18.389 --> 00:16:23.690
Or I would almost describe it as more of an architect mindset versus a carpenter mindset.
00:16:23.750 --> 00:16:30.203
Um, and we've already talked about that kind of, mean, plenty of people have an architect title already, and I think we're going to see a push more towards that.
00:16:30.374 --> 00:16:37.634
what are, what are the other industries that you expect to see a lot of changes, uh, coming due to agentic AI improvements?
00:16:37.996 --> 00:16:47.403
Well, I think in terms of roles within the knowledge work, the arena of knowledge work, I, I envision these agents climbing the career ladder.
00:16:47.549 --> 00:16:51.504
So initially they can do some data entry, then they can book meetings.
00:16:52.063 --> 00:17:00.604
You know, then they can send emails out and sort of monitor, maybe drip campaign and then eventually they kind of just go up this career ladder.
00:17:00.634 --> 00:17:04.503
They progress, um, maybe they write some basic applications.
00:17:04.804 --> 00:17:09.503
I do think it's a, we're, we're far from the agents necessarily solving for product market fit.
00:17:09.686 --> 00:17:12.576
Or doing sort of more, you know, or maybe designing
00:17:12.676 --> 00:17:12.916
Good.
00:17:12.916 --> 00:17:13.866
I need to keep my job.
00:17:13.866 --> 00:17:14.116
So,
00:17:14.920 --> 00:17:18.227
I think sort of the more, you know, the complex and creative.
00:17:18.653 --> 00:17:27.339
job functions, as you say, for example, architecture, it's not necessarily, that's a layer abstracted away from doing specific tasks.
00:17:27.579 --> 00:17:36.054
So in that regard, I see, a whole myriad of especially task based work being disrupted.
00:17:36.054 --> 00:17:50.155
I Something I'm realizing when I interact with companies is that there seems to be that of three or four stages prior to even engaging, but each of those stages, the customers are willing to pay for.
00:17:50.516 --> 00:17:59.199
And the first is enterprises they want people to explain to them how AI fits into everything.
00:17:59.991 --> 00:18:07.058
They don't have, let's say, necessarily, an immediate answer to the use case.
00:18:07.737 --> 00:18:10.616
Within the organization that can benefit from AI.
00:18:11.156 --> 00:18:16.247
And I actually think that type of reasoning is this design of experiment reasoning.
00:18:16.896 --> 00:18:17.126
Hmm.
00:18:17.402 --> 00:18:21.721
And you learn about that in human factors engineering, you learn about it in research methods.
00:18:22.321 --> 00:18:28.751
Certainly if you're doing machine learning work, you have to sort of say, okay, what is the use case?
00:18:28.751 --> 00:18:34.071
But you're reducing that use case down to a single variable called the target variable or the dependent variable.
00:18:34.645 --> 00:18:39.746
And you're aligning a bunch of data that you have to sort of bear results for that target variable.
00:18:39.836 --> 00:18:43.695
And that's a, that's a way of thinking about the world.
00:18:44.118 --> 00:18:47.169
And that is sort of like an experimental form of reasoning.
00:18:47.361 --> 00:18:58.527
And that lends itself to kind of going into an organization and figuring out how things are functioning and then figuring out how AI can benefit those systems and processes.
00:18:59.250 --> 00:19:02.090
And right now I notice that companies just want help with that.
00:19:02.513 --> 00:19:11.874
And some subset of those companies, I think, as a sort of an emerging AI startup, those are your pilots.
00:19:12.406 --> 00:19:15.176
And then eventually those are your MSAs.
00:19:16.568 --> 00:19:20.048
I think Bain Capital put out a report recently that I thought was interesting.
00:19:20.251 --> 00:19:25.538
showed some hurdles that have been overcome in so far as AI adoption.
00:19:25.568 --> 00:19:27.838
And some of those hurdles were InfoSec related,
00:19:28.171 --> 00:19:28.471
Hmm.
00:19:28.549 --> 00:19:29.296
It
00:19:29.445 --> 00:19:36.278
data private, but the other hurdles that have yet to be overcome, sort of like, what's the use case?
00:19:36.499 --> 00:19:38.108
It's like kind of a classic thing.
00:19:38.118 --> 00:19:39.098
Like what's the use case?
00:19:39.098 --> 00:19:40.269
How do we apply this?
00:19:40.868 --> 00:19:42.348
How do we wrap our minds around this?
