رونوشت 🔗
WEBVTT
00:00:00.400 --> 00:00:02.640
I think it's really an unsolved problem, largely, you know.
00:00:02.799 --> 00:00:07.440
Because a lot of times people have way too much confidence in the agent and they get very lazy, and that's what concerns me the most.
00:00:07.679 --> 00:00:18.079
Imagine a skill or something that does an action uh but then to fetch more information to fulfill that, they get that from like an authoritative source, which could have been prompt injected.
00:00:18.160 --> 00:00:21.920
That is the invisible part where that's like not entirely obvious how that works.
00:00:22.320 --> 00:00:26.719
At the end of the day, the agent is gonna be able to negotiate the guardrail.
00:00:26.879 --> 00:00:31.039
Eventually, this uh access is gonna get breached.
00:00:31.120 --> 00:00:33.280
And that's the hard truth we have to accept.
00:00:33.520 --> 00:00:40.880
The dependency on the GitHub workflow is going away because your code is just not that precious.
00:00:41.280 --> 00:00:45.679
The agents will actually find a way to get into your website.
00:00:45.759 --> 00:00:51.840
But if you want to make things easier, cheaper, and faster, well, WebMCP lets you do that.
00:00:52.079 --> 00:00:58.560
We don't need to apply the last 25 years of software engineering to the tools of the next 25 years.
00:00:58.799 --> 00:01:00.079
You will be left out.
00:01:00.320 --> 00:01:01.920
You will be replaced.
00:01:02.880 --> 00:01:09.599
The AI Native Dev is a podcast for developers and engineering leads at the cutting edge of AI and agentic coding.
00:01:09.760 --> 00:01:20.000
Join your host, Sky Fajani, and meet Simon Maple every week as we chat with the most exciting voices in AI and tackle the biggest questions facing developers today.
00:01:20.400 --> 00:01:22.959
This is the AI Native Dev.
00:01:26.000 --> 00:01:28.879
Hey everyone, hope you're enjoying the episode so far.
00:01:29.200 --> 00:01:37.280
Our team is working really hard behind the scenes to bring you the best guests so we can have the most informative conversations about agentics development.
00:01:37.439 --> 00:01:42.799
Whether that's talking about the latest tools, the most efficient workflows, or defining best practices.
00:01:43.040 --> 00:01:46.640
But for whatever reason, many of you have yet to subscribe to the channel.
00:01:46.799 --> 00:01:53.599
If you're enjoying the podcast and want us to continue to bring you the very best content, please do us a favour and hit that subscribe button.
00:01:53.760 --> 00:02:01.120
It really does make a difference and lets us continue to improve the quality of our guests and build an even better product for you.
00:02:01.359 --> 00:02:03.120
Alright, back to the episode.
00:02:03.359 --> 00:02:04.879
Hi, Simon Maple here.
00:02:04.959 --> 00:02:08.000
Welcome to another episode of the AI Native Dev.
00:02:08.159 --> 00:02:19.599
We've just wrapped up another amazing AI Native DevCon in London, and while we were there, we got some
Load full transcript
really smart people together to talk about some of the most interesting topics around agentic development today.
00:02:19.840 --> 00:02:31.520
Coming up, we have my conversation with Netlify's Dana Lawson, author and educator Max Fertman, and Tesla's very own James Moss, talking about the future of AI development for web applications.
00:02:31.759 --> 00:02:33.840
But first, the subject is security.
00:02:34.000 --> 00:02:40.400
And I sat down with Sneaks Liran Tal, Joseph Katsialudas from GitHub, and John Gretzinger from Cisco.
00:02:40.560 --> 00:02:41.280
Enjoy.
00:02:42.080 --> 00:02:42.879
All of you, welcome.
00:02:42.960 --> 00:02:44.159
How are you all doing?
00:02:44.479 --> 00:02:45.759
Great, thank you for having us.
00:02:45.919 --> 00:02:46.159
Absolutely.
00:02:46.400 --> 00:02:47.199
Enjoying DevCon?
00:02:47.520 --> 00:02:48.000
Of course.
00:02:48.479 --> 00:02:49.680
Fantastic energy here today.
00:02:49.840 --> 00:02:50.800
Oh, it's good fun, isn't it?
00:02:50.879 --> 00:02:52.879
So why don't we just do a very, very brief uh intro.
00:02:53.039 --> 00:02:55.360
Tell us a little bit about yourself, 10-15 seconds.
00:02:55.439 --> 00:02:56.240
Uh let's start.
00:02:56.400 --> 00:02:57.120
Left to right.
00:02:57.360 --> 00:02:57.520
Sure.
00:02:57.680 --> 00:03:01.759
Uh John Götzinger, Cisco, uh largely background in firewall.
00:03:01.840 --> 00:03:06.800
Um, but recently, the last couple years more focused on agentec development and AI solutions.
00:03:07.280 --> 00:03:07.759
Lauran?
00:03:08.080 --> 00:03:14.319
Yeah, I've been doing uh dev turned into uh security and DevRel thanks to Simon, who has had that influence on my life.
00:03:14.479 --> 00:03:15.039
Thank you.
00:03:15.199 --> 00:03:19.360
Uh I've been doing some AI security research recently, uh, so that's been fun.
00:03:20.000 --> 00:03:26.240
Similar background to Loran, security turning more into DevRel, helping developers to ship secure code.
00:03:26.560 --> 00:03:26.960
Awesome.
00:03:27.120 --> 00:03:35.919
Well actually, Loran, you kind of mentioned um a little bit around the kind of like, you know, the the the NPM securities and vulnerabilities and things like that.
00:03:36.080 --> 00:03:45.120
Why don't we kind of like start there and we just let's talk a little bit about, I guess, vulnerabilities from a typical traditional software space to what we're looking at now?
00:03:45.280 --> 00:03:52.319
And um I know you know you you technique and others are are disclosing MCP uh CVEs as well.
00:03:52.639 --> 00:04:10.719
Is MCP security and AI security generally now um where NPM was like 15-20 years ago where dependencies coming out and people just using them because they feel like there's a massive performance increase, but they're just lacking in security skills?
00:04:11.360 --> 00:04:13.360
Um yeah, I think there's an aspect of that.
00:04:13.520 --> 00:04:17.759
To an extent, somewhat it's the same issues and somewhat somewhere new.
00:04:18.000 -->
00:04:23.519
So there's whether the MCP server, for example, that you're using is actually like vulnerable or not.
00:04:23.600 --> 00:04:30.160
So there's the aspect of the security of the MCP server itself, and then there's like the consumer side of it.
00:04:30.319 --> 00:04:34.639
So whether the MCP server you're using, are you like taking it from like official places?
00:04:34.800 --> 00:04:35.839
Is it trusted?
00:04:36.000 --> 00:04:41.600
Uh when MCP came out, it was uh very high-pitched and like let's use the latest.
00:04:41.680 --> 00:04:49.439
So people would just like run it off as like command strings with like you know, UVX or NPX, and then like the MCP server at latest.
00:04:49.519 --> 00:04:56.480
So they'd get like the latest thing, and you could like rug pull them, and they would just get like new, I don't know, dependencies or whatever with code in there.
00:04:57.120 --> 00:05:04.560
So from that aspect, we kind of like haven't learned to like instill the right practices when we're like consuming MCPs.
00:05:04.800 --> 00:05:09.839
Um and it's when we talk about maybe skills later, it's kind of like the same issue again and again.
00:05:10.160 --> 00:05:24.399
Um so there are some aspects of like AI supply chain and traditional like NPM package management supply chain that really, really affect and um and implement in the same way to MCP security.
00:05:25.040 --> 00:05:27.360
And we've been through this route right before, Joseph.
00:05:27.439 --> 00:05:32.160
Like, do you feel like we have that muscle that has just been kind of like it's it's already built there?
00:05:32.319 --> 00:05:40.879
It's we we know what we should be looking for, but we just need to apply the same methodologies to other things like skills and MCP to the yeah.
00:05:40.959 --> 00:05:42.319
Yeah, I absolutely agree with that.
00:05:42.480 --> 00:05:48.800
I think that the the speed of development is so fast, and we are all of us trying to make the most out of it.
00:05:48.959 --> 00:05:52.560
Therefore, security keeps catching up behind this linearly.
00:05:52.879 --> 00:06:02.800
Therefore, we definitely learn from learn from the past, but awareness is not there yet, so that we are starting deploying a new technology securely from the beginning.
00:06:03.120 --> 00:06:06.160
Um that's why we have this gap every single time, as Liran mentioned.
00:06:06.240 --> 00:06:13.360
Yeah, and I guess uh John, a large organization like Cisco, like obviously we're aware of the threats.
00:06:14.240 --> 00:06:31.839
What's the balance, do you think, between wanting developers to actually play with this technology so they get that advancement in genetic development, but also being sure you it's done at a pace that can allow for security to best practices and things to be put in place.
00:06:32.399 --> 00:06:36.959
Yeah, I mean, it's it's hard to speak for you know the whole work, but we kind of go team by team.
00:06:37.120 --> 00:06:48.240
When one team tries some pattern that works well, you know, we kind of try to apply that broadly to other teams, but uh I mean it's really an unsolved problem largely, you know, it's it's there's a new problem every week, a different tool, different thing.
00:06:48.319 --> 00:06:54.399
But we do try to restrict access to things like you know, open claw, for example, is one of the very few tools we banned in Cisco.
00:06:54.639 --> 00:06:57.439
Um I was like, oh, I can't even get to this, which is a good thing, right?
00:06:57.600 --> 00:07:02.639
But um, our security and trust org is very intentional about what they allow and don't when it comes to coding, though.
00:07:02.720 --> 00:07:05.040
So it helps us kind of just get guidance from them.
00:07:05.360 --> 00:07:05.759
Yeah, yeah.
00:07:05.839 --> 00:07:11.680
And do you and do you feel is is it very similar to kind of like, you know, talking of security and the adoption of dev security?
00:07:11.839 --> 00:07:14.800
Um, do you find it similar in the the adoption of AI?
00:07:14.959 --> 00:07:26.000
You'll have different groups of teams where some are absolutely eager, want to get started, others are complete skeptics, don't want to even touch it, and then probably the big bit in the middle is willing to use it, but they want to see the value first.
00:07:26.319 --> 00:07:26.720
Absolutely.
00:07:26.800 --> 00:07:28.480
Yeah, I could I'd see that so much.
00:07:28.560 --> 00:07:31.519
I'd say it's way less unlike the skeptics.
00:07:31.600 --> 00:07:35.199
We don't really have many of those left, I'd say, in Cisco, at least I haven't really found them.
00:07:35.680 --> 00:07:39.839
Um top-down is very, you know, supportive of AI, and we know how productive it can make us.
00:07:39.920 --> 00:07:44.480
At the same time, it can be dangerous, but again, we have intentional security policies around that.
00:07:44.560 --> 00:07:50.560
So, but yeah, it's it's largely people that are kind of never really been coders, but they want to try now because they can.
00:07:50.720 --> 00:07:54.079
Those are really the ones that are kind of the most dangerous because they put stuff out there.
