WEBVTT
00:00:03.680 --> 00:00:04.639
Be honest.
00:00:04.879 --> 00:00:08.880
How has your working relationship with AI changed in the past few years?
00:00:09.119 --> 00:00:13.119
Are you still copying, pasting into ChatGPT and hoping for the best?
00:00:13.359 --> 00:00:19.120
Are you drowning in AI-generated drafts that need more editing than if you've just done it yourself?
00:00:19.359 --> 00:00:24.000
Or maybe you figured out a few tricks, but something still feels off.
00:00:24.640 --> 00:00:31.280
Here's the thing: most people are still treating AI like a junior version of themselves with better grammar.
00:00:31.519 --> 00:00:32.960
But the game has changed.
00:00:33.200 --> 00:00:35.759
In 2023, you learned to prompt.
00:00:36.000 --> 00:00:39.119
In 2024, hopefully you learn to iterate.
00:00:39.280 --> 00:00:44.240
And in 2025, you started wondering why this still feels so manual.
00:00:44.560 --> 00:00:48.719
And now in 2026, the skill gap isn't technical anymore.
00:00:48.880 --> 00:00:50.159
It's managerial.
00:00:50.320 --> 00:00:51.280
It's human.
00:00:51.439 --> 00:00:53.200
It's self-reflection.
00:00:53.359 --> 00:00:59.520
Because the teams winning with AI aren't better prompters or they don't have access to better models than you.
00:00:59.679 --> 00:01:07.680
They're just better orchestrators of AI, better at throwing away the old way of work and starting fresh with AI front and center.
00:01:07.840 --> 00:01:13.280
And that human AI collaboration shift changes everything about how you work.
00:01:15.760 --> 00:01:16.400
All right.
00:01:16.799 --> 00:01:23.040
Does that get you a little fired up to reexamine your working relationship with AI?
00:01:23.280 --> 00:01:24.079
I hope it is.
00:01:24.159 --> 00:01:29.120
And welcome to volume four of the Start Here series with Everyday AI.
00:01:29.200 --> 00:01:34.799
We're going to be looking at the human AI collaboration and best practices for working alongside AI.
00:01:35.040 --> 00:01:41.599
So if you're new here, after almost 700 episodes, we have kicked off our start here series.
00:01:41.680 --> 00:01:43.200
Like I said, this is volume four.
00:01:43.359 --> 00:01:48.640
This is essential beginner and advanced AI overviews to launch your year strong.
00:01:48.799 --> 00:01:56.480
So if one of your big goals in 2026 was to learn AI or to double down, maybe you've been there since day one.
00:01:56.640 --> 00:01:59.680
This Start Here series is for you.
00:02:00.000 --> 00:02:03.120
Uh and we have a special resource.
00:02:03.280 --> 00:02:06.719
So make sure you go to starthearseries.com.
00:02:06.879 --> 00:02:11.919
That is going to get you access for free to our inner circle AI community.
00:02:12.000 --> 00:02:15.120
And it's going to take you straight to our Start Here series page.
00:02:15.280 --> 00:02:18.960
So you can very easily catch up with all of the Start Here series.
00:02:19.039 --> 00:02:27.199
Uh, listen, additional resources, related episodes, anything you need to get either started or centered on your AI journey.
00:02:27.360 --> 00:02:31.039
And if you are very new here, this is part of Everyday AI.
00:02:31.120 --> 00:02:39.360
It's a daily live stream podcast and free daily newsletter, helping everyday business leaders like you and me keep up with AI and get ahead and use it to grow our companies and our careers.
00:02:39.520 --> 00:02:49.680
So the start here series, we're going to do about a dozen or so of these episodes in the first part of 2026, uh, that I hope is going to be a refreshing look at AI for all of us.
00:02:49.759 --> 00:02:56.639
Because, yeah, as someone that's done this now almost 700 times, I understand how it can be so hard to just start somewhere.
00:02:56.800 --> 00:02:58.400
So this is for you.
00:02:58.560 --> 00:03:04.719
And if you missed our last uh episode, it was volume three of the Start Here series, AI as an operating system.
00:03:05.039 --> 00:03:08.719
So if you are listening uh on the podcast, make sure to check your show notes.
00:03:08.800 --> 00:03:11.759
We're going to be updating the show notes of all the start here series.
00:03:11.840 --> 00:03:16.240
So you can very easily, if you're listening to the podcast, just flip around that way as well.
00:03:16.560 --> 00:03:21.439
All right, let's get straight into it and talk about well, why?
00:03:21.680 --> 00:03:25.840
Why is there this big skill gap in 2026?
00:03:26.000 --> 00:03:36.639
There's those people that are running away with AI, and then there's people that have been using it pretty much every day or every week since the Chat GPT moment of November 2022.
00:03:36.879 --> 00:03:39.680
And they're still just barely keeping up.
00:03:40.000 --> 00:03:42.560
So it's a shift.
00:03:42.800 --> 00:03:46.800
It's a shift from operator to orchestrator.
00:03:46.960 --> 00:03:51.360
And that's what I think that you have to start thinking about, uh, right?
00:03:51.599 --> 00:04:06.560
Your relationship with AI, and if you really want to stay ahead of the best practices, because your new job isn't doing the work, it's really defining on what the work looks like, how it gets done, what information the work needs.
00:04:06.719 --> 00:04:10.159
You set the parameters, the constraint, and the success criteria.
00:04:10.479 --> 00:04:15.439
And the AI is probably going to be the one doing the actual work, right?
00:04:15.599 --> 00:04:21.040
I talked about this literally in 2024, this concept of you know agent orchestration.
00:04:21.120 --> 00:04:23.680
And that's how I believe that that's where we're headed.
00:04:23.759 --> 00:04:29.199
And I think we've really seen that a lot in the last uh, gosh, only four to five weeks.
00:04:29.360 --> 00:04:34.000
That's really coming to kind of mainstream fruition uh on this big shift.
00:04:34.160 --> 00:04:41.759
And I think the winners of this era aren't those that who can effectively, you know, onboard, supervise, and audit digital agents.
00:04:41.920 --> 00:04:47.600
It is really thinking it like an entire team that you are now orchestrating, right?
00:04:47.759 --> 00:04:51.120
Not an entire team that you are a player coach.
00:04:51.279 --> 00:04:52.720
You are the manager.
00:04:52.879 --> 00:04:55.600
You're not out on the field doing all of the work.
00:04:56.560 --> 00:05:03.040
So here's what we're gonna cover in today's uh iteration of the Start Here series.
00:05:03.199 --> 00:05:12.000
We're gonna be talking about the comfortable, uncomfortable truths of AI in 2026 and our relationship as humans with AI.
00:05:12.240 --> 00:05:20.720
And I think what so many people do is they take their current antiquated processes that a lot of times are broken, and they just find ways to put AI in it, right?
00:05:20.800 --> 00:05:21.759
In the front and the middle.
00:05:21.920 --> 00:05:24.480
Oh, they find a broken joint, a leaky funnel.
00:05:24.639 --> 00:05:26.319
Let's slap some AI on it, right?
00:05:26.480 --> 00:05:29.920
Do you guys remember the uh uh the infomercial, right?
00:05:30.000 --> 00:05:31.839
You slap some flex seal on it.
00:05:31.920 --> 00:05:33.120
That's what people look at AI.
00:05:33.199 --> 00:05:34.720
They find something that's broken or something.
00:05:34.879 --> 00:05:36.399
Ah, this can be improved a little bit.
00:05:36.480 --> 00:05:38.000
Let's slap some AI on it.
00:05:38.160 --> 00:05:40.000
That's the wrong way to do it.
00:05:40.160 --> 00:05:45.040
Um, I think that it's about unlearning, right?
00:05:45.519 --> 00:05:49.120
I'm gonna be griping on a couple of these terms that I absolutely hate.
00:05:49.279 --> 00:05:50.480
One is human in the loop.
00:05:50.639 --> 00:05:51.759
I'll get to that later.
00:05:51.920 --> 00:05:54.240
One is upskilling or reskilling.