00:19:43.084 --> 00:20:03.570
And it seems like the role of the AI engineer right now is really that, and then eventually that translates into ingesting data, ingesting content, building systems, analyzing, sort of the output, based on the output, where's the refinement, and then based on all of that, can you get agents to kind of automate processes.
00:20:04.675 --> 00:20:08.726
um, there's a lot of work right now to be done just informationally.
00:20:08.726 --> 00:20:10.435
The work doesn't stop there, by the way.
00:20:10.945 --> 00:20:13.046
Um, and, and it really shouldn't just stop there.
00:20:13.066 --> 00:20:18.746
That's not, that's not the end all be all, but it's, um, it's certainly a, it's certainly a need right now,
00:20:19.041 --> 00:20:31.971
feels like part of what's happening is we know there's so much potential to apply AI, uh, and there's so many opportunities that zeroing in on where to start and kind of the steps to take along the path is a challenge for a lot of folks right now.
00:20:32.352 --> 00:20:32.872
precisely.
00:20:32.872 --> 00:20:36.362
One way I would think about this design experiment reasoning is that.
00:20:36.886 --> 00:20:42.186
You've identified the why and you can distill that through the what into the how.
00:20:42.621 --> 00:20:49.221
there's a lot of, um, content out there that, okay, the why is getting to the point of being self explanatory.
00:20:49.431 --> 00:20:51.250
How do you distill that down to the how?
00:20:51.419 --> 00:21:03.909
In the short run, you've noted that AI will encourage engineers to think more Socratically, to ask more questions, to really elicit more, what does that mean for day to day development now?
00:21:03.949 --> 00:21:07.229
And then I'm curious where you think it's going to go in the future.
00:21:15.597 --> 00:21:18.438
by engineers specifically.
00:21:18.478 --> 00:21:20.034
Um, I don't know.
00:21:20.074 --> 00:21:26.621
I mean, I, I think, more specifically, the agents themselves will take on that Socratic reasoning.
00:21:27.017 --> 00:21:27.196
And
00:21:27.281 --> 00:21:29.821
you think the shift is already happening to a more.
00:21:30.112 --> 00:21:35.172
Managerial mindset for engineers where it's like, Hey, I'm going to manage these agents.
00:21:35.392 --> 00:21:40.362
I need to make sure they're, they're functioning well versus let me elicit info from them.
00:21:40.825 --> 00:21:46.525
I think that the shift is happening where the human being commands the agents.
00:21:46.904 --> 00:21:47.795
That's the imperative.
00:21:48.633 --> 00:21:57.546
And then the reasoning that needs to happen between the imperative and the completed task, no pun intended, it begs a lot of questions.
00:21:58.214 --> 00:22:03.355
And it's, it's that interrogative process that I sort of think of as like the Socratic dialogue.
00:22:03.789 --> 00:22:04.079
Got it.
00:22:04.089 --> 00:22:07.970
And then obviously on the more, I'll call it individual level.
00:22:08.009 --> 00:22:13.250
We're all engaging in that Socratic dialogue when we simply use chat2BT to like rewrite an email for
00:22:13.309 --> 00:22:14.200
Precisely.
00:22:14.442 --> 00:22:19.303
And when we're engaging in that Socratic dialogue, we're maintaining context in our minds.
00:22:19.873 --> 00:22:40.382
To sort of guide the conversation towards an outcome, whether it's winnowing down the reasoning process towards something that we want, uh, with increased specificity, that process again depends on these, the access to long term memory, the access to short term memory for adaptability, training these agents on particular tasks.
00:22:40.731 --> 00:22:42.500
Again, there's a difference between knowledge, and knowledge.
00:22:42.741 --> 00:22:43.682
And expertise.
00:22:43.776 --> 00:22:49.185
So there's a knowledge layer here with these agents and there's sort of like a, an expertise layer.
00:22:49.726 --> 00:23:03.791
And for example, you see that, for example, if you want to have a contract written there, you can leverage chat GPT to edit various clauses and you'd probably want someone with expertise to just look it over, see if it, is cogent.
00:23:04.000 --> 00:23:08.557
that is sort of where a lot of this modeling context comes into play.
00:23:08.557 --> 00:23:17.637
And all of that modeling context, as I see it, is a bus between sort of the foundational layer and the, and the application layer.
00:23:17.647 --> 00:23:18.678
And it's, it's really.