00:07:54.160 --> 00:08:00.079
They don't understand the traditional developer security controls, like don't put public, you know, don't make your private project public.
00:08:00.160 --> 00:08:00.240
Yeah.
00:08:00.480 --> 00:08:04.560
These types of basic things that agents will just do um that they're not aware of.
00:08:05.040 --> 00:08:10.000
There's actually a uh an interesting data stat that I'd love to kind of like share with you both afterwards.
00:08:10.160 --> 00:08:25.040
But um you mentioned in a in a in a report, I think 83%, or this is maybe Cisco's data, 83% of enterprises plan to deploy agentic AI, but only 29% feel ready to do so securely.
00:08:25.279 --> 00:08:28.560
Uh I don't know if that's a stat that you're that you've that you've seen or Yeah, yeah, yeah.
00:08:29.600 --> 00:08:30.240
What do you what do you think?
00:08:30.319 --> 00:08:33.360
I want to start with you, John, but then we'll come across because this is a really interesting one.
00:08:33.519 --> 00:08:34.799
What does not ready look like?
00:08:34.879 --> 00:08:41.360
When we say when we say uh only 27, sorry, 29% feel ready to do so securely.
00:08:42.080 --> 00:08:45.600
What what what's a sign or a flag that says this is not ready?
00:08:46.159 --> 00:08:50.080
I mean, some teams don't even have traditional security controls in like their pipeline.
00:08:50.240 --> 00:08:51.120
So like that's number one.
00:08:51.200 --> 00:08:56.080
If you don't have that, it's it's just a disaster waiting to happen as a first thing.
00:08:56.240 --> 00:08:59.519
I mean, that's just a key sign, but I don't know.
00:08:59.759 --> 00:09:00.799
That's the basic one for me.
00:09:02.559 --> 00:09:15.679
A good signal that things are very chaotic at a company when they're adopting MCP is like basically having lots of adoption, different MCP servers, no one knows about them, people just rediscover them using latest.
00:09:15.840 --> 00:09:28.639
Um it's very fast to adopt, uh, as in like move fast, break things, but at the same time, it's it's early and not exactly uh as like good, confident uh security hygiene to basically adopt them.
00:09:28.720 --> 00:09:30.320
So what's the interesting?
00:09:31.440 --> 00:09:37.919
Well, I think at the at the very uh like we talk about like governance and policy, what if the first thing you want to have like is like observability, right?
00:09:38.000 --> 00:09:52.399
Like it's the thing that uh when we go back to talking about like how does this uh correlate with like NPM usage and like you know, traditional software development, if you don't really know like all the packages you're running, all the PyPIs packages and stuff across your team, you don't have that visibility.
00:09:52.720 --> 00:09:53.279
Visibility, yeah.
00:09:53.519 --> 00:09:58.320
Something that happens, like malicious packages, vulnerable packages, you know, new React, you know, zero day, whatever.
00:09:58.480 --> 00:10:01.519
If you don't really know where that's deployed, that's a problem to begin with.
00:10:01.600 --> 00:10:05.440
And that's kind of like the same parallel with MCPs and with skills later on and whatever.
00:10:05.519 --> 00:10:21.120
Like basically you want to have a deployed visibility layer for everything the team is running and your org is running, so you could really have control over if there's like a new vulnerability, new CVE, new malicious MCP server, whatever, you are able to actually like find it to begin with.
00:10:21.200 --> 00:10:23.919
And that's like those like core beginning points.
00:10:24.159 --> 00:10:26.960
But you shouldn't try to have it, you should do or do not.
00:10:27.039 --> 00:10:28.320
There is there is no try.
00:10:28.639 --> 00:10:29.679
Yes, there is no try.
00:10:30.000 --> 00:10:30.240
Indeed.
00:10:30.480 --> 00:10:31.039
Do or do not.
00:10:31.279 --> 00:10:35.919
But for those on audio only, Loran is wearing a Yoda hat, just so that it that doesn't sound ridiculous.
00:10:36.159 --> 00:10:37.279
Uh Joseph, how about you?
00:10:37.360 --> 00:10:39.039
What does not ready look like for?
00:10:39.919 --> 00:10:53.519
I feel what is uh where this stems from is from the fact that we are moving in the execution layer and AI is taking actions, it can send emails and do all these things that people are getting nervous about.
00:10:53.679 --> 00:10:59.360
So what is not is not ready feels, is visibility, governance, all these things that Lyra mentioned.
00:10:59.600 --> 00:11:07.279
I I think it's totally normal for people to have a lower risk appetite than accepting AI taking care of the execution layer.
00:11:07.519 --> 00:11:22.960
Therefore, if you see these uh cases of executives in in companies getting their emails being deleted, maybe other executives are getting nervous and the top-down gets OpenCloud being banned, which is the same as GitHub, by the way.
00:11:23.039 --> 00:11:25.759
We also don't have access to the website of OpenCloud.
00:11:26.000 --> 00:11:41.279
Therefore, I think not ready means that I don't feel empowered to leave these agents to take care of the stuff that we might lose clients for, we might get hurt reputation-wise, and all these cases.
00:11:41.600 --> 00:11:42.080
Got it.
00:11:42.320 --> 00:11:55.120
And interestingly, when we think about how we how we actually become more ready, and actually a lot of it is education, a lot of it is us, you know, making sure the developers on our team and on our organization know these types of attacks.
00:11:55.279 --> 00:12:05.120
And in fact, Joseph, the session that you literally just gave in the in the room, just behind that door uh behind you there, um the it was a very much an attack and then defend.
00:12:05.200 --> 00:12:13.440
So, like, let's try and work out how we can exploit, let's try and work out what the what the issues are, and then how can we write better prompts and how can we set up our system better to defend more against it?
00:12:13.519 --> 00:12:28.720
In fact, it's very similar to the to the Stranger Danger style sessions that that uh you and I both gave many times at Sneak and I think at uh at the recent conference uh Cisco Live in Amsterdam, I I did a little bit of a session there with uh uh with one of your colleagues as well.
00:12:28.960 --> 00:12:34.559
Um do you do you find when developers are more aware of the attacks, it actually helps them?
00:12:34.639 --> 00:12:36.080
It changes their behavior?
00:12:36.320 --> 00:12:41.919
Like when they understand the attacks, does it change the way they build, change the way they develop, change the way they write prompts?
00:12:42.159 --> 00:12:42.799
Definitely.
00:12:42.960 --> 00:12:48.000
Um that that's another workshop I gave in uh in Brooklyn in the previous conference of Tesla there.
00:12:48.159 --> 00:12:55.360
Uh, I think that
when developers are aware of those, they are gonna, they are not gonna trust AI to defend themselves.
00:12:55.519 --> 00:13:03.919
Um, they are gonna build applications that have mitigated controls there, such as, for example, output filtering or dual LM, input filtering.
00:13:04.080 --> 00:13:14.480
In general, they will not give access to anything sensitive to the AI agents because eventually this uh access is gonna get bridged.
00:13:14.639 --> 00:13:24.240
And by that, through sandboxes and all the zero trust um measures out there, sure they are aware of those, I feel they are gonna be able to do so.
00:13:24.399 --> 00:13:26.399
Otherwise, they are gonna find out the hard way.
00:13:26.480 --> 00:13:28.480
And that's the hard truth we have to accept.
00:13:30.159 --> 00:13:31.919
Um anything you gonna add?
00:13:32.399 --> 00:13:45.279
Um I I would just add I I think developers who are security-minded are probably gonna be in that uh like box where they're always like aware of what could go wrong.
00:13:45.600 --> 00:13:56.080
But even after many attacks and stuff, like I I still feel like people do like NPX CC usage latest, just like watch the cloud code usage uh stuff.
00:13:56.240 --> 00:14:07.200
And like uh I'm like completely scared to run that because I know the moment I do NPX what could happen, like when I bring all of those dependencies, and maybe that was like you know a compromised package or something.
00:14:07.440 --> 00:14:15.039
Um so I think it's it's gonna be hard to trust that devs and also beyond devs, just like developer citizens.
00:14:15.120 --> 00:14:17.519
Like now everyone's like empowered to be a developer, right?
00:14:17.679 --> 00:14:22.480
So you have people who do not even understand SDLC and just build apps on top of cloud code and whatever.
00:14:22.720 --> 00:14:28.720
And it's gonna be hard to expect them to follow security guidelines and practices.
00:14:28.879 --> 00:14:30.799
So we should put all of that out.
00:14:30.960 --> 00:14:38.399
And I think like, for example, Joseph is doing like a really good job of like you know, advocating all of those things, um, which is great for their community, for the education.
00:14:38.639 --> 00:15:07.919
At the same time, we have to like build that into the workflow itself, like have proper isolation and sandboxing, and how do we treat secrets and how do we you know secure MCPs and skills from like a very uh first principles kind of aspect so that that is already by itself sort of like taken care of and like invisible to devs because otherwise I don't I'm not entirely sure that people are gonna be stopping to think about what could go wrong versus wanting to ship uh new features.
00:15:08.080 --> 00:15:11.440
So hopefully the harnesses do get better.
00:15:11.519 --> 00:15:12.559
Uh that's something I've seen.
00:15:12.720 --> 00:15:18.960
We'll talk about it in my talk later, but um it's it's definitely something we have to bake into agentic tooling.
00:15:19.279 --> 00:15:20.080
Yeah, agree.
00:15:20.399 --> 00:15:26.320
Uh we talked a little bit about visibility, and which I think is a key piece of um of security gener uh generally.
00:15:26.480 --> 00:15:40.240
John, um Cisco's AI defense tool um now includes an MCP gateway that actually allows you to be able to track and identify uh messages that are being sent from agent to to tool.
00:15:40.480 --> 00:15:43.440
Um this is really, really important to actually identify, okay, what's happening?
00:15:43.519 --> 00:15:47.519
Is it is there is there some badness happening from the MCP or something like that?
00:15:47.679 --> 00:15:52.159
Uh at what stage though is this kind of like visibility versus security?
00:15:52.240 --> 00:16:00.480
Like, is this is this gonna stop issues or is this gonna at least show us something is happening and we need to stop using these types of things?
00:16:00.720 --> 00:16:01.519
Yeah, I mean it's hard.
00:16:01.600 --> 00:16:11.200
I think if you can somehow observe intent, I think that's really, you know, what was the intention going in, what the agent was trying to do, and what systems did it reach out to or talk to?
00:16:11.600 --> 00:16:13.039
Was it really necessary to talk to the systems?
00:16:13.200 --> 00:16:17.679
If you can focus on observing those patterns, it's good, but that's really hard to kind of set that up.
00:16:17.840 --> 00:16:19.360
I think it's still kind of new.
00:16:19.519 --> 00:16:21.679
Um, but like that's where I've been trying to focus.
00:16:21.759 --> 00:16:26.960
Like, hey, I need to add this rule to a firewall, I want an agent to do it for me or whatever, and has to do some LDAP lookup.
00:16:27.039 --> 00:16:31.919
It's like, I don't want you talking to LDAP for this, you should talk to this other system, like instilling that in and observing the patterns.