00:05:54.480 --> 00:05:58.879
That alone, I think, has set so many companies back years.
00:05:59.600 --> 00:06:08.079
Because when you, you know, the C-suite and the boardroom, uh, you know, go around HR and they're looking at, you know, AI investment.
00:06:08.160 --> 00:06:09.920
They're like, who needs upskilled?
00:06:10.079 --> 00:06:11.519
Who needs reskilled?
00:06:11.680 --> 00:06:19.519
They're looking at it almost as a um proactive or uh or sorry, a reactive measure, right?
00:06:19.600 --> 00:06:20.959
It's almost like a course correction.
00:06:21.040 --> 00:06:23.120
Oh, who needs who needs upskilled this year?
00:06:23.279 --> 00:06:24.879
Who needs reskilled in AI?
00:06:25.040 --> 00:06:27.199
No, you need to unlearn, right?
00:06:27.279 --> 00:06:43.519
I've been uh chirping out uh this unlearned word for a very long time because it's how I really started getting the most out of AI myself when I started unlearning good habits that had traditionally led to success.
00:06:43.920 --> 00:06:51.519
And I think that's when you look at your relationship with AI, your working relationship with AI, that's what you have to start to do.
00:06:51.680 --> 00:06:56.959
You have to start to say, no, why would I just upskill with AI?
00:06:57.120 --> 00:07:01.199
My skills are, again, they're not worthless, they're worthless.
00:07:01.360 --> 00:07:14.879
So if you're still holding on to those skill sets, right, whether you've been using them for two years or 30, it's probably time to let them go and to unlearn and then relearn and rebuild from scratch.
00:07:16.560 --> 00:07:17.600
Well, why?
00:07:17.759 --> 00:07:30.399
Well, because right now we have, I'm not gonna say superhuman AI, but I will say above human, above human expert AI that can make millions of decisions per second worldwide.
00:07:30.560 --> 00:07:32.399
And humans can't keep up, right?
00:07:32.480 --> 00:07:36.319
So we think of this concept of human in the loop, and that's our job, right?
00:07:36.480 --> 00:07:42.959
And that's a job that we thought was gonna last into the, you know, into the late 2020s.
00:07:43.040 --> 00:07:43.519
It's not.
00:07:43.759 --> 00:07:47.839
Human in the loop, in my opinion, is dead on arrival.
00:07:47.920 --> 00:07:56.160
It was dead on arrival, uh, right, because we thought that you could stick any human in the loop and okay, let's have this human go look what the AI is doing.
00:07:56.399 --> 00:07:56.720
No.
00:07:57.439 --> 00:07:58.720
It is dead, right?
00:07:58.800 --> 00:08:06.079
Because agentic workflows really create miles-long action traces that humans cannot realistically interpret.
00:08:06.240 --> 00:08:07.360
We can't keep up.
00:08:07.600 --> 00:08:12.480
The complexity has out far outpaced what a human can review, right?
00:08:12.639 --> 00:08:24.160
When you see a large language model, or if you're working uh with agents that have sub-agents and they can work around the clock, what good is a human gonna do at that point?
00:08:24.959 --> 00:08:25.360
Right?
00:08:25.519 --> 00:08:27.120
A human in the loop, right?
00:08:27.360 --> 00:08:29.120
Like, I mean, let's say what it is.
00:08:29.279 --> 00:08:35.919
It's it's it's your uh last, you know, your last hope that you didn't get something wrong.
00:08:36.000 --> 00:08:40.240
It's your last hope that a hallucination doesn't slip its way through to production, right?
00:08:40.320 --> 00:08:47.039
It's your last hope that uh you don't make a tragic mistake from uh what a bunch of agents did.
00:08:47.759 --> 00:08:53.759
Because if your human in the loop can't push back, pause, or ask hard questions, that's not even oversight.
00:08:53.919 --> 00:08:57.440
That's a fail safe, probably a name only, right?
00:08:58.480 --> 00:09:07.440
No human in the loop can keep up with what agentic AI can do today, let alone next week and next month.
00:09:07.600 --> 00:09:10.399
And yes, it does change that quickly.
00:09:10.559 --> 00:09:12.000
Uh, my gosh.
00:09:12.960 --> 00:09:15.919
So this isn't hype, this is measurable.
00:09:16.080 --> 00:09:19.279
And the problem is, well, models are smarter than us.
00:09:19.440 --> 00:09:25.279
So why would we want to keep a general or generic human in the loop?
00:09:25.519 --> 00:09:32.240
You need to find what your expertise is or you know, your team's expertise.
00:09:32.399 --> 00:09:38.159
What is that one thing that you are always better than the best large language models at?
00:09:38.320 --> 00:09:50.960
And that's what you should be focusing on and double down, uh doubling down on and seeing where the overlap is between that one thing and what ultimately makes your company or your team more revenue, uh, right?
00:09:51.120 --> 00:09:57.440
Because you have to stop comparing AI to perfection, compare it to the human speed and accuracy for the same tasks.
00:09:57.759 --> 00:10:01.120
And at that point, humans can't keep up.
00:10:01.360 --> 00:10:04.559
And I'm not gonna say every single use case, right?
00:10:04.639 --> 00:10:08.159
But I'll say the overwhelming majority of use cases.
00:10:08.799 --> 00:10:17.519
If you have an average skilled human, right, pick pick your pick your niche, pick your vertical, right?
00:10:17.759 --> 00:10:24.559
Financial analysis, uh, marketing, uh PR, whatever.
00:10:25.200 --> 00:10:26.159
Pick your job.
00:10:26.480 --> 00:10:27.919
Find an average person.
00:10:28.480 --> 00:10:38.399
Someone that's you know worked there for one of your colleagues, but that colleague that's amazing at AI and they know everything, and then pick the smartest person in that same vertical.
00:10:39.600 --> 00:10:44.879
The smartest person in that vertical cannot compete with the person who knows enough, right?
00:10:45.519 --> 00:10:48.879
They speak the language, but is a whiz at AI.
00:10:49.120 --> 00:10:52.080
It's not, it's it's not close.
00:10:52.240 --> 00:10:54.720
It's not even a competition, right?
00:10:55.120 --> 00:11:00.559
That's like having 20 Michael Jordans on a team versus you know, a Pee-Wee team.
00:11:00.639 --> 00:11:01.840
It's not fair.
00:11:02.240 --> 00:11:33.200
So I think that's what we have to get to is we have to understand that these models now that are agentic by default, that can reason, that can think, they can plan ahead, uh, they can they can iterate, loop back, take different paths, use tools all faster than humans and all spin off agents that do the same thing and you know can do it hundreds at a time, you know, spin off hundreds of sub-agents and then come back and you know, have multiple, you know, mixture of experts or you know, what I like to call mixture of models uh that that judge all of those inputs, right?
00:11:33.279 --> 00:11:35.840
You humans can't compete anymore.
00:11:40.159 --> 00:11:42.480
So why?
00:11:43.600 --> 00:11:49.360
Why are we still thinking that the human AI collaboration in 2026, when it comes to AI, is human in the loop?
00:11:49.440 --> 00:11:50.240
It's not.
00:11:50.559 --> 00:11:51.120
Uh, right?
00:11:51.279 --> 00:12:00.320
There was a uh a recent study that looked at 700 plus court cases worldwide, now involve AI hallucinations and fabricated citations.
00:12:00.399 --> 00:12:07.120
And that rate is accelerating to multi to a handful of new case uh new cases daily.
00:12:07.759 --> 00:12:16.480
And that's because the AI is too fast, humans are too lazy, and for the most part, organizations aren't training their people.
00:12:16.639 --> 00:12:23.919
A lot of people assume, oh, if my company pays for a good AI, okay, well, I'm just gonna have it do most of my work.
00:12:24.000 --> 00:12:27.840
It's gonna be right because my company's paying for it and they're not training me on it.
00:12:28.000 --> 00:12:30.480
And this is a huge risk.
00:12:30.639 --> 00:12:34.480
You know, JP Morgan, uh, they acknowledge this risk openly.
00:12:34.559 --> 00:12:41.039
They recently talked about that when systems perform correctly, most of the time human attention drifts, right?