00:23:19.548 --> 00:23:31.127
All of us right now with regard to these agentic frameworks, we're sort of just developing the bus that sits between that, the, the CPU analogy and the application layer now.
00:23:32.096 --> 00:23:38.027
Like, I think many folks listening, I am only at this point using kind of the Socratic approach.
00:23:38.027 --> 00:23:39.967
I'm leveraging maybe Copilot.
00:23:39.977 --> 00:23:43.686
Maybe I'm just leveraging ChatGPT to help me with certain test writing tasks.
00:23:43.936 --> 00:23:51.297
I'm not yet leveraging agents, but I see that it's going that way and I want to start leveraging agents within my software engineering team.
00:23:51.946 --> 00:23:54.457
What would be your advice to those folks who are listening and saying?
00:23:55.027 --> 00:23:55.307
Damn.
00:23:55.307 --> 00:23:55.666
Okay.
00:23:55.856 --> 00:23:56.557
Here's the next level.
00:23:56.557 --> 00:23:57.436
I need to start moving there.
00:23:58.020 --> 00:24:06.359
one advice is to say, with regard to the labor market transaction, there's two sides and there's a sell side of the labor market transaction.
00:24:06.359 --> 00:24:07.380
There's the buy side.
00:24:07.837 --> 00:24:14.137
it will be, I think, increasingly difficult to compete with agents On the sell side of that labor market transaction.
00:24:14.425 --> 00:24:23.983
So if you're, if you're on the buy side of that transaction, then these agents are great because they've reduced your operational expenditure, et cetera.
00:24:24.541 --> 00:24:30.086
so the way I see it that, we as humans will probably leverage these agents.
00:24:30.606 --> 00:24:44.578
To do work and if we can own relational structure, we own the product of the labor and we kind of own the relational structure, then the humans directly benefit from the agents.
00:24:45.398 --> 00:24:54.365
But if you don't have ownership over that relational structure, And you don't have ownership over that work product, then effectively you're competing with the agents.
00:24:55.132 --> 00:24:55.561
Yeah.
00:24:55.612 --> 00:25:04.491
And I think this is where a lot of the fear currently comes in for folks where they hear that and they go, wait, does that mean I need to run my own company?
00:25:04.511 --> 00:25:07.501
Does that mean I need to go be an entrepreneur?
00:25:07.741 --> 00:25:13.251
Does that mean I'm going to Not be as strong within the labor market in a couple of years as I am today.
00:25:13.261 --> 00:25:14.582
Like, what do I need to do to change that?
00:25:14.852 --> 00:25:20.521
Um, and obviously AI has extremely broad implications on what's happening with the labor economy.
00:25:21.001 --> 00:25:26.172
How do you see AI agents impacting the broader structure of work and labor contracts in the coming year?
00:25:26.703 --> 00:25:30.013
Well, I think the roles based economy is too rigid.
00:25:30.118 --> 00:25:43.809
So I, I see that companies have roles, and again, like I said, those roles map to jobs, to workflows, to tasks, et cetera, and roles offer a degree of economies of scale for these companies.
00:25:44.049 --> 00:25:47.809
Agents actually, by the way, offer economies of scope.
00:25:48.376 --> 00:25:50.096
Cause the way to think about an agent is.
00:25:50.537 --> 00:25:56.012
You're designing a tooling line that can effectively, manufacture various products.
00:25:56.596 --> 00:26:01.576
To your comparison earlier, I'm building the factory versus I'm building the digital piece on the line.
00:26:01.941 --> 00:26:02.641
Right, right.
00:26:03.170 --> 00:26:17.119
I generally think that if you're building agents into enterprise, into the enterprise, and that inclusive of, of workflows, but, but also sort of, um, culturally and structurally.
00:26:17.970 --> 00:26:22.730
Then you may want to rethink this sort of concept of a role.
00:26:24.400 --> 00:26:40.800
when you're starting your own company, when you're functioning as an entrepreneur, you're not engaging day in, day out at the task level, or at like a single workflow level, or you don't just have this one job that maps to this one role.
00:26:40.810 --> 00:26:45.105
I mean, that's, that's, that's, that's, that's, that's, In my opinion, that's very junior.
00:26:45.444 --> 00:26:51.242
And another way of thinking about it is the roles can be assigned to these agents.
00:26:51.637 --> 00:26:58.407
So if the roles are being imparted to agents, how does that redefine the roles based labor economy?