00:16:32.399 --> 00:16:40.320
Does it feel like still kind of like quite a like a when you look at like a web firewall or something like that, where you're constantly adding rules because things, you know, everyone it's constantly changing.
00:16:40.480 --> 00:16:46.639
Is this something that is we need to just get comfortable with it being kind of a a little bit more reactive and identifying these things are happening?
00:16:46.720 --> 00:16:47.600
We need to adjust.
00:16:48.000 --> 00:16:53.679
Is it is that I mean I wouldn't say in production reactive, but more, you know, maybe sandbox environments.
00:16:54.720 --> 00:16:58.480
To pull it there, test it out, see what it tries to do before you put it in production, observe that pattern.
00:16:58.639 --> 00:17:06.880
I mean, that's obviously what you should do first, but unfortunately, I think largely our learnings come from the reactive of it happening, and then we put our role in place.
00:17:06.960 --> 00:17:08.880
So that's just kind of how security is.
00:17:09.119 --> 00:17:10.559
But you can try to be proactive.
00:17:10.799 --> 00:17:11.839
It's just not
easy.
00:17:12.240 --> 00:17:21.759
Um so I'm sure during the next couple of days we're going to be talking about things like OWASP top tens, and there's an agentic, uh, there's an agentic applications uh top 10.
00:17:21.839 --> 00:17:25.119
I think there's a new one coming out for skills as well, skills development, which is super interesting.
00:17:25.200 --> 00:17:27.680
I think it's coming out Q3, Q4 time maybe.
00:17:27.839 --> 00:17:29.440
Uh very, very interesting.
00:17:29.680 --> 00:17:37.839
When we look at some of the new styles of attacks that we need to be thinking about, um, not just during development, but production as well.
00:17:38.079 --> 00:17:49.920
What would you say are kind of, when we go down the line again, start with you, uh maybe Joseph, what would you say are the are the are the uh the most interesting attacks today that we're not paying enough attention to?
00:17:50.079 --> 00:17:52.799
So the things that exist that developers don't know enough about.
00:17:53.119 --> 00:17:55.599
For me, this is prone injection uh by far.
00:17:56.000 --> 00:17:58.400
I believe that it's a risk that we have to accept.
00:17:58.799 --> 00:18:09.839
Because if we constrain too much the input there, maybe we don't give the chance to people use AI in the way that they want to use, which is used through natural language.
00:18:10.160 --> 00:18:20.079
Therefore, uh I think we should be taking active steps towards the direction of what happens from the moment that AI gets that information.
00:18:20.240 --> 00:18:29.119
Therefore, um assuming the worst that maybe sensitive data can be licked, we can apply output filtering.
00:18:29.359 --> 00:18:43.279
Before that, we can use the technique of dual LLM or LLM assiduity, therefore, second LLM with instructions to check the work of the first LLM, which is something that um Microsoft will have dual production.
00:18:43.599 --> 00:18:48.160
And by the way, when I mentioned that it's a risk that we have to accept, that's something that we did as well.
00:18:48.240 --> 00:18:49.920
Yeah, we accepted that risk.
00:18:50.160 --> 00:19:15.440
And for that, we made sure that uh since AI can be steered in ways that we don't expect through natural language and other approaches from attackers, it's all about not giving access to data that it shouldn't have, in the same way an employee shouldn't have root access or and apply the least privilege principle in that case.
00:19:15.599 --> 00:19:20.000
And at the same time, making sure that uh these actions are sandboxed.
00:19:20.160 --> 00:19:29.839
Because at the end of the day, the agent is gonna be able to negotiate the guardrail, is gonna be able to escape from stuff that you put around.
00:19:30.079 --> 00:19:48.480
Therefore, prompt injection is a risk that can come from external, but also from internal, through the execution layer instructions that we want to give internally in order to do our work faster, maybe, or become more productive, achieve business outcomes in general.
00:19:48.720 --> 00:19:56.559
And even from the inside, we can prompt inject in ways that we don't expect, or intentionally or unintentionally.
00:19:56.720 --> 00:20:01.680
Therefore, for me, Simon, this is the most interesting risk that AI brings uh by far.
00:20:02.240 --> 00:20:08.480
And people and ourselves, we are getting slowly, slowly more exposed into it and more educated, learning the hard way.
00:20:08.720 --> 00:20:12.160
And sandboxing is like something which just needs to become a default.
00:20:12.240 --> 00:20:21.440
And I probably tell more people about it than I do it myself, which is maybe a uh a confession I'm not happy to make, but yeah, it's something that we should just be natural.
00:20:21.519 --> 00:20:21.759
Yeah.
00:20:22.000 --> 00:20:22.880
How about you, Loran?
00:20:22.960 --> 00:20:28.000
What's your what's your favorite uh you have to pick one, what's your Yeah.
00:20:28.160 --> 00:20:30.640
Um I'm gonna try and squeeze two of them.
00:20:30.720 --> 00:20:37.680
I'm gonna build on Joseph's uh prompt injection, but I think indirect prompt injection is an interesting part of it.
00:20:37.839 --> 00:20:48.799
So imagine a skill or something that does an action, uh, but then to fetch more information to fulfill that, they get that from like an authoritative source, which could have been prompt injected.
00:20:48.960 --> 00:20:54.720
So I think that is like the invisible part where that's like not entirely obvious to how that works.
00:20:54.880 --> 00:21:10.720
And also the more you go into like complicated workflows and have to like trust this like third-party upstream kind of like indirect uh input and context into the action, um that becomes like a bit more messier to figure out how to clean the data right.
00:21:11.039 --> 00:21:12.400
And how do you even defend that, right?
00:21:12.480 --> 00:21:18.880
Because let's let's say I'm I'm reaching out to let's say a GitHub issue or something, because you know, I I need some more data.
00:21:18.960 --> 00:21:21.599
Yeah, it's a perfectly reasonable place to go.
00:21:21.839 --> 00:21:22.160
Correct.
00:21:22.480 --> 00:21:28.000
But it's how does, you know, anyone can just like write some stuff on a GitHub issue or a get or a pull request or whatever.
00:21:28.160 --> 00:21:37.759
And as long as it's rate, as long as it's created, you know, even with a trusted source, you could get potentially malicious data that someone's just putting that in.
00:21:37.839 --> 00:21:40.240
So how do we even think about protecting it?
00:21:40.480 --> 00:21:41.039
It's a hard one.
00:21:41.119 --> 00:21:44.640
I've I've seen a couple of like examples of how to try and defend that.
00:21:44.880 --> 00:21:59.759
One of them is um just recently, they're like just like this research that actually like runs, like actually like uh proxies the fetch of the data to like a dummy, like isolated proxy, and then watches the behavior of what happened.
00:22:00.160
--> 00:22:01.759
So it's not the actual one.
00:22:01.839 --> 00:22:06.160
It's like a it's like a proxied agent that you can't really protect it.
00:22:06.240 --> 00:22:11.359
So you actually tell it to do it, and then you watch what happens afterwards in like an isolated environment and then judge that.
00:22:11.519 --> 00:22:17.839
So you kind of have to kind of f like a honeypot, you have to like follow it and then like judge it because you don't really know in advance what would happen.
00:22:18.079 --> 00:22:22.319
And also just because it runs well then doesn't necessarily mean in a day it's gonna come into changes.
00:22:22.640 --> 00:22:23.039
True, true.
00:22:23.200 --> 00:22:24.400
And there's like so many, exactly.
00:22:24.480 --> 00:22:31.759
And there's like you could like fetch code and like what happens when you fetch code is like, you know, not I don't like understand the code, maybe that's like malicious, maybe not.
00:22:31.920 --> 00:22:42.480
Um there are some signals like when we did uh the toxic skills research at Snake, we saw like that some like suspicious downloads would have like zip files that are password protected.
00:22:42.640 --> 00:22:46.640
So if you put something on like GitHub that's password protected, like why would you do it?
00:22:46.799 --> 00:22:53.119
And like the reason is to like evade antiviruses and like malware protections that need to like uh inspect what's inside.
00:22:53.200 --> 00:22:54.240
So you password protected.
00:22:54.880 --> 00:22:58.319
Um so there's like signals, but again, there's like signals, they're not entirely there.
00:22:58.480 --> 00:23:05.039
And there's like like those, there's like a bunch of like uh methods around it, but essentially it's like it's a hard problem to figure out.
00:23:05.200 --> 00:23:10.079
So it's behavioral analysis plus whatever the agent is doing in its business logic.
00:23:10.799 --> 00:23:11.039
Awesome.
00:23:11.119 --> 00:23:12.240
So what what what do you think, John?
00:23:12.400 --> 00:23:13.440
What's your what's your favorite?
00:23:14.799 --> 00:23:22.640
I mean, I think uh being intentional about the questions that you ask is really something I want my developers to learn because but that takes experience, right?
00:23:22.720 --> 00:23:23.839
Knowing what questions to ask.
00:23:23.920 --> 00:23:30.319
Uh because a lot of times people have way too much confidence in the agent and they get very lazy, and that's what concerns me the most is just people letting it do it.
00:23:30.400 --> 00:23:31.759
And then they just ask it, is this secure?
00:23:31.920 --> 00:23:33.359
And of course it'll be like, I reviewed it.
00:23:33.440 --> 00:23:33.920
Yes, it is.
00:23:34.000 --> 00:23:39.599
And it's like, no, what you should be able to know what questions to ask it, like what did you check to afford this?
00:23:39.759 --> 00:23:43.440
And like again, understanding the architecture, how it fits into when it's deployed, right?
00:23:43.519 --> 00:23:51.279
It might not know it's sitting in front of some load balancer which strips headers, and you know, so it's like, oh, it's not it's secure here because this could strip and you know assumptions about that.
00:23:51.519 --> 00:23:54.240
Yeah, I think that's uh overagency in OS variable.
00:23:54.559 --> 00:23:55.599
There it is, then that's then that one.
00:23:55.759 --> 00:24:12.319
That one that one in a year from now, we're gonna be uh I know it's ridiculous doing things beyond a week, but in a year from now, um organizations are gonna be deploying AI agents into production much, much more, using it in development much, much more.
00:24:13.039 --> 00:24:19.920
But do you think we're gonna be uh more or less secure in our usage and in our deployment?
00:24:22.079 --> 00:24:24.240
I think more, uh especially in the enterprise setting.
00:24:24.400 --> 00:24:26.960
Maybe in the more smaller individual, probably less so.
00:24:27.279 --> 00:24:31.440
But in enterprise, it's it's getting locked down a lot more and there's more policies and tolling coming out.
00:24:31.839 --> 00:24:34.559
But the tolling costs money, so individuals you know, don't really purchase that.
00:24:34.640 --> 00:24:37.839
But I have hope for enterprise, but the individual is gonna be tough.
00:24:38.640 --> 00:24:39.759
So more from Germany.
00:24:40.319 --> 00:24:47.119
I would say more secure, but at the same time, we're gonna watch some uh I think very chaotic like incidents out there.
00:24:48.640 --> 00:24:50.559
I agree, I agree with both.
00:24:51.119 --> 00:24:53.920
Uh I think we are in an interesting world.
00:24:54.240 --> 00:24:58.960
Uh again, technology is growing exponentially and we are trying our best to catch up linearly.