00:12:41.200 --> 00:12:50.480
But if AI is right 85 to 95% of the time, well, your human in the loop falls asleep, right?
00:12:50.559 --> 00:12:56.480
Or your human in the loop, maybe if they're a good one, they double down and just produce twice as much.
00:12:57.679 --> 00:13:01.279
But they just wave them wendell, right?
00:13:01.519 --> 00:13:03.519
Chicago reference right there, right?
00:13:03.600 --> 00:13:05.600
They just let everyone go in.
00:13:06.000 --> 00:13:06.720
They don't care.
00:13:06.799 --> 00:13:08.080
They're like, all right, everyone's safe.
00:13:08.159 --> 00:13:09.919
You're in, you're in, you're in, you're in.
00:13:10.080 --> 00:13:15.440
It's a passive approval of just automating uh things that need more oversight.
00:13:17.200 --> 00:13:21.840
And there's a a concept, uh, a concept here that I want to talk about.
00:13:21.919 --> 00:13:24.240
Uh, it's Mdahl's law.
00:13:24.399 --> 00:13:33.759
And that's essentially you can speed up one part of the process or many parts of the process, but the whole system is bottlenecked by whoever can't keep up, right?
00:13:33.840 --> 00:13:41.600
It's the uh the old sports, you know, cliche or the you know teamwork cliche that you're only, you know, strong is your weakest link.
00:13:41.759 --> 00:13:47.360
And this is true for the human AI collaboration, right?
00:13:47.440 --> 00:13:49.679
And and this is why this is so important.
00:13:50.799 --> 00:13:55.919
So few business leaders zoom out and think about this.
00:13:57.120 --> 00:14:01.600
So few people think of the human AI relationship, right?
00:14:02.080 --> 00:14:10.639
So few people working on on front-end AI strategy, back end AI uh implementation go through and ask these questions.
00:14:11.759 --> 00:14:12.240
Right?
00:14:13.039 --> 00:14:20.000
What happens when multi-agentic systems are way smarter than humans and the humans don't know how to run them?
00:14:20.879 --> 00:14:21.440
Right?
00:14:21.840 --> 00:14:30.240
And the human that two or three years ago was integral is now a liability, right?
00:14:30.320 --> 00:14:32.559
And and how do we start to change that human role?
00:14:32.639 --> 00:14:35.200
Because yes, humans are still needed in all this, right?
00:14:35.279 --> 00:14:37.600
I never said that, and I I'm not implying that.
00:14:37.840 --> 00:14:40.960
Human roles are changing, and I'm gonna get to what that looks like.
00:14:41.679 --> 00:14:45.279
But that's why we need human reviews everything.
00:14:46.399 --> 00:14:48.639
That's why we need that, right?
00:14:48.799 --> 00:14:49.679
That's why you have to.
00:14:49.840 --> 00:14:51.679
I'm a huge advocate, right?
00:14:51.840 --> 00:14:55.360
If you're using front-end large language models, what do you do?
00:14:55.600 --> 00:14:58.720
Like people, like people are like, okay, what do you do with all this time savings?
00:14:58.799 --> 00:15:01.360
Or what do you do if you're using a thinking model while you wait?
00:15:01.519 --> 00:15:03.519
Well, you read the chain of thought, right?
00:15:03.600 --> 00:15:13.120
And you rerun it and you correct it and you bring in more of your company's data, right, at the right points before the model goes too deep into its dive.
00:15:15.919 --> 00:15:20.720
And that's why the fix is expert-driven loops.
00:15:20.879 --> 00:15:27.840
That's what I've been advocating for, you know, EDL uh for a long time now, not just generic oversight.
00:15:28.080 --> 00:15:34.720
Because the difference is when you embed experts in building, right?
00:15:34.879 --> 00:15:41.360
Whether it's, you know, we don't have we don't even have to have to talk about, you know, get too technical and talk about, you know, multi-agentic loops.
00:15:41.440 --> 00:15:50.799
Let's just talk about embedding the right experts and setting up your team's AI processes in your ChatGPT Teams account, right?
00:15:51.679 --> 00:15:55.200
Or how you're gonna tackle work and claude co-work.
00:15:55.519 --> 00:16:04.799
A lot of times you have one person, sometimes it's IT, sometimes it's you know, your your AI champion team, which is important.
00:16:05.600 --> 00:16:09.120
And they kind of do it for everyone and they set it up for everyone.
00:16:09.360 --> 00:16:11.759
And that's again the wrong way.
00:16:11.919 --> 00:16:22.639
You know, one thing that was interesting, there was a legal on technology study kind of looking at this concept of the, you know, human in the loop versus an expert loop.
00:16:22.720 --> 00:16:28.799
Um, and they saw that one law firm that they looked at put senior partners in the loop, right?
00:16:28.960 --> 00:16:35.600
Where normally this is uh, you know, not in a bad way, but you know, kind of the human in the loop overseeing AI.
00:16:35.759 --> 00:16:42.240
A lot of times they put younger people or more inexperienced people, uh, people who cost less because that's what companies think, right?
00:16:42.320 --> 00:16:46.000
They're like, okay, well, this is just someone clicking a button, clicking approve.
00:16:46.159 --> 00:16:48.159
Why am I going to put my senior people on this?
00:16:48.240 --> 00:16:50.320
Why am I gonna put my smartest people on this?
00:16:50.559 --> 00:17:06.720
So, in this case study from Legalon Technologies, they found 86% faster contract review and 65% better issue detection when they had senior partners instead of junior reviewers.
00:17:08.160 --> 00:17:12.480
86% faster and 65% better.
00:17:12.720 --> 00:17:16.559
And that's not like versus the human-only baseline.
00:17:16.880 --> 00:17:23.839
That is better than the uh AI, the augmented, right, junior reviewers plus AI.
00:17:24.000 --> 00:17:27.759
So, I mean, you're getting compoundingly better and better results.
00:17:28.000 --> 00:17:34.720
The smarter and the more expert uh people that you put in the right places.
00:17:34.880 --> 00:17:45.279
And I think it's it's it's it's again, it's not putting one expert, you know, on one AI-powered workflow or one, you know, agent run.
00:17:45.599 --> 00:17:49.519
It's putting multiple people in there at the right place, right?
00:17:49.839 --> 00:17:55.359
It's experts driving the loop, not a single human overseeing.
00:17:56.079 --> 00:18:06.880
And other studies show this, other organizations are seeing the ROI triple when they move from generic oversight to expert-driven collaboration.
00:18:08.640 --> 00:18:12.400
So how does this get to this point?
00:18:13.039 --> 00:18:13.680
Right?
00:18:14.400 --> 00:18:23.440
And it's almost like the better the technology gets and the more advanced it gets, and the less tech know-how it requires, right?
00:18:23.519 --> 00:18:30.799
That anyone can go in there and click a button and you know, set up AI agents that are connected to your data, literally, right?
00:18:30.880 --> 00:18:33.680
This show's been going for 19 minutes.
00:18:33.920 --> 00:18:37.599
You could have set the hundreds of agents up in 19 minutes.
00:18:37.680 --> 00:18:38.960
I kid you not, right?
00:18:39.200 --> 00:18:41.839
Um, one click, very easy.
00:18:42.240 --> 00:18:57.200
But poorly implemented AI can crush productivity because you just end up spending more time correcting your errors and managing expectations that maybe went awry, and then running multiple parallel backups.
00:18:57.440 --> 00:19:09.920
So I like to say this if you had a bad workflow with AI and you upskilled or reskilled that workflow, right?
00:19:11.680 --> 00:19:16.799
You you can't just throw makeup on an ugly process and think it's gonna be pretty, it's still ugly.
00:19:17.680 --> 00:19:20.240
Now it's just like, you know, got some makeup on it.
00:19:20.880 --> 00:19:23.200
Got a little shine that it doesn't deserve.
00:19:24.160 --> 00:19:29.200
All this does is it recreates workflows that weren't working.