00:26:58.664 --> 00:27:08.193
And if, if again, you're going back and saying these, these effective workers who can sort of think Socratically, some, in many cases, they're polymaths.
00:27:08.970 --> 00:27:19.164
And if they can go in between, Various functions and the agents are sort of doing the work and why should they sit and just do one role all day long?
00:27:19.454 --> 00:27:20.924
This doesn't make any sense whatsoever.
00:27:21.059 --> 00:27:34.227
So all of that, I think if you're thinking about updating the frameworks of the labor economy, I think you slowly want to start thinking about the rigidity of the roles based labor market.
00:27:34.227 --> 00:27:39.297
I mean, we're not, we're not all sort of sitting on a factory line and sort of.
00:27:39.950 --> 00:27:42.890
at that level in, in sort of the knowledge work economy.
00:27:43.319 --> 00:27:48.825
And yet a lot of how we've built work is based off of that factory model, to your point.
00:27:49.085 --> 00:27:51.434
Uh, which is very early 20th century.
00:27:51.663 --> 00:27:52.151
fascinating.
00:27:52.227 --> 00:28:07.926
I wonder if this is going to exacerbate or kind of remove the trend of like freelance work of gig work that we've seen increasingly over the last, you know, 10 years, do you think that agents are really going to replace freelancers?
00:28:07.926 --> 00:28:14.787
Or are we going to see, uh, this kind of, you know, freelancer gig work approach where, Hey, I'm running agents on behalf of a company now.
00:28:15.211 --> 00:28:36.196
Um, because I, I know, and you've talked about this before that many companies, many large enterprises are taking this insourcing approach and saying, well, if I can build an agent to solve these tasks for me, I'd rather be able to have broader observability and potentially cost savings compared to, say, passing this basic data task off to another country where it's cheaper.
00:28:36.675 --> 00:28:42.113
I think there's a, even, let me go one, one, step more, more broadly.
00:28:42.272 --> 00:28:44.932
I think there's a case to be made for vertical integration.
00:28:45.286 --> 00:28:49.268
I think that these agentic frameworks inherently have a multi layer stack.
00:28:49.897 --> 00:29:21.132
You have the foundational layer, you have the long term memory layer, You have the short term memory for adaptability, you have these agents that are trained on tasks, you have this integration layer, and there are feedback loops that, in fact, there's a two fold feedback loop that just comes, you know, comes to mind right now, and that is to say, if you're allocating these agents to specific workflows within an organization, and they're improving processes, and they're improving processes.
00:29:21.560 --> 00:29:27.330
Okay, there's an improvement in so far as the returns to scale and so far as those process improvements.
00:29:28.016 --> 00:29:40.960
But then also the, the, the outcome of those process improvements and the knowledge derived therein can go back and train the models and effectively train the agents to kind of further that process.
00:29:41.480 --> 00:29:48.650
And it's very, very difficult to have those feedback processes without vertical integration.
00:29:48.936 --> 00:29:50.426
So that's one, that's one point.
00:29:50.457 --> 00:30:07.883
There's a second point here, which is that if you think of this multi layered stack, if you will, Again, and go back, let's say 10 years and think about an analogy to AI, which a very good analogy to AI is sort of like machine learning, set of companies in the early 2010s.
00:30:08.282 --> 00:30:13.813
And those companies were, the ML models were like an afterthought to the use case.
00:30:14.307 --> 00:30:15.337
And the application.
00:30:15.817 --> 00:30:30.106
And so you're sort of building at the application layer and then, yeah, you're, you may have an ML model somewhere and, you know, and that, that ML model is benefiting the application, but you're really thinking at a top down, uh, from a top down perspective.
00:30:30.836 --> 00:30:34.833
With these AI, systems, everything's really happening bottoms up.
00:30:34.942 --> 00:30:38.742
The foundational layer comes about first, then the rag layer.
00:30:39.315 --> 00:30:43.694
Then the short term memory layer, then these agentic frameworks.
00:30:44.194 --> 00:30:51.171
And frankly speaking, the application layer, the AI application layer is sort of maybe still around the corner.
00:30:51.459 --> 00:30:58.818
So if all of that stuff is sort of happening bottoms up, well then that begs two interesting questions.
00:30:58.878 --> 00:31:14.586
One is, Where are these AI first companies with vertical go to markets, and then how is that different from an existing incumbent, and that's redundant, but an incumbent at the present moment?