00:24:59.279 --> 00:25:06.079
Therefore, I'm expecting some chaos out there, but good chaos in the sense that it's gonna be helpful.
00:25:06.480 --> 00:25:10.160
Uh I don't think mature enterprises are gonna face problems.
00:25:10.400 --> 00:25:22.480
I think that um small medium enterprises that maybe when I move faster, they might be on the biggest risk because they are facing the least amount of they have the less the least maturity.
00:25:22.640 --> 00:25:31.119
And at the same time, vibe coders out there uh might be exposing themselves to a lot of security problems.
00:25:31.680 --> 00:25:32.480
Let's see.
00:25:33.039 --> 00:25:35.519
Joseph, Laurent, John, it's been an absolute pleasure.
00:25:35.680 --> 00:25:36.480
Thank you very much for joining.
00:25:37.440 --> 00:25:37.839
Thank you.
00:25:39.359 --> 00:25:44.880
Really fascinating conversation there, and from such smart people in our security industry today.
00:25:45.039 --> 00:25:49.200
Um let's change tack a little bit and talk about all things web AI.
00:25:49.599 --> 00:25:54.400
Hi there, Simon Maple here, uh, hosting another episode at the AI Native DevCon.
00:25:54.640 --> 00:25:56.960
Uh, and joining me here is another amazing panel.
00:25:57.039 -->
00:26:01.039
Uh, we have Dana Lawson, the Chief Technology Officer at Netlify.
00:26:01.119 --> 00:26:08.079
We have Max Fertman, the founder and director of Codemia, and we have James Moss, a member of technical staff at TESL.
00:26:08.240 --> 00:26:10.880
And we're going to be talking about AI on the web today.
00:26:10.960 --> 00:26:13.119
So, first of all, welcome to the episode.
00:26:13.440 --> 00:26:13.759
Thank you.
00:26:14.640 --> 00:26:16.960
I think everyone's first time on Oh no, sorry.
00:26:17.039 --> 00:26:18.960
You did a session with Guypone, right?
00:26:19.200 --> 00:26:20.400
So WebMCP.
00:26:20.640 --> 00:26:25.440
WebMCP, and I think uh we had we had um who did we have?
00:26:25.599 --> 00:26:28.880
We had someone on the uh someone from Netlify on one of the earlier.
00:26:29.440 --> 00:26:30.240
Sean Roberts?
00:26:30.319 --> 00:26:32.079
No, I think it's been Matt Billman?
00:26:32.240 --> 00:26:33.759
Uh it was, it was Matt Bilman.
00:26:34.640 --> 00:26:36.240
Yes, I better do better than him.
00:26:36.480 --> 00:26:39.920
So we're gonna send this to him and uh and tell him all the ways that you're better than him.
00:26:40.319 --> 00:26:42.559
You definitely do because it's review cycle right now.
00:26:42.720 --> 00:26:44.079
Oh, excellent, we'll do that, absolutely.
00:26:44.240 --> 00:26:44.559
Awesome.
00:26:44.720 --> 00:26:47.119
Well, what do we start off with, kick off with a little bit of intros.
00:26:47.200 --> 00:26:51.920
Uh we go left to right here, um, and uh tell us a little bit about yourselves and your role.
00:26:52.400 --> 00:26:55.839
Well, um, as you said, I'm Dana Lawson, CTO of Netlify.
00:26:55.920 --> 00:27:00.160
Um I've been in tech for, well, I don't even want to say, since the 90s.
00:27:00.400 --> 00:27:04.799
Uh I've been obsessed with developer tools and platforms for decades now.
00:27:04.880 --> 00:27:18.160
I started my career um as an individual contributor, slinging code, and have been working in this developing for developers system since New Relic, GitHub, Heptio, Envision, and now at Netlify.
00:27:18.240 --> 00:27:21.440
So what a set of set of companies to work for as well.
00:27:21.839 --> 00:27:22.319
I know, right?
00:27:22.400 --> 00:27:22.960
There's some good ones.
00:27:23.200 --> 00:27:23.599
Yeah, yeah.
00:27:23.680 --> 00:27:24.319
And Max?
00:27:24.559 --> 00:27:28.319
Well, before the AI era, so I was a mobile and web developer.
00:27:28.400 --> 00:27:29.200
So I was doing apps.
00:27:29.359 --> 00:27:31.599
I've been doing websites for 30 years.
00:27:31.839 --> 00:27:37.279
And mobile app since Nokia, BlackBerry, I'm not sure if you remember that, Symbian and all the stuff.
00:27:37.359 --> 00:27:40.880
And so I've been a developer, also a speaker.
00:27:41.039 --> 00:27:48.319
Uh I've been to 75 countries doing speaking and training, doing trainings, and also a book author.
00:27:48.480 --> 00:27:50.880
So trying to explain new stuff around.
00:27:51.039 --> 00:27:52.480
And now it's AI, everything is AI.
00:27:52.720 --> 00:27:55.279
And what's the country that you haven't been to that you really want to?
00:27:55.839 --> 00:27:57.359
There are a lot because there are 200.
00:27:57.519 --> 00:27:58.160
So yeah.
00:27:58.319 --> 00:28:02.559
But in Europe, most of I think I have all the European, all the South American countries.
00:28:02.960 --> 00:28:05.359
Is there one that you haven't checked off yet that you really want to go to?
00:28:05.519 --> 00:28:08.720
No, I I mean it can be uh South Korea.
00:28:08.960 --> 00:28:09.519
Never been there.
00:28:09.920 --> 00:28:10.319
Nice, nice.
00:28:10.559 --> 00:28:10.799
Yeah.
00:28:11.119 --> 00:28:12.079
Vietnam, maybe.
00:28:13.039 --> 00:28:14.240
Nice, nice, nice.
00:28:15.039 --> 00:28:15.680
James.
00:28:16.000 --> 00:28:16.400
Hey.
00:28:16.640 --> 00:28:20.240
Um I um have been building on the web since 1999.
00:28:20.319 --> 00:28:25.759
So built my very first GeoCity site back in back in the day for my planetarian gang uh clan.
00:28:25.839 --> 00:28:27.519
Does anyone remember Planetarian?
00:28:27.680 --> 00:28:28.799
No, just me?
00:28:30.079 --> 00:28:35.440
Um yeah, and then for the last past 20 years, kind of been building uh products mostly for startups.
00:28:35.680 --> 00:28:42.559
Right now, I'm a member of technical staff at TESL, so I'm working on our registry product, amongst many other things, uh, as you kind of have to at startup.
00:28:42.640 --> 00:28:46.079
So yeah, love building on the web, love building products that customers love.
00:28:46.640 --> 00:28:51.759
It sounds weird, but we've got over a century of development just between the four of us.
00:28:52.640 --> 00:28:53.039
I know.
00:28:53.119 --> 00:29:00.160
I I uh did a uh recent talk and it was 150 years, and it was just three of us, and I was just like, oh my gosh, you guys are old.
00:29:00.559 --> 00:29:01.440
Yeah, yeah, yeah, yeah.
00:29:01.680 --> 00:29:02.160
Awesome.
00:29:02.240 --> 00:29:04.319
So actually, Dana, why don't we start with you?
00:29:04.480 --> 00:29:07.359
Like uh you mentioned GitHub, Netlify.
00:29:07.440 --> 00:29:11.759
Um, that's like two incredible like developer platform companies.
00:29:12.000 --> 00:29:18.480
Um what would you say is different about building the web platform versus a developer platform?
00:29:18.880 --> 00:29:23.359
I think the difference is really the web is a lot simpler in a lot of ways.
00:29:23.519 --> 00:29:29.359
Um, there's not as many gotchas that you have to consider except caching, and that's really freaking hard.
00:29:29.599 --> 00:29:33.359
I think the big hard challenge with web is honestly just caching.
00:29:33.440 --> 00:29:34.240
Trust me on this.
00:29:34.319 --> 00:29:34.960
It's hard.
00:29:35.119 --> 00:29:38.079
Um, but a development platform, it's a lot more open-ended.
00:29:38.160 --> 00:29:44.480
People can build anything they dream of, and it's usually something very technical
that's feeding one of these experiences.
00:29:44.640 --> 00:29:52.160
And so the real shift is creating this obfuscation of stuff that doesn't matter to a web developer and to how you power the web.
00:29:52.240 --> 00:29:54.799
And so finding that line, but very similar.
00:29:54.880 --> 00:30:00.880
And with the rise of AI, the lines are definitely not they're they're blurry.
00:30:00.960 --> 00:30:04.400
You know, development platform, web platform, it's an AI platform now.
00:30:04.480 --> 00:30:08.400
Um it's just how how deep in what you're building do you need to actually go?
00:30:08.640 --> 00:30:13.119
And my hot take is you don't need a lot of the nonsense for building experiences.
00:30:13.279 --> 00:30:19.599
If you're building a tech company, yes, maybe, but most people are just building applications and websites.
00:30:21.119 --> 00:30:26.319
Also, I think that um users never care too much about web or native.
00:30:26.400 --> 00:30:28.720
They just care about using the experience.
00:30:28.799 --> 00:30:28.880
Yeah.
00:30:29.279 --> 00:30:29.599
That's right.
00:30:30.000 --> 00:30:31.359
Sometimes they don't even know.
00:30:31.680 --> 00:30:32.799
I mean it's a QR code.
00:30:32.880 --> 00:30:33.440
They use the Q.
00:30:33.920 --> 00:30:35.759
Scan the QR code and they use the app.
00:30:35.920 --> 00:30:38.960
I don't know if it's web or native or what.
00:30:39.119 --> 00:30:39.359
Yeah.
00:30:39.599 --> 00:30:42.000
Any differences that you uh you code to?
00:30:42.319 --> 00:30:48.319
Um yeah, I guess when it comes to those uh those kind of development tools, you kind of want to build a sandbox, right?
00:30:48.400 --> 00:30:51.839
Where you want to build all these tools, this environment that people can go and play.
00:30:52.000 --> 00:30:54.160
You don't want to have those kind of guardrails in place.
00:30:54.319 --> 00:30:56.400
You want to enable that sort of creativity.
00:30:56.559 --> 00:31:03.279
Um like Dana said, you kind of want to enable folks to experiment and get the tools out of the way, right?
00:31:03.440 --> 00:31:09.680
Build something interesting, build something fun, um, something that you're kind of proud of, uh, and not put those guardrails up.
00:31:10.000 --> 00:31:10.640
Yeah, cool.
00:31:10.880 --> 00:31:14.799
You uh we talked a little bit about kind of like leaning into a little bit like agent experience there.
00:31:14.880 --> 00:31:17.200
Um and this is something that I think Netlify coined, right?
00:31:17.279 --> 00:31:17.680
Agent experience?
00:31:17.920 --> 00:31:18.160
That is right.
00:31:18.400 --> 00:31:18.640
Yeah, yeah.
00:31:18.720 --> 00:31:20.880
So tell us a little bit about um that concept.
00:31:20.960 --> 00:31:31.039
What does it practically mean for when you're you know developing uh an app and you need to think about what the agent experience is like, how does that differ from?