00:19:29.440 --> 00:19:50.640
And if anything, it just creates this augmentation debt that ultimately tanks productivity because now you just have to go back, you know, and instead of getting more things done in a better way, now you're just getting more things that need fixed faster with potentially more airs because you're not putting the right people in the right processes, right?
00:19:50.799 --> 00:19:59.119
You're putting anyone in old processes, you have to rebuild that to be AI native.
00:20:00.960 --> 00:20:03.440
So, how do you do this?
00:20:06.480 --> 00:20:08.400
It's a mindset shift.
00:20:09.039 --> 00:20:19.839
Like I said, you're not the one if your team wants to excel and outrun the competition in 2026 and beyond.
00:20:22.400 --> 00:20:26.880
You can't just use AI, you can't leverage AI.
00:20:28.079 --> 00:20:29.839
You have to orchestrate it.
00:20:30.240 --> 00:20:30.720
Right.
00:20:30.880 --> 00:20:36.319
What's funny is most most times, this is one of the times I you know, I'm looking at my other screen here.
00:20:36.480 --> 00:20:42.559
Um, you know, I actually don't have agents running right now, uh, where normally I would, right?
00:20:42.880 --> 00:20:52.240
For the most part, we have to start thinking of ourselves as orchestrators, or I like to say tastemakers sometimes, right?
00:20:52.720 --> 00:20:55.599
You need to provide that context, right?
00:20:55.759 --> 00:21:03.519
Context is gonna be one of those, you know, context engineering uh is gonna be one of those buzzwords of 2026, right?
00:21:03.680 --> 00:21:07.839
In 2024, I called it uh first company data, right?
00:21:08.000 --> 00:21:12.160
Um, but I still think that that's more realistic in what you need.
00:21:12.319 --> 00:21:13.599
Context can be anything, right?
00:21:13.759 --> 00:21:15.680
Context needs more context to be defined.
00:21:15.839 --> 00:21:25.839
But you know, context engineering, uh to put simply, you know, when I said, hey, in 2023, you were prompting and then you were iterating, and then you were providing more context, right?
00:21:26.000 --> 00:21:32.000
That's just more data, more data, more direction to a model before it goes off and do its uh does its thing.
00:21:32.240 --> 00:21:41.200
Because, you know, before when the models were just, you know, non-reasoning models, non-thinking models, they would just spit something back pretty quickly, right?
00:21:41.359 --> 00:21:44.720
Now they might go work for five minutes, right?
00:21:44.880 --> 00:21:47.039
Uh, 10 minutes longer.
00:21:47.680 --> 00:21:51.759
Uh, I had one run the other day on the front end, not on the back end, right?
00:21:51.920 --> 00:21:56.720
On the back end, if you're using it via the API, it's pretty easy to get these things working for hours.
00:21:56.799 --> 00:21:59.759
On the front end, you know, yeah, you might have a model work for five, 10, 15.
00:22:00.160 --> 00:22:00.960
20 minutes.
00:22:01.039 --> 00:22:08.640
Uh, I had an actual model, not a deep research run, uh, run for I think 92 minutes uh the other day.
00:22:08.880 --> 00:22:09.359
Right.
00:22:09.519 --> 00:22:12.079
You have to give it the context.
00:22:12.240 --> 00:22:15.680
And then at that point, you're orchestrating, right?
00:22:16.000 --> 00:22:21.599
Um I'm I'm looking at different models, what they bring to me, different agents, what they bring to me.
00:22:21.680 --> 00:22:23.920
And I'm saying this is good, this is good, this isn't.
00:22:24.079 --> 00:22:25.599
Let's rebuild that skill, right?
00:22:25.680 --> 00:22:29.599
If I'm in Claude, I'm sitting, let's I'm saying let's update this skill, right?
00:22:29.680 --> 00:22:33.519
If I'm in uh Chad GPT, I'm going and updating that GPT, right?
00:22:33.599 --> 00:22:41.920
You always have to be improving the processes and not just trying to sprinkle some AI on an old process that's broken.
00:22:42.079 --> 00:22:46.400
And that's why we're talking about these things like Asian supervisors, op orchestrators.
00:22:46.640 --> 00:22:50.480
This is a fundamental shift in how work is getting done.
00:22:50.640 --> 00:22:59.200
This is the difference in the human AI collaboration and why I think you need to rethink your working relationship with AI.
00:22:59.599 --> 00:23:07.039
Because if you're still using it like you were in, you know, late 2022 or 2023, right?
00:23:07.200 --> 00:23:13.839
If you're going out of your way, we tackled this earlier in our start here series about treating AI like an operating system.
00:23:15.839 --> 00:23:19.920
So one, that part covers the tech, right?
00:23:20.240 --> 00:23:22.480
Treating AI like an operating system.
00:23:23.039 --> 00:23:28.960
This part, volume four here, while we're tackling the human AI collaboration, that's the mindset shift.
00:23:29.039 --> 00:23:38.799
That's going from an operator, I'm the one pushing the buttons, to nope, I'm orchestrating an agent that's technically going out there and pushing the buttons and coming back.
00:23:38.880 --> 00:23:41.519
And then I'm telling it how to improve, right?
00:23:41.759 --> 00:23:44.640
I am building that expertise, even if you don't know.
00:23:44.799 --> 00:23:54.079
Hey, what is that one thing that you know you're definitely smarter than these AI models that are genius level on offline IQ tests, right?
00:23:54.240 --> 00:24:00.400
You might be saying, okay, what could I be smarter than an AI model that knows everything if you give it the right context?
00:24:00.559 --> 00:24:03.440
Well, you don't know until you look at its chain of thought.
00:24:03.599 --> 00:24:13.680
You don't know until you know you've put in, you know, five, 10, 15, 30 hours on a project, having multiple AI models go and do something that you know how to do front to back.
00:24:13.839 --> 00:24:16.319
That's how you carve out your expertise.
00:24:16.480 --> 00:24:22.880
And then you really have to then deploy that and duplicate it across your team or your organization.
00:24:23.279 --> 00:24:34.000
That is how you go about shifting your mindset to go from an operator to an orchestrator, because then that output compounds, right?
00:24:34.240 --> 00:24:37.279
Your ability to generate revenue compounds.
00:24:37.519 --> 00:24:44.559
Your ability to do things that you didn't have time to do last year, all of a sudden that frees up, right?
00:24:44.640 --> 00:24:49.359
And this is where the most advanced businesses are shifting toward right now.
00:24:50.240 --> 00:25:03.279
So the last thing I want to talk about last two things is where our expertise actually belongs.
00:25:04.000 --> 00:25:04.319
Right?
00:25:04.640 --> 00:25:06.319
Where should we be spending our time?
00:25:06.799 --> 00:25:14.319
Well, you have to go do what I just said, you have to make that transition successfully from operator to orchestrator.
00:25:14.480 --> 00:25:18.160
But then I want you to think about kind of this jagged frontier of capabilities, right?
00:25:18.240 --> 00:25:26.400
Because there's things that humans are good at, and then there's things that we're terrible at.
00:25:26.480 --> 00:25:29.039
And those things are changing all the time, right?
00:25:29.279 --> 00:25:30.799
And the same thing with AI models.
00:25:30.880 --> 00:25:34.160
There's things that AI models are amazing at that we never thought would be possible.
00:25:34.319 --> 00:25:36.160
Then there's things that they kind of fail at.
00:25:36.319 --> 00:25:38.480
So right now, this is today, right?
00:25:38.559 --> 00:25:43.039
So if you're listening to this in January, February 2026, probably still safe.
00:25:43.119 --> 00:25:47.519
If you're listening to it in, you know, July, September, this could be different.
00:25:47.920 --> 00:25:58.960
But right now, humans win at high context empathy, ambiguous decisions with incomplete data, accountability, and novel judgment, right?
00:25:59.200 --> 00:26:05.759
Reading between the lines where there's not structured data or there's not company context to fill those cracks.
00:26:05.920 --> 00:26:09.200
Right now, AI wins at pretty much everything else, right?
00:26:09.440 --> 00:26:13.680
Data synthesis, first drafts, pattern recognition, repetitive cognition, right?
00:26:13.759 --> 00:26:15.119
All of those things.