00:31:15.645 --> 00:31:26.174
Acquiring sort of the AI capabilities because they have already distribution and they have sort of monopoly pricing power over those distribution channels.
00:31:26.855 --> 00:31:32.484
And even a vertical go to market AI first company, they have to solve for distribution.
00:31:33.224 --> 00:31:37.099
And arguably they have to solve for distribution faster than the incumbent solves for innovation.
00:31:37.481 --> 00:31:38.991
So there's like a difference of race problem.
00:31:39.521 --> 00:31:44.402
So maybe in the short run you're going into these large enterprises and agentifying them.
00:31:44.832 --> 00:31:52.467
But you have to do that vertically, or at least it would seem that there's a case to be made for vertical integrations that you get those feedback loops.
00:31:52.856 --> 00:31:53.696
That's one thought.
00:31:53.787 --> 00:31:59.729
And then the other thought is those vertical AI first go to market companies.
00:32:00.121 --> 00:32:02.810
This whole analogy is very similar to like Blockbuster.
00:32:03.580 --> 00:32:19.685
So there's an argument to be made with a sort of a smarter company going to Blockbuster and saying, Hey, we're going to, bring something analogous to Netflix, but have, have that inside of Blockbuster and sort of bring Blockbuster up to code for the current decade.
00:32:20.816 --> 00:32:28.756
And alternatively, there's just the, the paradigm where Netflix disrupts Blockbuster,
00:32:29.502 --> 00:32:36.942
So it sounds like what you're saying is if I'm at the startup level, I need to kind of focus on the innovation side.
00:32:37.002 --> 00:32:39.502
Uh, I don't have that vertical integration opportunity quite yet.
00:32:39.502 --> 00:32:44.492
I can start to move towards it, but I don't have the vast stores of data that, you know, the belt and resources.
00:32:44.731 --> 00:32:51.071
and what I can do is I can, like many startups have in the past, out innovate these enterprises if they're not careful.
00:32:51.386 --> 00:32:56.067
Whereas the enterprise layer needs to look at it and say, Hey, I need to insource.
00:32:56.096 --> 00:33:05.816
I need to, really vertically integrate so that I can take advantage of these, you know, monopolistic prices, uh, these data advantages, these scale economies that I have already.
00:33:06.115 --> 00:33:09.226
Um, and that's kind of how you would think about it from, from those perspectives.
00:33:09.567 --> 00:33:17.209
Yeah, when I think about a startup in general, I think of the stock of a startup having an inherent option premium.
00:33:17.509 --> 00:33:23.239
It's not an explicit options contract, but there is no profit that's distilling down to shareholder value.
00:33:23.308 --> 00:33:25.519
So what is the, what is the equity buyer, i.
00:33:25.538 --> 00:33:25.548
e.
00:33:25.558 --> 00:33:30.019
the investor, Buying the generally buying the option premium.
00:33:30.392 --> 00:33:43.632
And as you solve for product market fit, IE, you solve for distribution as a startup, you're exercising that option premium and converting that option premium to enterprise value or effectively equity value.
00:33:43.662 --> 00:33:50.616
And, and sort of that, that analogy is sort of like potential about potential energy being transferred, transferred into kinetic energy.
00:33:50.944 --> 00:33:58.285
And I don't think, you know, generally, a lot of startups are going to fail because they can't innovate.
00:33:58.295 --> 00:34:01.444
They fail because they can't solve for distribution.
00:34:01.824 --> 00:34:06.473
And meanwhile, the vertical integration strategy says the following.
00:34:07.213 --> 00:34:10.403
The acquiring entity has distribution.
00:34:10.664 --> 00:34:36.900
Now, interestingly, there's a There's something called a reverse acquihire now, and that's an interesting form of vertical integration where the larger company, let's say the incumbent, is not really buying your stock, they're actually employing the founders and signing a non exclusive licensing deal for the underlying AI technology, and exclusive share.
00:34:37.351 --> 00:34:39.510
Licensing deal in U.
00:34:39.510 --> 00:34:39.650
S.
00:34:39.670 --> 00:34:51.139
copyright law is a transfer of IP, but a non exclusive licensing deal allows your product still to exist, and you can kind of have non exclusive licensing with many different, uh, vendors.
00:34:51.340 --> 00:35:01.585
But here, there's an interesting hybrid, because the founders are sort of being employed to Work at the incumbent and sort of build that AI sort of from the inside.