00:31:31.680 --> 00:31:34.480
I mean, many of us don't have green-field applications.
00:31:34.640 --> 00:31:41.039
We're dealing with a lot of stuff that was written for humans, and now with agents it has to be at machine scale.
00:31:41.200 --> 00:31:43.440
And so rethinking about how you approach that.
00:31:43.519 --> 00:32:03.359
You know, for instance, with CLIs, a lot of it's an interactive, and you can write scripts, bash scripts around to utilize the CLI, but the way that it was codified was for human eyes and for sequential movements of data and systems versus you know this new world that we live in where agents just work in parallel.
00:32:03.519 --> 00:32:08.559
Shoot, they work in all directions, and now you have different boundaries that you didn't have before.
00:32:08.799 --> 00:32:08.960
Yeah.
00:32:09.200 --> 00:32:19.920
Well, to say like a you know it's actually funny when I when I play with Tesla, actually, James being here, when I play with Tesla, half the time I'm like, oh yeah, I'm I'm just using, I'm gonna I'm gonna use it through Claude.
00:32:20.240 --> 00:32:22.880
Uh and then I'm like, oh hang on, is this using the CLI?
00:32:23.039 --> 00:32:24.000
Is it using MCP?
00:32:24.079 --> 00:32:26.799
And I just realized actually I don't know, and I don't care.
00:32:27.119 --> 00:32:27.599
I don't even care.
00:32:27.759 --> 00:32:28.559
Yeah, I don't think it matters.
00:32:28.880 --> 00:32:29.519
You d it doesn't matter.
00:32:29.599 --> 00:32:38.160
And I think really agent experience is about thinking about how those agents interact with each other, and again, utilizing intent.
00:32:38.319 --> 00:32:46.960
And so agent experience is truly just going back and re-looking at some of the systems that you built, or how you're gonna build the new ones, knowing that they're gonna be used at machine speed.
00:32:47.119 --> 00:32:53.519
And so thinking about that, and it's like you're trying to, you know, run out all your threads, you're trying to kill the systems.
00:32:53.599 --> 00:32:54.640
Well, you're not trying to.
00:32:54.720 --> 00:33:04.319
So agent experience for me really comes down to two parts is putting in guardrails, writing intent with specs, and making it human readable as well as machine readable.
00:33:04.720 --> 00:33:06.720
And what does that mean when we talk about machine readability?
00:33:06.799 --> 00:33:14.640
Because I think for us, you know, documentation, well-documented APIs and things like that, is it the same levels of documentation that an agent needs?
00:33:14.880 --> 00:33:17.599
Or do agents need something else to understand how to interact with it?
00:33:17.839 --> 00:33:19.200
It's the way that it's actually written.
00:33:19.279 --> 00:33:22.160
Instead of doing gits and posts, agents don't understand that.
00:33:22.240 --> 00:33:26.240
When you want to understand what's actually being pushed through, it's writing in.
00:33:26.559 --> 00:33:28.559
I I go back to some of the basics of Python.
00:33:28.720 --> 00:33:31.759
When I first learned Python, I was like, oh my
God, this is so easy.
00:33:31.920 --> 00:33:36.640
Um and a lot of machine learning was based upon Python because it's understandable.
00:33:36.720 --> 00:33:39.200
It's not anything that you have to go and compile.
00:33:39.359 --> 00:33:40.000
You can just read it.
00:33:40.079 --> 00:33:45.680
And I think a lot of when we talk about what agents need, they want that human view because humans are speaking to the agent.
00:33:45.839 --> 00:33:49.920
So you have to speak in these parameters that aren't in machine speak.
00:33:50.079 --> 00:33:55.680
It's it's now spoken in what's the intent and the experience that you're gonna have ultimately.
00:33:55.839 --> 00:33:56.880
That's really the shift.
00:33:56.960 --> 00:33:57.839
It's not magical.
00:33:57.920 --> 00:34:00.960
It's just it's gonna feel like dumbing it down, frankly.
00:34:01.119 --> 00:34:01.759
That's how I feel.
00:34:01.839 --> 00:34:06.720
I'm like, oh man, I've got to make this a lot more clear so that the agent can take the correct action.
00:34:06.880 --> 00:34:13.039
Because when it's not, it's gonna have to parse through so much to figure out what those endpoints, what those services are trying to do.
00:34:13.519 --> 00:34:14.559
Yeah, very much so.
00:34:14.719 --> 00:34:20.960
I think coming back to kind of the CLI point you were talking about there, about like thinking about um making those suitable for agents.
00:34:21.119 --> 00:34:31.039
Um they are reading, like you say, they are reading um things like help documents in terms of um how to use those TLIs, you know, passing dash H in on the command line to read that.
00:34:31.119 --> 00:34:52.000
Um but I think a lot right now a lot of people are still focused on human experience and they're not thinking about the agent experience, um, and they're not detecting that, hey, this TLI is being run by an agent, not by a human, and actually I need to tailor the experience for that agent and give it perhaps more information, less information, allow it to kind of navigate through the information that you're getting out of that CLI as well.
00:34:52.320 --> 00:34:55.199
It's like Max was saying, like, it doesn't matter anymore.
00:34:55.360 --> 00:35:00.400
I mean, somebody was asking me, they're like, well, what if you want to get in there, look under the hood, because you're a developer?
00:35:00.559 --> 00:35:03.760
I I would ask them, why are you wasting company money if nothing's broke?
00:35:04.000 --> 00:35:05.440
Like, why are you spending your time?
00:35:05.679 --> 00:35:07.760
Be curious, like be curious.
00:35:08.079 --> 00:35:21.119
But this this version of like spelunking for spelunking's sake is really just, I think, a personal fear of developers losing their identity because now anybody can participate because the agents are doing the heavy lifting.
00:35:21.280 --> 00:35:29.199
So now when you go and you you look at what's happening, humans are going, oh my gosh, where where's my where's where's my purpose now?
00:35:29.760 --> 00:35:31.760
And let's take it one level deeper.
00:35:32.239 --> 00:35:32.719
Code.
00:35:33.039 --> 00:35:44.159
Obviously, and I I was chatting with uh Ryan on another episode, actually, uh Lapopoli, and and we were talking about code and the human readability of code, the importance of that versus agent readability.
00:35:44.480 --> 00:35:52.800
And the parallel I drew was, you know, years, years ago when we were writing an assembly language and binary.
00:35:53.440 --> 00:35:56.320
This is machine, machine readable, not human readable.
00:35:56.400 --> 00:36:06.480
But now we've uh abstracted you know into different languages where actually the maintainability was more important, and so we made things intentionally more uh human-readable than machine readable.
00:36:06.800 --> 00:36:08.719
But now we're getting into the AI space.
00:36:09.199 --> 00:36:11.760
We're actually going almost like the need is the other way around.
00:36:11.840 --> 00:36:23.599
We're actually something it's more important for it to be agent readable than human readable, although the human readability, I guess, it's more about levels of trust and level, you know, when we need to look at it.
00:36:23.840 --> 00:36:31.840
Where would you say you are, I guess, individually today in terms of the needing something to be like code to be agent readable or human readable?
00:36:32.320 --> 00:36:35.760
Well, I think that um let me give you an example on the web.
00:36:35.840 --> 00:36:36.800
Let's say the web.
00:36:36.960 --> 00:36:39.360
So I'm writing a book, and the book is on vanilla web.
00:36:39.599 --> 00:36:46.639
I was always an advocate of using vanilla web, but why are we using React these days with coordination, for example?
00:36:47.199 --> 00:36:50.320
So is React really needed at one point?
00:36:50.719 --> 00:36:55.920
So because uh React was actually created for developer experience, for a human developer experience.
00:36:56.079 --> 00:36:57.360
It was actually needed.
00:36:57.599 --> 00:37:03.280
But when you think about that, well, now we have other problems, you know, security issues with dependencies and all that stuff.
00:37:03.519 --> 00:37:07.679
So maybe we are going to see coming back to some base.
00:37:08.159 --> 00:37:16.480
Of course, we can get the extreme version and say, like Elon Musk said at some point, that code will disappear and agents will actually write binary code directly.
00:37:17.440 --> 00:37:28.159
But despite that extreme vision, maybe we are going to start removing some of those uh layers at the top that we had for human interaction.
00:37:28.239 --> 00:37:33.039
And then maybe we can talk to the platform directly using just uh native APIs.
00:37:33.280 --> 00:37:33.519
Yeah.
00:37:33.679 --> 00:37:36.559
James, can you sleep at night now and uh you couldn't read the code?
00:37:36.800 --> 00:37:41.119
I think we're in a kind of a weird sort of uh half state at the moment, which is really hard to reason
about.
00:37:41.280 --> 00:37:45.519
I often think about it in terms of self-driving cars, those different kind of levels that you have.
00:37:45.679 --> 00:37:48.639
I think it's at level four, or level five is self-driving.
00:37:48.800 --> 00:37:51.840
Um, we're like full self-driving, we're kind of at that level four right now.
00:37:52.000 --> 00:37:55.599
You have to keep your hands near the steering wheel, you kind of have to keep your eyes on the road.
00:37:55.679 --> 00:37:58.880
Um and it feels like it's in a similar way, we're kind of there with code, right?
00:37:59.039 --> 00:38:12.079
You're you're still you're still slightly involved with that, you're still kind of doing reviews, um, and you still need to be able to read that code, but ultimately um you kind of do want the agents to take more control and you for you to step back and not kind of care about some of those details so much.
00:38:12.159 --> 00:38:15.039
Um but because we're not in that for kind of full state, you have to.
00:38:15.119 --> 00:38:27.599
And I think we're definitely kind of wasting uh time uh in some cases thinking about um uh these kind of interactions between humans and agents, especially when it comes to code and how it's structured and kind of the quality of it.
00:38:28.559 --> 00:38:34.880
Um Max, you've you've been teaching web and and mobile development for 20 plus years, as you mentioned.
00:38:35.119 --> 00:38:35.519
25 years.
00:38:36.000 --> 00:38:36.400
25 years.
00:38:36.639 --> 00:38:37.599
Yeah, in June 25 years.
00:38:37.920 --> 00:38:38.960
You earn that half century.
00:38:39.440 --> 00:38:40.000
Yeah, yeah, yeah.
00:38:40.880 --> 00:38:54.000
Um what would you say is the kind of like the biggest mind shift uh in recent times now that you're you're telling students that they need to make to be truly not AI native, but now AI agents have entered the picture.
00:38:54.239 --> 00:38:57.039
What's the biggest shift that you've had to make in your life?
00:38:57.280 --> 00:39:06.880
I think that the biggest shift that I see right now is that when you you have new people coming to the web environment, that they ask you, but do I need to learn something?
00:39:07.360 --> 00:39:09.440
Or why do I need to learn something?
00:39:09.679 --> 00:39:14.639
So my take on that is that we need to understand the fundamentals, how how everything works.
00:39:14.719 --> 00:39:16.239
You need to understand HTTP requests.
00:39:16.719 --> 00:39:19.360
So how a fetch is being done, that you have a get post.
00:39:19.840 --> 00:39:24.239
You need to understand those stuff, even if you're not going to write that, but at least you understand what's going on.