00:26:15.920 --> 00:26:24.640
Those are the danger zones for if those are your competitive motes, or you think that's your competitive advantage right now, right?
00:26:24.720 --> 00:26:38.559
If you think uh one of those things, data synthesis, first drafts, pattern recognition, if that's what your department, your career, your company is built on right now, you you you've got to find your pivot.
00:26:40.960 --> 00:26:46.000
Because right now, the number one predictor of human AI success isn't just the model you use.
00:26:46.400 --> 00:26:53.200
It's the quality of the context and kind of those procedures that you create and you provide.
00:26:53.440 --> 00:26:59.599
Because if garbage context goes in, a worse outcome, a worse outcome comes out.
00:27:01.119 --> 00:27:03.839
So here's some quick takeaway advice.
00:27:04.640 --> 00:27:05.119
Right?
00:27:05.519 --> 00:27:11.519
I can't sit here and give you uh advice for how to set up your agent orchestration.
00:27:11.599 --> 00:27:17.039
What I can tell you, your individual using different platforms, right?
00:27:17.519 --> 00:27:22.960
One of the biggest mistakes is skipping over repeatable and scalable context.
00:27:23.279 --> 00:27:26.400
So stop starting from zero, right?
00:27:26.880 --> 00:27:31.839
And you need to take this, start this at the individual level, but then take this to your team.
00:27:32.079 --> 00:27:34.240
Stop starting at zero every single prompt.
00:27:34.319 --> 00:27:37.359
You have to start building uh context faults.
00:27:38.000 --> 00:27:38.400
Right.
00:27:38.559 --> 00:27:40.319
You can think of those as skills, right?
00:27:40.400 --> 00:27:44.559
You can have a markdown file with all your skills with your company knowledge, right?
00:27:44.720 --> 00:27:54.319
If you use uh Claude skills, you know, you're probably familiar with these markdown files, but start creating these skill files based on the tasks that you and your team repeatedly do.
00:27:54.480 --> 00:27:56.960
This is something I'm constantly updating mine, right?
00:27:57.039 --> 00:28:02.960
If you're using, as an example, Claude, you can update them in um kind of in chat.
00:28:03.119 --> 00:28:07.200
If you're using uh the GPT builder in ChatGPT, you can do it there as well.
00:28:07.359 --> 00:28:07.839
Uh right.
00:28:07.920 --> 00:28:17.599
But you need to be building and reusing these uh, you know, custom GPTs inside of uh inside of ChatGPT, the Claude projects and skills, Google Gems, whatever.
00:28:18.000 --> 00:28:21.759
You need to have that rag for your personal use, right?
00:28:21.839 --> 00:28:31.839
We think of when we think about retrieval augmented generation, we usually think about these you know vector databases that would cost millions of dollars three, four, five years ago to build.
00:28:32.000 --> 00:28:33.599
And we think about these very complex things.
00:28:33.759 --> 00:28:38.480
I want you to think about your personal uh rag, your personal retrieval augmented generation.
00:28:38.640 --> 00:28:39.359
What are those things?
00:28:39.440 --> 00:28:41.680
I update mine all the time, right?
00:28:41.920 --> 00:28:46.319
I have my, you know, as a small business owner, these are the things that I'm thinking about.
00:28:46.480 --> 00:28:51.920
These are uh, you know, my important facts, stats about my role, about what I'm trying to drive.
00:28:52.079 --> 00:28:55.440
Here's my KPIs, and I'm constantly, you know, updating these things.
00:28:55.519 --> 00:29:02.480
But you need to have your personal context, your team context, uh, your company context, your competitive landscape context.
00:29:02.640 --> 00:29:08.400
You have to have these things that are reusable because if you're just starting at zero, you're wasting time.
00:29:09.039 --> 00:29:11.680
That's being a button pusher, right?
00:29:12.000 --> 00:29:17.039
Instead of agents that don't need to have the button pushed.
00:29:17.119 --> 00:29:19.119
They're already doing it, they're already delivering it.
00:29:19.279 --> 00:29:23.359
They have all that and they're able to use it in a repeatable and scalable way.
00:29:24.240 --> 00:29:30.079
And then last but not least, you need to elevate your champions.
00:29:30.480 --> 00:29:32.240
Okay, here's what I mean by that.
00:29:32.480 --> 00:29:40.720
Uh uh funny, I had uh Chris Caldwell, the CEO from Concentrics on uh a couple of weeks ago.
00:29:40.880 --> 00:29:44.240
And he kind of said, Hey, you don't want 100 Jordans running around, do you?
00:29:44.400 --> 00:29:45.200
And you don't.
00:29:45.440 --> 00:29:47.119
But I'll tell you this.
00:29:49.519 --> 00:29:56.880
And I don't want this to come off in the wrong way, but you need people like me on your team, and you need a lot of them.
00:29:57.279 --> 00:30:05.200
Here's what I need: you need people whose mainly their only job is to keep up with AI every single day, right?
00:30:05.680 --> 00:30:11.200
Every large organization needs dozens of people who are in my position.
00:30:11.440 --> 00:30:14.720
All they do, they they read about AI every day.
00:30:14.799 --> 00:30:20.400
Uh, they're scoping out new projects, uh, measuring, building modular backups, right?
00:30:20.480 --> 00:30:26.960
Because you don't want all of your uh you know systems in one model, the model changes and your whole company comes to comes to a halt.
00:30:27.039 --> 00:30:31.279
Uh, and then you need to constantly have those people training the non-champions.
00:30:31.519 --> 00:30:35.039
So you need to find where your time savings opportunities are.
00:30:35.200 --> 00:30:41.440
You need to scope those, measure those, and then deploy the ones that can easily gain back time the quickest.
00:30:41.599 --> 00:30:49.359
And then you apply those hour saves into creating those, you know, go create your dozen Jordans uh on your team, right?
00:30:49.519 --> 00:30:51.839
Your domain experts, right?
00:30:52.079 --> 00:31:04.799
It might feel weird to start automating some of their work or automating some of their roles, but then you find those people that are actually able to automate parts of their job and then teach others, build scalable systems.
00:31:04.960 --> 00:31:09.519
There's still new lines of revenue uh to build in your company, right?
00:31:09.680 --> 00:31:28.559
Uh, people, when they look at AI, yes, I do ultimately think AI will um cause a net negative in the job market, but there's millions, tens of millions of new jobs that we have no clue that are gonna exist in three years.
00:31:28.880 --> 00:31:36.799
And you need those people, those champions on your team, you need to elevate them, challenge them, and you need to deploy them, right?
00:31:36.960 --> 00:31:50.160
They need to be listening to this show every day and other, you know, AI podcasts and other, you know, YouTube people on AI reading newsletters, and you need to be scoping, breaking, and training people every single day.
00:31:50.319 --> 00:31:53.519
But the largest organizations aren't even doing that.
00:31:53.920 --> 00:31:56.880
So stop looking for cool AI tricks.
00:31:57.119 --> 00:32:00.240
Focus on automating the dull stuff first, right?
00:32:00.559 --> 00:32:03.279
Get rid of shiny AI object syndrome.
00:32:03.519 --> 00:32:11.920
Go, the invoices, the summarization, the filing, all those boring AI things beat the flashy AI every single time.
00:32:12.079 --> 00:32:24.720
Because as AI starts to handle more and more digital interaction, face-to-face and high empathy relationships are going to become your company's differentiators, right?
00:32:24.880 --> 00:32:28.400
But you can't have that if you're still the one pushing buttons.
00:32:28.640 --> 00:32:29.359
You can't.
00:32:29.599 --> 00:32:32.240
The human premium is rising.
00:32:32.640 --> 00:32:34.240
So you have to use it wisely.
00:32:34.319 --> 00:32:58.720
But you can only do that if you go back and follow the steps that we just laid out in re-examining the human AI collaboration, going through the best practices of shifting away from being the operator, from being the button, the button pusher, the chat GPT prompter into being the orchestrator, the tastemaker, and the champion that's pushing your organization to do the same top to bottom.
00:32:58.880 --> 00:32:59.119
All right.