00:35:02.378 --> 00:35:12.028
And that's a form of vertical integration that, all of this is really kind of interesting in the short run until those AI first vertical go to market companies come about.
00:35:12.759 --> 00:35:12.858
Yeah.
00:35:12.869 --> 00:35:14.429
And that's certainly coming.
00:35:14.777 --> 00:35:27.387
So we've been talking about this from kind of the, the company perspective, the company's strategy side, but we've also spoken a little bit about the individual side and trying to make that shift to, Hey, I'm managing these agents.
00:35:27.398 --> 00:35:31.878
Um, I'm helping them, uh, versus, Hey, I'm individually executing tasks.
00:35:32.657 --> 00:35:42.427
How do you think folks who aren't on that founder level, uh, but are maybe You know, an engineering manager listening to this, who, who listened to the show thinking, Hey, I'm trying to improve in my career.
00:35:42.427 --> 00:35:43.338
I'm trying to grow.
00:35:43.722 --> 00:35:55.472
Or even like an individual engineer right now, maybe a junior or senior engineer who has been thinking, Hey, I want to move into engineering management, but I am, you know, maybe I'm worried now that, you know, is my role going to go away?
00:35:55.472 --> 00:35:56.592
Am I going to have this opportunity?
00:35:56.762 --> 00:36:00.242
How should they be thinking about their approach to the labor market?
00:36:00.503 --> 00:36:05.617
You know, now in the next couple of years, as this AI transformation wave continues to rock through us,
00:36:05.699 --> 00:36:10.927
Um, there are two ways that just, I think, in the short run, They come to mind.
00:36:11.168 --> 00:36:22.818
One is that you leverage the AI agents or just the AI frameworks for marginal improvement over your current productivity.
00:36:23.054 --> 00:36:28.014
So you can get maybe 20, 30 percent improvement over otherwise.
00:36:28.130 --> 00:36:33.980
which I think where a lot of us are today where it's like, Hey, I'm, I'm leveraging Claude or cursor or whatever else to help
00:36:34.034 --> 00:36:35.094
That that's right.
00:36:35.445 --> 00:36:46.567
The other way, again, in the short run, that comes to mind is effectively designing these AI systems that effectively that which you're tasked to work on.
00:36:46.666 --> 00:36:52.262
So it's building the meta models as opposed to building the models, building the, the, the systems that build systems.
00:36:53.413 --> 00:37:01.413
And that is at least right now, not something with which, these AI agents.
00:37:02.242 --> 00:37:03.833
Are, um, best suited.
00:37:04.043 --> 00:37:06.193
Maybe that'll change in short order.
00:37:06.193 --> 00:37:12.864
I mean, things are moving very, very fast right now, but those are the two ways I would say, to the, the person whom you mentioned.
00:37:13.253 --> 00:37:14.623
It is certainly moving really fast.
00:37:14.634 --> 00:37:16.804
We've got to make sure we get this episode out quickly so that not too
00:37:16.853 --> 00:37:16.873
yeah,
00:37:16.994 --> 00:37:18.884
changes before we go live.
00:37:19.454 --> 00:37:23.463
I know another thing that I've heard you talk about is this idea of a synthetic marketplace.
00:37:23.659 --> 00:37:28.798
I'd love for you to expand on that concept as I'm starting to increasingly see that as kind of where we're going.
00:37:28.869 --> 00:37:32.969
Well, you think of two sided marketplaces, so Uber.
00:37:33.688 --> 00:37:35.699
Airbnb, etc.
00:37:35.871 --> 00:37:37.561
What is a two sided marketplace?
00:37:37.572 --> 00:37:40.326
It's a, it's a, there are two demand curves.
00:37:40.639 --> 00:37:47.018
So in the case of Uber, there's a demand function for passengers and there's a demand function for the drivers.
00:37:47.295 --> 00:37:59.425
And there's an interesting, recursive cross product price elasticity of demand that defines how those two demand curves interact with one another, or I'm sorry, interact with each other.
00:38:00.434 --> 00:38:04.440
An example of this would be Adobe Acrobat, Reader and Writer.
00:38:04.646 --> 00:38:06.456
Once Upon a Time, the Reader had a price.
00:38:07.141 --> 00:38:08.771
And the Writer had a price.
00:38:08.951 --> 00:38:17.460
Now, if we're not familiar with the Writer, Adobe Acrobat Writer would create the PDFs, and it was sort of the de facto way that you would create a PDF.