00:39:24.400 --> 00:39:24.480
Okay.
00:39:24.719 --> 00:39:25.440
What are you asking?
00:39:25.679 --> 00:39:30.559
If you're asking if it's possible or not, because if it's not possible, the HTML will try to figure that out with that.
00:39:30.639 --> 00:39:37.679
Maybe the solution is so complex that it will not be performance or it will not be really the best possible solution.
00:39:37.760 --> 00:39:42.239
So I think that we still need to keep for now that human criteria.
00:39:42.719 --> 00:39:45.039
And for that you need to understand the basics at least.
00:39:45.440 --> 00:39:48.320
Maybe you're not going to write code anymore from scratch.
00:39:48.960 --> 00:39:54.880
You're you don't need to remember the fetch API arguments and how to add an abort controller or some things like that.
00:39:55.280 --> 00:40:03.760
But you need to understand the basics that when you're making an HTTP request and you are going you're pressing the back button, maybe you should abort that thing.
00:40:04.320 --> 00:40:05.039
Things like that.
00:40:05.119 --> 00:40:10.320
So some basic ideas, I think that they are still important about software development.
00:40:10.639 --> 00:40:13.199
Uh unless until I don't know, AGI appears.
00:40:14.239 --> 00:40:17.039
I don't know if that will happen or not, or when if it happens.
00:40:17.280 --> 00:40:26.000
But at least so far, the the difference is that people actually starting from scratch right now, uh they say, but do I want to learn?
00:40:26.159 --> 00:40:32.719
Do I do I need to understand the difference between uh a div and a section and things like that?
00:40:32.880 --> 00:40:44.960
So um and I say yes, because if you don't know that there isn't a semantic HTML and a non-semantic HTML and what's the difference in terms of accessibility, maybe your Asian is not going to do the right thing by default.
00:40:45.280 --> 00:40:51.199
So we went very quickly from code not needing to be human readable to now needing to understand the details of a div there, though, didn't we?
00:40:51.519 --> 00:40:57.599
Yeah, but but but it's not a div because of the div, it's because of accessibility, accessibility, um, things like that.
00:40:57.840 --> 00:41:04.320
Even even when you are you are making a semantic HTML, it's gonna be better for uh Asians reading that website.
00:41:04.559 --> 00:41:10.159
But uh it depends on the agent that you're using, if you're not using the right skill or things like that.
00:41:10.239 --> 00:41:12.639
It's just taking the average website out there.
00:41:12.800 --> 00:41:22.960
So uh by default, when you when you ask, when you you are doing bycoding and you ask uh for an app, it's probably it's going to make an X.js application by default if you don't know what's going on.
00:41:23.280 --> 00:41:28.719
So it's a little bit like actually Martin Thompson, I think it was, who kind of like mentioned, or maybe it's Martin Fowler, actually, mechanical sympathy.
00:41:29.280 --> 00:41:42.639
The idea of you don't need to like as a as a Formula One race car driver, you don't need to know the inner workings or uh and the and the detail of how to rebuild a gears uh uh uh a gearbox, but you need to know exactly how it works generally so that you can get the most out of it.
00:41:42.719 -->
00:41:44.159
It's that kind of uh that kind of an update.
00:41:44.400 --> 00:41:45.119
It's being an operator.
00:41:45.280 --> 00:41:49.039
I think you I think you have to know enough to be an effective operator.
00:41:49.199 --> 00:41:55.840
And I think that though, you know, there's not gonna be as big of a need to have experts as it gets better.
00:41:55.920 --> 00:41:57.679
And it's gonna happen before AGI.
00:41:57.920 --> 00:42:01.920
We're gonna we're gonna see this before there's generalized artificial intelligence.
00:42:02.079 --> 00:42:09.199
And I think that the the bitter truth is adding in a bunch of glue agents and a bunch of glue code, you're gonna have to unwind.
00:42:09.280 --> 00:42:11.920
Because I do believe your your point that it will get simpler.
00:42:12.079 --> 00:42:23.360
We created a bunch of frameworks and harnesses and layers around code for utilities for humans and human developers to go again connect the workflows.
00:42:23.599 --> 00:42:34.159
But now that the models are getting better because we're doing great with skills, and we keep feeding skills, the context of the skills and the in the intuitive nature of what you're saying is becoming learned.
00:42:34.239 --> 00:42:39.119
And I think people sometimes forget is it's large language model, but it's machine learning.
00:42:39.199 --> 00:42:42.559
And every time we prompt, we are making the models better.
00:42:42.800 --> 00:42:53.440
And as developers utilize the models better and we create more abstractions that train the models, the question is where do you spend your time, effort, and energy getting ready for this?
00:42:53.519 --> 00:42:56.719
And I think we're a little ways away, maybe forever away.
00:42:56.880 --> 00:42:57.280
I don't know.
00:42:57.360 --> 00:42:58.000
I don't believe that.
00:42:58.079 --> 00:43:05.920
I think we're gonna we're gonna get to the point where building web experiences is as easy as sending a text message.
00:43:06.159 --> 00:43:08.239
And that's not too far off in the future.
00:43:08.320 --> 00:43:18.719
And so coming back to the laws of physics and science, understanding HTTP and the way that the internet works, I do believe are valuable, but you're not probably ever gonna touch that layer of the stack ever again.
00:43:19.039 --> 00:43:19.199
Yeah.
00:43:19.599 --> 00:43:20.400
And you shouldn't have to.
00:43:20.639 --> 00:43:21.280
No, no.
00:43:21.440 --> 00:43:25.280
I kind of see it in the same way that things like the kernel, for example, right?
00:43:25.360 --> 00:43:30.800
Like I've been mostly building on the web, I've got some understanding of operating systems, but I don't know how the Linux kernel works, right?
00:43:30.880 --> 00:43:38.400
Uh and I don't need to, uh, because it's so abstracted away from the layer that I'm working at that I, you know, it would be a waste of time to think about that.
00:43:38.480 --> 00:43:44.000
But I do need to know how HTTP requests work and how a fetch works, and that kind of one layer down is important.
00:43:44.079 --> 00:43:46.480
And I kind of see it being a similar thing, right?
00:43:46.559 --> 00:43:55.119
Like you might want to know some of the details and mechanics of how these apps fit together and um and how they're talking to each other, but you don't need to know some of the lower level details.
00:43:55.199 --> 00:43:59.119
And that'll just get more and more abstracted as we're building kind of more and more layers on top.
00:44:00.400 --> 00:44:05.840
Let's talk a little bit about what that means for when we talk about agents being primary users of that.
00:44:06.400 --> 00:44:12.800
When um what does like user testing, what does user testing even look like when agents are agents?
00:44:13.360 --> 00:44:16.480
Well, I mean, obviously the big thing that everyone talks about right now is taste, right?
00:44:16.559 --> 00:44:18.880
In terms, do you want the most generic app?
00:44:18.960 --> 00:44:25.599
Because you've got agents to do all your user testing, and the feedback that you've got there is just the average of the training data that they've been trained on.
00:44:25.760 --> 00:44:26.559
Probably not.
00:44:26.719 --> 00:44:29.679
Um I think that's where taste kind of comes in more than ever.
00:44:30.079 --> 00:44:32.639
Um but uh it's a new skill.
00:44:32.960 --> 00:44:34.320
Isn't that good testing?
00:44:34.719 --> 00:44:35.840
Okay, well, it's solved.
00:44:39.280 --> 00:44:40.239
It's on GitHub.
00:44:40.639 --> 00:44:41.920
Okay, I'll check that one out.
00:44:42.079 --> 00:44:46.400
Um but yeah, it's it's one of those things where maybe again it comes down to abstraction, right?
00:44:46.480 --> 00:44:57.599
Like user testing uh works a different way in terms of you're utilizing agents to maybe user test things around like accessibility or maybe some of the kind of uh large flows that you have within your app.
00:44:57.760 --> 00:45:06.239
Um but for things like specific details around uh you know look and feel, things like that, maybe that's where you're getting a human in the loop to make those decisions.
00:45:06.400 --> 00:45:14.239
Um and they're able to kind of uh sort of supplant what the agent is doing to be able to enable you to do things like that kind of at a much larger scale.
00:45:15.679 --> 00:45:19.599
I'd love to talk about existing uh applications.
00:45:19.760 --> 00:45:30.719
For uh for uh, I guess uh a web developer who's maintaining that, what's the what's the practical, first practical thing they need to do to make that web uh agent ready?
00:45:31.119 --> 00:45:32.719
For make it agent ready.
00:45:32.880 --> 00:45:42.639
So um I think that the the first thing if it's a legacy app, okay, it's uh it's different that the I mean that means an app that I coded like two hours ago, right?
00:45:43.199 --> 00:45:43.760
Okay, yeah, yeah.
00:45:43.920 --> 00:45:44.159
Yeah.
00:45:44.320 --> 00:45:47.039
But maybe it's uh Java-based or things like that.
00:45:47.119 --> 00:45:48.159
So it can be really legacy.
00:45:48.400 --> 00:45:49.760
There are a lot of websites out there.
00:45:49.920 --> 00:45:55.360
So I think that uh I mean for the future, WebMCP is one of the solutions.
00:45:55.679 --> 00:46:03.599
I mean, probably the future is um means that different browsers and uh agents should uh implement that.
00:46:03.920 --> 00:46:06.159
So right now you can start implementing WebMCP.
00:46:06.320 --> 00:46:13.760
WebMCP lets the website expose tools to agents, tools being just uh HTML forms or JavaScript functions.
00:46:14.000 --> 00:46:31.679
Okay um but the first um thing, so one of the so my my my session at the conference was on WebMCP, and one of the questions I I got after the session is that but uh what about people that actually don't want Asians to access your website?
00:46:31.920 --> 00:46:41.440
So the first question is that we will have that situation where a lot of websites or a lot of companies they want want patients that we can use the web notice.
00:46:41.760 --> 00:46:42.480
Yeah, something like that.
00:46:42.559 --> 00:46:53.039
So well first, if you actually want Asians to use I mean, if you don't want I mean you are trying to uh I don't know, try to turn off the sun, you it's not gonna work.
00:46:53.280 --> 00:46:57.760
And Asians will actually find a way to to to get into your website.
00:46:58.000 --> 00:47:17.440
But if you want to make things easier, cheaper, and faster, well, WebMCP lets you do that by exposing functions that the agent can call directly on the front end on the browser session with uh the user's context, local storage, index TV, with uh accessing the camera, the user's camera.
00:47:17.599 --> 00:47:28.320
Think about the I think that one example, the one good example is the uh the meta display glasses that looking into the future if we are going to use those AR glasses.
00:47:28.800 --> 00:47:33.360
Uh the meta display glasses, they have uh web app support.
00:47:33.679 --> 00:47:38.000
So you can run a web app these days, and they have an AI agent inside the glass.
00:47:38.400 --> 00:47:46.559
So then uh sometimes I mean you yeah, you can use your finger to move around, but you probably the Asian will do more most of the stuff.
00:47:46.719 --> 00:47:52.480
So then if you have a web app, an application that is running there, you want to be more Asian ready.