00:32:59.200 --> 00:33:02.400
I hope this Start Here series was helpful.
00:33:02.559 --> 00:33:04.720
Volume four done in the books.
00:33:04.799 --> 00:33:09.359
Uh, so if this was helpful, remember please go to startheseries.com.
00:33:09.519 --> 00:33:14.160
That's gonna give you access for free to our inner circle community.
00:33:14.319 --> 00:33:18.000
And you're gonna see all of the start here uh episodes right there.
00:33:18.079 --> 00:33:22.640
We're gonna be throwing in and adding more and more additional resources to help you with your journey.
00:33:22.720 --> 00:33:27.359
So whether you're just starting out, I hope this episode helped you better understand some things.
00:33:27.519 --> 00:33:35.599
If you're an expert doing this every day, I hope this challenged you to look at AI a little bit differently and to push even your own human AI collaboration.
00:33:35.759 --> 00:33:37.200
So, thank you for tuning in.
00:33:37.279 --> 00:33:40.799
I hope to see you back tomorrow and every day for more everyday AI.
00:33:40.960 --> 00:33:42.079
Thanks, y'all.
00:00:03.680 --> 00:00:04.639
Be honest.
00:00:04.879 --> 00:00:08.880
How has your working relationship with AI changed in the past few years?
00:00:09.119 --> 00:00:13.119
Are you still copying, pasting into ChatGPT and hoping for the best?
00:00:13.359 --> 00:00:19.120
Are you drowning in AI-generated drafts that need more editing than if you've just done it yourself?
00:00:19.359 --> 00:00:24.000
Or maybe you figured out a few tricks, but something still feels off.
00:00:24.640 --> 00:00:31.280
Here's the thing: most people are still treating AI like a junior version of themselves with better grammar.
00:00:31.519 --> 00:00:32.960
But the game has changed.
00:00:33.200 --> 00:00:35.759
In 2023, you learned to prompt.
00:00:36.000 --> 00:00:39.119
In 2024, hopefully you learn to iterate.
00:00:39.280 --> 00:00:44.240
And in 2025, you started wondering why this still feels so manual.
00:00:44.560 --> 00:00:48.719
And now in 2026, the skill gap isn't technical anymore.
00:00:48.880 --> 00:00:50.159
It's managerial.
00:00:50.320 --> 00:00:51.280
It's human.
00:00:51.439 --> 00:00:53.200
It's self-reflection.
00:00:53.359 --> 00:00:59.520
Because the teams winning with AI aren't better prompters or they don't have access to better models than you.
00:00:59.679 --> 00:01:07.680
They're just better orchestrators of AI, better at throwing away the old way of work and starting fresh with AI front and center.
00:01:07.840 --> 00:01:13.280
And that human AI collaboration shift changes everything about how you work.
00:01:15.760 --> 00:01:16.400
All right.
00:01:16.799 --> 00:01:23.040
Does that get you a little fired up to reexamine your working relationship with AI?
00:01:23.280 --> 00:01:24.079
I hope it is.
00:01:24.159 --> 00:01:29.120
And welcome to volume four of the Start Here series with Everyday AI.
00:01:29.200 --> 00:01:34.799
We're going to be looking at the human AI collaboration and best practices for working alongside AI.
00:01:35.040 --> 00:01:41.599
So if you're new here, after almost 700 episodes, we have kicked off our start here series.
00:01:41.680 --> 00:01:43.200
Like I said, this is volume four.
00:01:43.359 --> 00:01:48.640
This is essential beginner and advanced AI overviews to launch your year strong.
00:01:48.799 --> 00:01:56.480
So if one of your big goals in 2026 was to learn AI or to double down, maybe you've been there since day one.
00:01:56.640 --> 00:01:59.680
This Start Here series is for you.
00:02:00.000 --> 00:02:03.120
Uh and we have a special resource.
00:02:03.280 --> 00:02:06.719
So make sure you go to starthearseries.com.
00:02:06.879 --> 00:02:11.919
That is going to get you access for free to our inner circle AI community.
00:02:12.000 --> 00:02:15.120
And it's going to take you straight to our Start Here series page.
00:02:15.280 --> 00:02:18.960
So you can very easily catch up with all of the Start Here series.
00:02:19.039 --> 00:02:27.199
Uh, listen, additional resources, related episodes, anything you need to get either started or centered on your AI journey.
00:02:27.360 --> 00:02:31.039
And if you are very new here, this is part of Everyday AI.
00:02:31.120 --> 00:02:39.360
It's a daily live stream podcast and free daily newsletter, helping everyday business leaders like you and me keep up with AI and get ahead and use it to grow our companies and our careers.
00:02:39.520 --> 00:02:49.680
So the start here series, we're going to do about a dozen or so of these episodes in the first part of 2026, uh, that I hope is going to be a refreshing look at AI for all of us.
00:02:49.759 --> 00:02:56.639
Because, yeah, as someone that's done this now almost 700 times, I understand how it can be so hard to just start somewhere.
00:02:56.800 --> 00:02:58.400
So this is for you.
00:02:58.560 --> 00:03:04.719
And if you missed our last uh episode, it was volume three of the Start Here series, AI as an operating system.
00:03:05.039 --> 00:03:08.719
So if you are listening uh on the podcast, make sure to check your show notes.
00:03:08.800 --> 00:03:11.759
We're going to be updating the show notes of all the start here series.
00:03:11.840 --> 00:03:16.240
So you can very easily, if you're listening to the podcast, just flip around that way as well.
00:03:16.560 --> 00:03:21.439
All right, let's get straight into it and talk about well, why?
00:03:21.680 --> 00:03:25.840
Why is there this big skill gap in 2026?
00:03:26.000 --> 00:03:36.639
There's those people that are running away with AI, and then there's people that have been using it pretty much every day or every week since the Chat GPT moment of November 2022.
00:03:36.879 --> 00:03:39.680
And they're still just barely keeping up.
00:03:40.000 --> 00:03:42.560
So it's a shift.
00:03:42.800 --> 00:03:46.800
It's a shift from operator to orchestrator.
00:03:46.960 --> 00:03:51.360
And that's what I think that you have to start thinking about, uh, right?
00:03:51.599 --> 00:04:06.560
Your relationship with AI, and if you really want to stay ahead of the best practices, because your new job isn't doing the work, it's really defining on what the work looks like, how it gets done, what information the work needs.
00:04:06.719 --> 00:04:10.159
You set the parameters, the constraint, and the success criteria.
00:04:10.479 --> 00:04:15.439
And the AI is probably going to be the one doing the actual work, right?
00:04:15.599 --> 00:04:21.040
I talked about this literally in 2024, this concept of you know agent orchestration.
00:04:21.120 --> 00:04:23.680
And that's how I believe that that's where we're headed.
00:04:23.759 --> 00:04:29.199
And I think we've really seen that a lot in the last uh, gosh, only four to five weeks.
00:04:29.360 --> 00:04:34.000
That's really coming to kind of mainstream fruition uh on this big shift.
00:04:34.160 --> 00:04:41.759
And I think the winners of this era aren't those that who can effectively, you know, onboard, supervise, and audit digital agents.
00:04:41.920 --> 00:04:47.600
It is really thinking it like an entire team that you are now orchestrating, right?
00:04:47.759 --> 00:04:51.120
Not an entire team that you are a player coach.
00:04:51.279 --> 00:04:52.720
You are the manager.
00:04:52.879 --> 00:04:55.600
You're not out on the field doing all of the work.
00:04:56.560 --> 00:05:03.040
So here's what we're gonna cover in today's uh iteration of the Start Here series.
00:05:03.199 --> 00:05:12.000
We're gonna be talking about the comfortable, uncomfortable truths of AI in 2026 and our relationship as humans with AI.
00:05:12.240 --> 00:05:20.720
And I think what so many people do is they take their current antiquated processes that a lot of times are broken, and they just find ways to put AI in it, right?
00:05:20.800 --> 00:05:21.759
In the front and the middle.
00:05:21.920 --> 00:05:24.480
Oh, they find a broken joint, a leaky funnel.