00:38:17.840 --> 00:38:20.610
And the Reader is, you just read the PDF.
00:38:21.246 --> 00:38:27.766
And you would price these two products in accordance with the point on demand curve where there's unitary elasticity.
00:38:27.960 --> 00:38:41.702
But what Adobe figures out is, okay, if we give the Reader a way for free, and therein maximize consumer surplus, what's for each unit downloaded of the reader, you're pushing out the demand curve for the writer and the demand curve for the writer is fairly inelastic.
00:38:42.481 --> 00:38:51.911
So price times quantity on the writer side, that revenue differential more than offsets the fact that you completely subsidize the reader.
00:38:52.516 --> 00:38:58.766
And that's that cross product sort of price elasticity of demand, that's sort of recursive, if you will.
00:38:59.286 --> 00:39:01.237
And that defines a platform.
00:39:01.757 --> 00:39:06.054
And so a platform, or maybe a marketplace, has those two demand curves.
00:39:06.331 --> 00:39:08.481
And they're sort of feeding off of each other.
00:39:08.902 --> 00:39:14.012
Now, if you think of Uber versus Waymo, Uber has two demand curves.
00:39:14.362 --> 00:39:17.612
It doesn't seem like Waymo has two demand curves because there's no driver in the car.
00:39:17.889 --> 00:39:20.869
But they, but it functions like a marketplace.
00:39:21.251 --> 00:39:26.702
And the user, from the user's perspective, they're calling a car and the car comes and picks them up, as does an Uber.
00:39:27.112 --> 00:39:30.681
So that's what I'm thinking of in terms of a synthetic marketplace.
00:39:31.007 --> 00:39:37.367
You're not, you don't have that inherent chicken and egg problem that you would when you're trying to see a marketplace.
00:39:38.012 --> 00:39:56.672
with human beings and with regard to these agents, if these agents have the ability to automate entire jobs because they've been trained on workflows, then you can effectively build a synthetic marketplace where humans describe the work that is to be done.
00:39:57.376 --> 00:40:15.072
That's very different than posting a job for a role, but if you could describe the work that is to be done and then sort of the synthetic marketplace itself allocates agents to doing that type of work, that to me is sort of a Waymo meets Upwork.
00:40:15.842 --> 00:40:21.824
And that goes back to your earlier point about how does this stuff disrupt the freelancing space.
00:40:22.434 --> 00:40:27.224
Yeah, it's interesting too, because it also, mean, obviously Waymo has other challenges, right?
00:40:27.224 --> 00:40:38.610
They have capital challenges, they have resource needs, but those are needs that companies Can scale more effectively without some of the complications that come with this double sided marketplace.
00:40:38.940 --> 00:40:41.860
And, uh, frankly, companies already have that challenge.
00:40:41.860 --> 00:40:42.590
Do I have the cash?
00:40:42.590 --> 00:40:43.949
Do I have the resource I need?
00:40:43.949 --> 00:40:48.757
So it does simplify it for the entity and gives them opportunities.
00:40:49.047 --> 00:40:51.751
Uh, you know, I love the Adobe example you use.
00:40:51.751 --> 00:40:53.172
I think that's a great way to explain it.
00:40:53.222 --> 00:40:57.382
So, uh, this has been a fantastic conversation on the labor economics side.
00:40:57.661 --> 00:40:59.461
I want to close this conversation.
00:40:59.461 --> 00:41:07.041
I wish we had another hour, but it is what it is, uh, to kind of dive into a bit more about where you see the future of AI and agentic AI going.
00:41:07.766 --> 00:41:14.306
How do you see AI impacting, particularly code development and software engineering in the kind of mid to long term?
00:41:14.586 --> 00:41:22.815
think in the mid to long term, humans are in the loop and they're, they are contributing to the code base and they are doing so quite meaningfully.
00:41:23.041 --> 00:41:28.204
And they're benefiting from AI and the two work as, as counterparts.
00:41:29.284 --> 00:41:32.784
And that will be probably, a flow.
00:41:33.204 --> 00:41:42.197
And then there's another flow that's probably happening at the same time where humans are effectively building these frameworks to automate the other flow.
00:41:42.420 --> 00:41:48.041
And then the question then becomes, you know, in the not so distant future, how does that bear out?