00:47:52.719 --> 00:47:55.840
And for that, I think that WebMCP will be the way to go.
00:47:56.239 --> 00:48:03.519
Yeah, and that's again, it's it's you have to just accept that it's gonna be artificial users across the entire experience.
00:48:03.679 --> 00:48:11.280
From those that are accessing the site, pulling the information, utilizing WebMC to give the experience in another view, or for the developer behind the keyboards.
00:48:11.360 --> 00:48:25.119
And I've been thinking about this a lot, and so at Netlify, we started up an open source project called Access, which is just a scoring system that is completely open to everybody to come contribute to get your systems agent ready.
00:48:25.199 --> 00:48:30.559
It goes through a series of checks and balances in some libraries that can come and see, like, are your endpoints ready?
00:48:30.719 --> 00:48:32.239
Have you made this change to your site?
00:48:32.400 --> 00:48:36.719
And really, you know, we're just trying to get the community because nobody has the answers right now.
00:48:36.880 --> 00:48:40.719
I think the interesting part is like, how do you make a website the web agent ready?
00:48:40.800 --> 00:48:42.239
We're all doing that right now.
00:48:42.400 --> 00:48:47.119
Everybody, I mean, somebody upstairs is probably getting the web agent ready, if I had to guess.
00:48:47.280 --> 00:49:00.639
Um, the reality is there's not one technologist or company to rule them all, that this is a time and space where the development community globally are coming together and defining these workflows, defining how to become agent ready.
00:49:00.800 --> 00:49:04.880
And I think that's an actually really exciting part of our job that didn't exist.
00:49:05.039 --> 00:49:06.800
I mean, I've been writing code forever.
00:49:06.960 --> 00:49:13.119
Finally, we're getting to do something new and fresh, even though I learned that you know it's the birthday of AI and it's 70 years old.
00:49:13.199 --> 00:49:15.280
But it feels new and fresh to me, I'll tell you.
00:49:16.960 --> 00:49:19.679
Do you think there'll be a time when the browser will be legacy?
00:49:20.480 --> 00:49:23.840
Oh man, I don't know.
00:49:24.000 --> 00:49:25.599
I'm I I I love the browser.
00:49:25.679 --> 00:49:26.480
I love the internet.
00:49:26.719 --> 00:49:28.400
Like my partners at Netscape.
00:49:28.639 --> 00:49:29.760
You know, I work it at Microsoft.
00:49:30.239 --> 00:49:33.119
Working at Microsoft was not great in my family.
00:49:33.280 --> 00:49:34.559
Uh they're like, what?
00:49:34.880 --> 00:49:36.320
I I don't think it's going anywhere.
00:49:36.400 --> 00:49:38.320
I think the browser will just be in a new form.
00:49:38.400 --> 00:49:40.000
You know, you're gonna have more heads-up displays.
00:49:40.079 --> 00:49:42.719
It's still gonna be a browser behind the scenes, it's still gonna be the web.
00:49:42.800 --> 00:49:49.440
Um but I do feel like HUD's gonna come in and not Dorky Oculus VR, but but true heads-up displays.
00:49:49.519 --> 00:49:51.599
We're seeing that with Meta as an example.
00:49:51.840 --> 00:49:54.719
But again, that's just another, that's another version of a browser.
00:49:54.960 --> 00:49:56.800
It's just it's it's just HTTP.
00:49:57.039 --> 00:49:58.480
TCP IP, man.
00:49:59.280 --> 00:50:00.079
All the way down.
00:50:00.320 --> 00:50:01.840
Um That's right.
00:50:02.960 --> 00:50:07.199
Yeah, I think the thing for me right now is accessibility.
00:50:07.360 --> 00:50:13.280
Actually, it feels like a really um kind of like a powerful skill that we should
kind of lean into when it comes to agents accessing the web.
00:50:13.519 --> 00:50:18.079
We spent years and years and years making the web semantic, like it's inherently semantic.
00:50:18.239 --> 00:50:25.599
Um, and um agents have allowed us to kind of like make our sites more accessible almost for free in some ways, right?
00:50:25.760 --> 00:50:31.519
They're the their um the training data covers a lot of kind of accessibility techniques and best practices.
00:50:31.679 --> 00:50:37.199
And so um, if you ask an agent, say, make this website accessible, it will do a pretty good job right now.
00:50:37.280 --> 00:50:39.039
It won't be amazing, but it'll do a pretty good job.
00:50:39.199 --> 00:50:44.400
And to me, that kind of accessibility layer is something that uh works great for agents, right?
00:50:44.559 --> 00:50:46.400
Um agents don't want to read the whole DOM.
00:50:46.639 --> 00:50:51.199
Like they don't care about a lot of the parts, but what they do care about is um the kind of accessibility tree.
00:50:51.280 --> 00:50:54.239
And they can use that to look at elements on the page, interact with it.
00:50:54.400 --> 00:50:58.079
Um uh and so they're able to use it kind of as a human would do.
00:50:58.320 --> 00:51:02.800
Um, but uh I think, yeah, more and more of that will get abstracted away.
00:51:02.880 --> 00:51:06.320
And you again, just like code, um some of these things just kind of won't exist.
00:51:06.639 --> 00:51:13.840
And let's talk about the value of skills here, because I think that's that's all of a sudden becomes very, very important with things like accessibility and those types of things.
00:51:14.000 --> 00:51:25.119
Um given that skills are kind of quite language agnostic, um, do we do we feel like there's gonna be kind of like front-end specific skills to help with things like that, like accessibility and things like that?
00:51:25.199 --> 00:51:25.280
Yeah.
00:51:26.639 --> 00:51:30.719
What even is front-end in a world where uh MCP, right?
00:51:32.400 --> 00:51:34.719
Where where a lot of this stuff disappears.
00:51:34.800 --> 00:51:39.280
Um, you know, as someone who is less technical, do I need to know exactly what front end is?
00:51:39.360 --> 00:51:44.239
You know, I just want to build a cool app that's gonna help me to do uh specific task or make my life easier.
00:51:44.400 --> 00:52:00.639
And so maybe what we're calling front end just becomes more of a kind of generic design uh element, whereas you don't need this kind of specialized skill of uh uh you know understanding HTML and being able to take a set of design or a set of kind of um requirements and turn that into something that can be rendered in a browser.
00:52:00.800 --> 00:52:01.280
So yeah.
00:52:01.679 --> 00:52:02.159
So it's interesting.
00:52:02.239 --> 00:52:08.239
Actually, it's the the front end and web, the front end and back end just completely there, there's no there's no we're all full stack.
00:52:08.559 --> 00:52:09.039
There's no back.
00:52:09.519 --> 00:52:12.559
There's no content management, there's no git, there's just code.
00:52:12.800 --> 00:52:12.960
Yeah.
00:52:13.360 --> 00:52:26.960
And it and to people that don't know, I mean, it's again, they they do not understand that's under the hood, and you have to start asking yourself, were these things that we built now needed, or can there be simplification across the board to build and generate those experiences?
00:52:27.119 --> 00:52:33.360
Because as agents become more prevalent, you're gonna have hyper-personalized experiences that are just for you.
00:52:33.519 --> 00:52:37.840
And those experiences and those websites will be created on the fly geared towards you.
00:52:37.920 --> 00:52:41.199
And I think it's gonna be really interesting on how we do approach that.
00:52:41.360 --> 00:52:52.400
And I don't know if it's gonna be front-end or back end, but regarding skills, I think for some stuff you don't need it, but that's still very much needed for guardrails and governance outside of the web.
00:52:52.559 --> 00:52:57.280
Like, I would not I would not YOLO without a skill for a database.
00:52:57.440 --> 00:52:58.480
Like, trust me on this.
00:52:58.559 --> 00:53:01.920
Like, you would have somebody that doesn't know anybody bork that database.
00:53:02.000 --> 00:53:06.559
I've seen it happen because I was working on a skill for a database and uh broke it.
00:53:06.719 --> 00:53:15.840
Um, I think that there's still a need in especially larger uh environments where it really is a team sport I'm building to have a good skills library.
00:53:15.920 --> 00:53:25.760
Because if you want to make your company participate and really think about the experiences that you are bringing, I think skills are required because you're gonna unleash them.
00:53:25.920 --> 00:53:36.000
And so that's becoming more of the developer and the architect's you know reign of control is ensuring that those guardrails are put in place so that people can safely do the things they need to do.
00:53:36.159 --> 00:53:40.800
And and in fact, people already assume that there's trust by default in these platforms.
00:53:40.960 --> 00:53:44.639
And so, you know, do one wrong thing, you're gonna be in a news rail.
00:53:44.719 --> 00:53:59.679
So yeah, I mean the blinds are blurring of of what we say are front end and back end because now the experiences are ephemeral and and they can just happen magically, it feels magically, but there's a lot of guardrails to make that magic happen in the background, and that's where we come in and put skills and specs.
00:54:00.079 --> 00:54:02.079
I I think the content of the skill will evolve, right?
00:54:02.159 --> 00:54:10.079
Right now it's very direct, and it's like, you know, use these HML elements, use this design language, this design library, these components, this front-end stack.
00:54:10.159 --> 00:54:14.079
Um I think what we'll see is a move towards the skills being more about intent, right?
00:54:14.559 --> 00:54:20.079
Um, if we think about things like the SOL document for Claude, for example, it doesn't have a lot of do this, don't do that.
00:54:20.159 --> 00:54:24.960
Um it's more around um what the intent is and kind of the kind of end goal and what you care about.
00:54:25.039 --> 00:54:34.079
And I think we'll start to see a translation of um skills uh into those kinds of documents rather than just being a specific list of do's and don'ts and a kind of checklist of how you do it.
00:54:34.320 --> 00:54:37.199
Everything is evolving really quickly these days, even the web.
00:54:37.679 --> 00:54:41.280
And the problem is that right now we don't have models that are being trained every day.
00:54:41.440 --> 00:54:51.360
So we are working with models that were trained a year ago or something like that, which means they don't have the last year of uh web evolution.
00:54:51.519 --> 00:54:53.440
So that's when skills appear.
00:54:53.599 --> 00:55:02.320
So they can teach the model, hey, we need you have this API now and that API, so you don't need to fall back to a library or things like that.
00:55:02.480 --> 00:55:04.000
You should use that library.
00:55:04.159 --> 00:55:10.880
That base baseline library that we have right now, baseline is how compatibility works across different browsers this year.
00:55:11.119 --> 00:55:22.880
So I think skills are are are going to be useful for for web design and web development on CSS or JavaScript APIs uh to modernize uh what the model already knows.
00:55:23.599 --> 00:55:24.559
I won't argue about that.
00:55:24.719 --> 00:55:25.840
I agree with skills.
00:55:26.000 --> 00:55:32.239
Um Dana, I think it was about five minutes ago you said uh there is no GitHub, only code.
00:55:32.639 --> 00:55:33.280
There's only code.
00:55:34.159 --> 00:55:40.480
I just want to I just want to all I wanted to say was I think it was I think it was Dana from Ghostbusters who I know.
00:55:40.719 --> 00:55:42.559
So there there is no Dana only Ghostbusters.