00:05:24.639 --> 00:05:26.319
Let's slap some AI on it, right?
00:05:26.480 --> 00:05:29.920
Do you guys remember the uh uh the infomercial, right?
00:05:30.000 --> 00:05:31.839
You slap some flex seal on it.
00:05:31.920 --> 00:05:33.120
That's what people look at AI.
00:05:33.199 --> 00:05:34.720
They find something that's broken or something.
00:05:34.879 --> 00:05:36.399
Ah, this can be improved a little bit.
00:05:36.480 --> 00:05:38.000
Let's slap some AI on it.
00:05:38.160 --> 00:05:40.000
That's the wrong way to do it.
00:05:40.160 --> 00:05:45.040
Um, I think that it's about unlearning, right?
00:05:45.519 --> 00:05:49.120
I'm gonna be griping on a couple of these terms that I absolutely hate.
00:05:49.279 --> 00:05:50.480
One is human in the loop.
00:05:50.639 --> 00:05:51.759
I'll get to that later.
00:05:51.920 --> 00:05:54.240
One is upskilling or reskilling.
00:05:54.480 --> 00:05:58.879
That alone, I think, has set so many companies back years.
00:05:59.600 --> 00:06:08.079
Because when you, you know, the C-suite and the boardroom, uh, you know, go around HR and they're looking at, you know, AI investment.
00:06:08.160 --> 00:06:09.920
They're like, who needs upskilled?
00:06:10.079 --> 00:06:11.519
Who needs reskilled?
00:06:11.680 --> 00:06:19.519
They're looking at it almost as a um proactive or uh or sorry, a reactive measure, right?
00:06:19.600 --> 00:06:20.959
It's almost like a course correction.
00:06:21.040 --> 00:06:23.120
Oh, who needs who needs upskilled this year?
00:06:23.279 --> 00:06:24.879
Who needs reskilled in AI?
00:06:25.040 --> 00:06:27.199
No, you need to unlearn, right?
00:06:27.279 --> 00:06:43.519
I've been uh chirping out uh this unlearned word for a very long time because it's how I really started getting the most out of AI myself when I started unlearning good habits that had traditionally led to success.
00:06:43.920 --> 00:06:51.519
And I think that's when you look at your relationship with AI, your working relationship with AI, that's what you have to start to do.
00:06:51.680 --> 00:06:56.959
You have to start to say, no, why would I just upskill with AI?
00:06:57.120 --> 00:07:01.199
My skills are, again, they're not worthless, they're worthless.
00:07:01.360 --> 00:07:14.879
So if you're still holding on to those skill sets, right, whether you've been using them for two years or 30, it's probably time to let them go and to unlearn and then relearn and rebuild from scratch.
00:07:16.560 --> 00:07:17.600
Well, why?
00:07:17.759 --> 00:07:30.399
Well, because right now we have, I'm not gonna say superhuman AI, but I will say above human, above human expert AI that can make millions of decisions per second worldwide.
00:07:30.560 --> 00:07:32.399
And humans can't keep up, right?
00:07:32.480 --> 00:07:36.319
So we think of this concept of human in the loop, and that's our job, right?
00:07:36.480 --> 00:07:42.959
And that's a job that we thought was gonna last into the, you know, into the late 2020s.
00:07:43.040 --> 00:07:43.519
It's not.
00:07:43.759 --> 00:07:47.839
Human in the loop, in my opinion, is dead on arrival.
00:07:47.920 --> 00:07:56.160
It was dead on arrival, uh, right, because we thought that you could stick any human in the loop and okay, let's have this human go look what the AI is doing.
00:07:56.399 --> 00:07:56.720
No.
00:07:57.439 --> 00:07:58.720
It is dead, right?
00:07:58.800 --> 00:08:06.079
Because agentic workflows really create miles-long action traces that humans cannot realistically interpret.
00:08:06.240 --> 00:08:07.360
We can't keep up.
00:08:07.600 --> 00:08:12.480
The complexity has out far outpaced what a human can review, right?
00:08:12.639 --> 00:08:24.160
When you see a large language model, or if you're working uh with agents that have sub-agents and they can work around the clock, what good is a human gonna do at that point?
00:08:24.959 --> 00:08:25.360
Right?
00:08:25.519 --> 00:08:27.120
A human in the loop, right?
00:08:27.360 --> 00:08:29.120
Like, I mean, let's say what it is.
00:08:29.279 --> 00:08:35.919
It's it's it's your uh last, you know, your last hope that you didn't get something wrong.
00:08:36.000 --> 00:08:40.240
It's your last hope that a hallucination doesn't slip its way through to production, right?
00:08:40.320 --> 00:08:47.039
It's your last hope that uh you don't make a tragic mistake from uh what a bunch of agents did.
00:08:47.759 --> 00:08:53.759
Because if your human in the loop can't push back, pause, or ask hard questions, that's not even oversight.
00:08:53.919 --> 00:08:57.440
That's a fail safe, probably a name only, right?
00:08:58.480 --> 00:09:07.440
No human in the loop can keep up with what agentic AI can do today, let alone next week and next month.
00:09:07.600 --> 00:09:10.399
And yes, it does change that quickly.
00:09:10.559 --> 00:09:12.000
Uh, my gosh.
00:09:12.960 --> 00:09:15.919
So this isn't hype, this is measurable.
00:09:16.080 --> 00:09:19.279
And the problem is, well, models are smarter than us.
00:09:19.440 --> 00:09:25.279
So why would we want to keep a general or generic human in the loop?
00:09:25.519 --> 00:09:32.240
You need to find what your expertise is or you know, your team's expertise.
00:09:32.399 --> 00:09:38.159
What is that one thing that you are always better than the best large language models at?
00:09:38.320 --> 00:09:50.960
And that's what you should be focusing on and double down, uh doubling down on and seeing where the overlap is between that one thing and what ultimately makes your company or your team more revenue, uh, right?
00:09:51.120 --> 00:09:57.440
Because you have to stop comparing AI to perfection, compare it to the human speed and accuracy for the same tasks.
00:09:57.759 --> 00:10:01.120
And at that point, humans can't keep up.
00:10:01.360 --> 00:10:04.559
And I'm not gonna say every single use case, right?
00:10:04.639 --> 00:10:08.159
But I'll say the overwhelming majority of use cases.
00:10:08.799 --> 00:10:17.519
If you have an average skilled human, right, pick pick your pick your niche, pick your vertical, right?
00:10:17.759 --> 00:10:24.559
Financial analysis, uh, marketing, uh PR, whatever.
00:10:25.200 --> 00:10:26.159
Pick your job.
00:10:26.480 --> 00:10:27.919
Find an average person.
00:10:28.480 --> 00:10:38.399
Someone that's you know worked there for one of your colleagues, but that colleague that's amazing at AI and they know everything, and then pick the smartest person in that same vertical.
00:10:39.600 --> 00:10:44.879
The smartest person in that vertical cannot compete with the person who knows enough, right?
00:10:45.519 --> 00:10:48.879
They speak the language, but is a whiz at AI.
00:10:49.120 --> 00:10:52.080
It's not, it's it's not close.
00:10:52.240 --> 00:10:54.720
It's not even a competition, right?
00:10:55.120 --> 00:11:00.559
That's like having 20 Michael Jordans on a team versus you know, a Pee-Wee team.
00:11:00.639 --> 00:11:01.840
It's not fair.
00:11:02.240 --> 00:11:33.200
So I think that's what we have to get to is we have to understand that these models now that are agentic by default, that can reason, that can think, they can plan ahead, uh, they can they can iterate, loop back, take different paths, use tools all faster than humans and all spin off agents that do the same thing and you know can do it hundreds at a time, you know, spin off hundreds of sub-agents and then come back and you know, have multiple, you know, mixture of experts or you know, what I like to call mixture of models uh that that judge all of those inputs, right?
00:11:33.279 --> 00:11:35.840
You humans can't compete anymore.
00:11:40.159 --> 00:11:42.480
So why?
00:11:43.600 --> 00:11:49.360
Why are we still thinking that the human AI collaboration in 2026, when it comes to AI, is human in the loop?