00:41:48.657 --> 00:41:58.121
And then what flows are actually allocated to the agents, the agents and the agentic frameworks and the humans that are effectively designing those frameworks.
00:41:58.463 --> 00:42:05.471
And then what flows are going to sort of this, with a co pilot versus an organization with an autopilot.
00:42:05.701 --> 00:42:17.704
this seems like it aligns really well to what you're doing with Memra, uh, your company, where you're helping enterprises to leverage gigantic AI, and kind of scale out operations internally versus outsourcing.
00:42:18.224 --> 00:42:18.813
Correct.
00:42:18.987 --> 00:42:26.331
I think tactically there's some work that has been going, maybe offshore or to various outsourcing vendors.
00:42:26.454 --> 00:42:36.534
That I see as low hanging fruit that can kind of come back into the enterprise, but, brought back to the enterprise for AI to, for AI to take on.
00:42:36.728 --> 00:42:37.307
Fantastic.
00:42:37.307 --> 00:42:39.527
I really appreciate all these insights, Amir.
00:42:39.597 --> 00:42:41.197
It's been a wonderful conversation.
00:42:41.802 --> 00:42:54.514
As we wrap up, maybe if you could share with me any closing thoughts you have for You know, software engineers or AI practitioners listening who are looking to stay ahead and kind of get a jump on leveraging AI agents.
00:42:54.525 --> 00:42:55.364
What should they be doing?
00:42:55.532 --> 00:42:58.222
I think the, the answer to that is do.
00:42:58.697 --> 00:43:07.081
I mean, I think with a lot of things, maybe there are thinkers and doers, and they're not the same, like the thinkers are allocating tasks.
00:43:07.262 --> 00:43:10.422
They're contemplating and then they're allocating tasks to the doers.
00:43:11.027 --> 00:43:27.516
But here, specifically, I think the doers are the thinkers, and specifically there's a lot of thinking that you can do as an engineer as to how to sort of shape the future of agentic frameworks or the future of work just by doing.
00:43:27.824 --> 00:43:34.025
I think a lot of the very successful AI companies, the startups, they're developer led right now.
00:43:34.391 --> 00:43:45.284
I think this model of, delegating work, I think it doesn't necessarily work, it doesn't necessarily translate, with AI as well as it does with other areas.
00:43:45.494 --> 00:43:47.135
And that's okay in the short run.
00:43:47.144 --> 00:44:16.867
I think in the long run, yeah, you may want to, uh, benefit from, um, the returns to scale that come from delegation, but in the short run, there's so much to learn by interacting with these models, by developing the infrastructure to augment the models, by learning about the processes the organizations have, and sort of improving on those processes and linking them to the agentic frameworks and the effectively the LLMs, there's so much learning that can take place with the LLMs.
00:44:17.079 --> 00:44:23.079
and in fact, some of that learning can fuel these vertical go to market AI first companies.
00:44:23.579 --> 00:44:28.469
And then some of that learning, um, maybe speaks to the other point about vertical integration.
00:44:28.663 --> 00:44:30.472
And there's a lot of work to be done here.
00:44:30.853 --> 00:44:32.432
It's super interesting work.
00:44:32.432 --> 00:44:38.672
It's really kind of fun to be honest with you as well, both to produce and to consume this stuff is super exciting.
00:44:39.014 --> 00:44:39.534
Perfect.
00:44:39.664 --> 00:44:41.735
Well, thank you so much for coming on the show, Amir.
00:44:41.744 --> 00:44:43.894
It's been wonderful having this conversation with you.
00:44:43.905 --> 00:44:44.925
I know I've learned a lot.
00:44:44.925 --> 00:44:46.485
I hope our audience does as well.
00:44:46.864 --> 00:44:52.244
And for those who want to learn more, want to follow Amir's work on these topics, you can find him on LinkedIn.
00:44:52.255 --> 00:44:53.764
He shares a lot of incredible stuff there.
00:44:53.965 --> 00:44:55.454
And you can also check out Memra at memra.
00:44:55.485 --> 00:44:58.228
co, to learn more about AI and agentic systems.
00:44:58.543 --> 00:45:05.222
As always, thanks everyone for tuning in to Dev Interrupted, uh, and be sure to describe to our newsletter on Substack for deeper dives into the topics we discussed today.
00:45:05.632 --> 00:45:06.842
Amir, thank you so much for coming on.
00:45:07.152 --> 00:45:07.842
Thank you very much.