00:55:42.719 --> 00:55:43.280
There is only Zool.
00:55:44.159 --> 00:55:45.280
And it was it fits so perfectly.
00:55:45.599 --> 00:55:48.000
Well, that is my GitHub handle, Dana is Zool.
00:55:48.159 --> 00:55:50.639
So check out my there is amazing.
00:55:50.880 --> 00:56:03.199
So No, I mean I I'm an ex-hubber and it it and it pains me to say that, but like the dependency on the GitHub workflow is going away because your code is just not that precious.
00:56:03.519 --> 00:56:12.880
Um it is and it isn't because now with it with truly atomic rollbacks and the ability to fix fast and create software factories underneath, there is still a need for revisioning.
00:56:12.960 --> 00:56:13.920
Do not get me wrong.
00:56:14.079 --> 00:56:16.800
But the whole Git workflow, like that's for developers.
00:56:17.360 --> 00:56:18.000
That's for developers.
00:56:18.960 --> 00:56:28.400
Well, what what's the key thing an agent requires in some
kind of uh, you know, whether it's consistency, whether it's persistence in something?
00:56:28.639 --> 00:56:31.199
What is that, first of all, what is that thing they want to persist?
00:56:31.280 --> 00:56:35.840
Is it like uh specs, skills, context, uh instructions?
00:56:36.000 --> 00:56:36.800
What is the thing?
00:56:37.199 --> 00:56:39.199
I I still believe in source control.
00:56:39.360 --> 00:56:39.519
Yeah.
00:56:39.760 --> 00:56:45.360
I'm just saying like I believe that source control as a company won't need to be a thing.
00:56:45.519 --> 00:56:51.199
It's gonna be more localized to the experience, or it's gonna be more localized closer to the builder.
00:56:51.360 --> 00:56:58.239
And it's gonna be in the systems that they're utilizing to make the changes versus calling back and creating developer workflows that are traditional.
00:56:58.559 --> 00:57:12.559
Now, a lot of us have dependencies on GitHub actions and all of the things that we utilize for our CI CD platform or CI CD pipelines to ship code, but as CI CD becomes easier, and really that's all that matters.
00:57:12.639 --> 00:57:14.159
So, what does an agent need to know?
00:57:14.320 --> 00:57:16.480
How to roll back and how to push forward.
00:57:16.800 --> 00:57:19.840
I don't care how it happens, it just that's all it needs to know.
00:57:19.920 --> 00:57:29.119
And you can use you know git as your um version control, or you can use anything else because most users or most new builders are not gonna know.
00:57:29.280 --> 00:57:31.679
They just are gonna say, oh, that change doesn't look good.
00:57:31.840 --> 00:57:32.320
Roll it back.
00:57:32.719 --> 00:57:33.760
They're going back to FTP.
00:57:34.159 --> 00:57:35.679
Oh, we are yeah, you know what?
00:57:35.840 --> 00:57:37.199
What's old is new again.
00:57:37.360 --> 00:57:38.159
I'm here for it.
00:57:38.480 --> 00:57:40.079
Just SCP direct into production.
00:57:40.480 --> 00:57:40.880
You know what?
00:57:41.039 --> 00:57:41.599
Why not?
00:57:41.840 --> 00:57:42.480
Why not?
00:57:42.880 --> 00:57:45.119
Because you say you say this is what agents need to know.
00:57:45.280 --> 00:57:47.840
Is this gonna be completely abstracted from developers?
00:57:48.079 --> 00:57:50.239
Do developers even need to know about this now?
00:57:50.800 --> 00:57:51.920
All happening through the agent?
00:57:52.239 --> 00:57:54.239
I think it's kind of a different set of requirements, right?
00:57:54.320 --> 00:57:57.119
We keep talking about abstraction, and I think Git is one of those things.
00:57:57.280 --> 00:57:58.159
I wanted to be abstracted.
00:57:58.400 --> 00:57:59.440
Git frustrates me all the time.
00:58:00.400 --> 00:58:04.559
It's an incredible piece of software, but it's also very, very annoying at times.
00:58:04.719 --> 00:58:10.079
And when I'm teaching folks how to do engineering, it's often the thing that is always tripping them up, right?
00:58:10.239 --> 00:58:11.599
But it's great for agents, right?
00:58:11.760 --> 00:58:15.840
Um and so I think building layers over the top of Git that that do the abstraction.
00:58:16.000 --> 00:58:20.800
Um, I also think we're kind of at a point where we can start rethinking how these tools work, right?
00:58:20.880 --> 00:58:27.679
We don't need to apply the last 25 years of software engineering um to the tools of the next 25 years.
00:58:28.000 --> 00:58:36.880
We should kind of use this as a bit of an opportunity to rethink things completely and not follow those same patterns, but kind of just tweak them a bit and get them to work for agents.
00:58:37.280 --> 00:58:41.599
But are we going to rethink that, or the agent will do that?
00:58:41.679 --> 00:58:42.079
We don't know.
00:58:42.400 --> 00:58:43.039
Oh, we get to.
00:58:43.199 --> 00:58:43.760
That's the fun bit.
00:58:44.719 --> 00:58:45.599
I think it's gonna be us.
00:58:45.679 --> 00:58:50.480
I think that's again where we we put the humans back in because I know y'all have all probably been there too.
00:58:50.639 --> 00:58:57.360
You're sitting there vibe coding and you're up to like 500, 600 lines, and then you look and you're like, what is going on here?
00:58:57.599 --> 00:59:00.320
Like 400 of this can be, it's just garbage.
00:59:00.480 --> 00:59:04.639
And you tell the agent, hey, 400 of the 400 lines of code, this is garbage.
00:59:04.719 --> 00:59:07.039
You need to go in there, clean this up because I don't know what you're doing.
00:59:07.119 --> 00:59:07.599
Start over.
00:59:07.760 --> 00:59:08.079
Guess what?
00:59:08.159 --> 00:59:08.880
It can do it.
00:59:09.119 --> 00:59:12.880
Um, but it also can do it bad, bad, and bad, and bad, and then good.
00:59:13.199 --> 00:59:14.000
You're absolutely right.
00:59:14.400 --> 00:59:16.320
We're not quite there, but we're gonna get there.
00:59:16.400 --> 00:59:21.360
Because every time me, you, and others go through that workflow, we're making it better.
00:59:21.519 --> 00:59:24.159
We're we're we're training it, we're training it to be better.
00:59:24.239 --> 00:59:27.119
And so where where's the line gonna blur?
00:59:27.199 --> 00:59:29.119
But again, we're pretty far off from there.
00:59:29.360 --> 00:59:29.599
Yeah.
00:59:30.159 --> 00:59:36.320
Let's ask one quick fire question uh where I'd love an answer and then a brief why.
00:59:36.559 --> 00:59:46.719
Uh the question is: what would you say is the most important thing that uh a web developer should either learn uh or change over the next 12 months or so?
00:59:46.960 --> 00:59:48.719
So why don't we start with yourself, Dana?
00:59:49.119 --> 00:59:52.079
In the next 12 months, a web developer should embrace AI.
00:59:52.239 --> 01:00:03.440
I see a lot of resistance, a lot of people saying, I can do better, a lot of fear, and the best thing they can do is embrace it, be curious, and start using it and becoming a force in nature because you will be left out.
01:00:03.679 --> 01:00:04.880
You will be replaced.
01:00:05.119 --> 01:00:08.079
The world is now AI native and our space is as well.
01:00:08.239 --> 01:00:19.679
And so the best thing I can do is like stop obsessing about your frameworks and start obsessing about your workflows and start becoming AI native because the expectations for you to produce just got amplified.
01:00:20.480 --> 01:00:24.559
Yeah, I would say that maybe being a better AI user.
01:00:24.719 --> 01:00:31.199
And that means uh as a technical person, understand a little bit about AI, so how the model works.
01:00:31.519 --> 01:00:32.239
What is the model?
01:00:32.320 --> 01:00:47.199
Maybe trying with a local model, even if you're not going to use that one, but understanding how it works, why context matters, so understanding the context window, all the stuff that maybe if you're a web developer, you have never done machine learning or things like that, you don't really know.
01:00:47.679 --> 01:00:53.760
And if you if you're trying to use it as a magic solution, uh you will probably get worse results.
01:00:53.920 --> 01:01:08.719
So I think that uh learning a little bit about AI, uh a little bit about prompting, understanding why prompting is important, the context, how the context is being organized and managed and compacted and all that stuff will make you a better developer.
01:01:09.840 --> 01:01:13.920
Um so I would say probably using uh AI to enhance your learning.
01:01:14.079 --> 01:01:20.880
AI doesn't just have to be about generating code for you, but what about using it as a way to kind of interview about topics you might be interested in learning?
01:01:21.039 --> 01:01:25.119
We've kind of talked about using fundamentals, you still kind of need to know a little bit of those right now.
01:01:25.199 --> 01:01:30.719
Um thinking about using AI in that sense, I think, is probably super, super powerful and not be scared of it.
01:01:31.039 --> 01:01:37.840
Really interesting, actually, because actually one of the one of the folks on our team, uh Alan Pope, uh, he's building out some of the some of the TESL learning.
01:01:37.920 --> 01:01:40.000
So we're trying to build a learning platform.
01:01:40.480 --> 01:01:43.760
Kind of like one of the things I mentioned to him is we need to be skills first.
01:01:43.920 --> 01:01:53.519
Like create skills that are module that are education modules that you can kind of like plug into your agent and then do your learning through the agent versus a traditional learning platform.
01:01:53.599 --> 01:01:55.760
So super, super curious about that.
01:01:56.559 --> 01:02:09.519
Dana, Max, James, thank you so so much uh for not just presenting amazing sessions and about to present amazing sessions, but uh but also for joining me today and sharing all your wisdom and knowledge on the on the podcast.
01:02:09.599 --> 01:02:10.159
I really appreciate that.
01:02:10.400 --> 01:02:10.559
Thanks.
01:02:12.400 --> 01:02:15.039
And uh and thanks for listening and tune into the next session.
01:02:15.280 --> 01:02:16.159
Bye for now.
01:02:16.880 --> 01:02:21.440
Really fascinating conversations uh throughout in both uh of the discussions here.
01:02:21.679 --> 01:02:22.960
Hope you enjoyed the episode.
01:02:23.119 --> 01:02:30.320
If you want to hear more from the guests featured in this episode, their full DevCon talks have been linked in the descriptions below.
01:02:30.559 --> 01:02:31.760
So that's it for this episode.
01:02:31.920 --> 01:02:32.639
See you next time.
01:02:32.880 --> 01:02:37.519
The AI Native Dev is brought to you by Tesla, the package manager for skills and context.
01:02:37.760 --> 01:02:40.639
Your hosts are Guy Pajani and me, Simon Maple.
01:02:40.880 --> 01:02:42.400
Our producer is Tom Dowler.
01:02:42.719 --> 01:02:45.840
The AI Native Dev is not just a podcast, it's a community.
01:02:46.000 --> 01:02:49.599
And we host a monthly meetup of Tescofficts at Temple Legends.
01:02:49.760 --> 01:02:53.920
It's etc.io forward community to learn more about this.