00:11:49.440 --> 00:11:50.240
It's not.
00:11:50.559 --> 00:11:51.120
Uh, right?
00:11:51.279 --> 00:12:00.320
There was a uh a recent study that looked at 700 plus court cases worldwide, now involve AI hallucinations and fabricated citations.
00:12:00.399 --> 00:12:07.120
And that rate is accelerating to multi to a handful of new case uh new cases daily.
00:12:07.759 --> 00:12:16.480
And that's because the AI is too fast, humans are too lazy, and for the most part, organizations aren't training their people.
00:12:16.639 --> 00:12:23.919
A lot of people assume, oh, if my company pays for a good AI, okay, well, I'm just gonna have it do most of my work.
00:12:24.000 --> 00:12:27.840
It's gonna be right because my company's paying for it and they're not training me on it.
00:12:28.000 --> 00:12:30.480
And this is a huge risk.
00:12:30.639 --> 00:12:34.480
You know, JP Morgan, uh, they acknowledge this risk openly.
00:12:34.559 --> 00:12:41.039
They recently talked about that when systems perform correctly, most of the time human attention drifts, right?
00:12:41.200 --> 00:12:50.480
But if AI is right 85 to 95% of the time, well, your human in the loop falls asleep, right?
00:12:50.559 --> 00:12:56.480
Or your human in the loop, maybe if they're a good one, they double down and just produce twice as much.
00:12:57.679 --> 00:13:01.279
But they just wave them wendell, right?
00:13:01.519 --> 00:13:03.519
Chicago reference right there, right?
00:13:03.600 --> 00:13:05.600
They just let everyone go in.
00:13:06.000 --> 00:13:06.720
They don't care.
00:13:06.799 --> 00:13:08.080
They're like, all right, everyone's safe.
00:13:08.159 --> 00:13:09.919
You're in, you're in, you're in, you're in.
00:13:10.080 --> 00:13:15.440
It's a passive approval of just automating uh things that need more oversight.
00:13:17.200 --> 00:13:21.840
And there's a a concept, uh, a concept here that I want to talk about.
00:13:21.919 --> 00:13:24.240
Uh, it's Mdahl's law.
00:13:24.399 --> 00:13:33.759
And that's essentially you can speed up one part of the process or many parts of the process, but the whole system is bottlenecked by whoever can't keep up, right?
00:13:33.840 --> 00:13:41.600
It's the uh the old sports, you know, cliche or the you know teamwork cliche that you're only, you know, strong is your weakest link.
00:13:41.759 --> 00:13:47.360
And this is true for the human AI collaboration, right?
00:13:47.440 --> 00:13:49.679
And and this is why this is so important.
00:13:50.799 --> 00:13:55.919
So few business leaders zoom out and think about this.
00:13:57.120 --> 00:14:01.600
So few people think of the human AI relationship, right?
00:14:02.080 --> 00:14:10.639
So few people working on on front-end AI strategy, back end AI uh implementation go through and ask these questions.
00:14:11.759 --> 00:14:12.240
Right?
00:14:13.039 --> 00:14:20.000
What happens when multi-agentic systems are way smarter than humans and the humans don't know how to run them?
00:14:20.879 --> 00:14:21.440
Right?
00:14:21.840 --> 00:14:30.240
And the human that two or three years ago was integral is now a liability, right?
00:14:30.320 --> 00:14:32.559
And and how do we start to change that human role?
00:14:32.639 --> 00:14:35.200
Because yes, humans are still needed in all this, right?
00:14:35.279 --> 00:14:37.600
I never said that, and I I'm not implying that.
00:14:37.840 --> 00:14:40.960
Human roles are changing, and I'm gonna get to what that looks like.
00:14:41.679 --> 00:14:45.279
But that's why we need human reviews everything.
00:14:46.399 --> 00:14:48.639
That's why we need that, right?
00:14:48.799 --> 00:14:49.679
That's why you have to.
00:14:49.840 --> 00:14:51.679
I'm a huge advocate, right?
00:14:51.840 --> 00:14:55.360
If you're using front-end large language models, what do you do?
00:14:55.600 --> 00:14:58.720
Like people, like people are like, okay, what do you do with all this time savings?
00:14:58.799 --> 00:15:01.360
Or what do you do if you're using a thinking model while you wait?
00:15:01.519 --> 00:15:03.519
Well, you read the chain of thought, right?
00:15:03.600 --> 00:15:13.120
And you rerun it and you correct it and you bring in more of your company's data, right, at the right points before the model goes too deep into its dive.
00:15:15.919 --> 00:15:20.720
And that's why the fix is expert-driven loops.
00:15:20.879 --> 00:15:27.840
That's what I've been advocating for, you know, EDL uh for a long time now, not just generic oversight.
00:15:28.080 --> 00:15:34.720
Because the difference is when you embed experts in building, right?
00:15:34.879 --> 00:15:41.360
Whether it's, you know, we don't have we don't even have to have to talk about, you know, get too technical and talk about, you know, multi-agentic loops.
00:15:41.440 --> 00:15:50.799
Let's just talk about embedding the right experts and setting up your team's AI processes in your ChatGPT Teams account, right?
00:15:51.679 --> 00:15:55.200
Or how you're gonna tackle work and claude co-work.
00:15:55.519 --> 00:16:04.799
A lot of times you have one person, sometimes it's IT, sometimes it's you know, your your AI champion team, which is important.
00:16:05.600 --> 00:16:09.120
And they kind of do it for everyone and they set it up for everyone.
00:16:09.360 --> 00:16:11.759
And that's again the wrong way.
00:16:11.919 --> 00:16:22.639
You know, one thing that was interesting, there was a legal on technology study kind of looking at this concept of the, you know, human in the loop versus an expert loop.
00:16:22.720 --> 00:16:28.799
Um, and they saw that one law firm that they looked at put senior partners in the loop, right?
00:16:28.960 --> 00:16:35.600
Where normally this is uh, you know, not in a bad way, but you know, kind of the human in the loop overseeing AI.
00:16:35.759 --> 00:16:42.240
A lot of times they put younger people or more inexperienced people, uh, people who cost less because that's what companies think, right?
00:16:42.320 --> 00:16:46.000
They're like, okay, well, this is just someone clicking a button, clicking approve.
00:16:46.159 --> 00:16:48.159
Why am I going to put my senior people on this?
00:16:48.240 --> 00:16:50.320
Why am I gonna put my smartest people on this?
00:16:50.559 --> 00:17:06.720
So, in this case study from Legalon Technologies, they found 86% faster contract review and 65% better issue detection when they had senior partners instead of junior reviewers.
00:17:08.160 --> 00:17:12.480
86% faster and 65% better.
00:17:12.720 --> 00:17:16.559
And that's not like versus the human-only baseline.
00:17:16.880 --> 00:17:23.839
That is better than the uh AI, the augmented, right, junior reviewers plus AI.
00:17:24.000 --> 00:17:27.759
So, I mean, you're getting compoundingly better and better results.
00:17:28.000 --> 00:17:34.720
The smarter and the more expert uh people that you put in the right places.
00:17:34.880 --> 00:17:45.279
And I think it's it's it's it's again, it's not putting one expert, you know, on one AI-powered workflow or one, you know, agent run.
00:17:45.599 --> 00:17:49.519
It's putting multiple people in there at the right place, right?
00:17:49.839 --> 00:17:55.359
It's experts driving the loop, not a single human overseeing.
00:17:56.079 --> 00:18:06.880
And other studies show this, other organizations are seeing the ROI triple when they move from generic oversight to expert-driven collaboration.
00:18:08.640 --> 00:18:12.400
So how does this get to this point?
00:18:13.039 --> 00:18:13.680
Right?
00:18:14.400 --> 00:18:23.440
And it's almost like the better the technology gets and the more advanced it gets, and the less tech know-how it requires, right?
00:18:23.519 --> 00:18:30.799
That anyone can go in there and click a button and you know, set up AI agents that are connected to your data, literally, right?
00:18:30.880 --> 00:18:33.680
This show's been going for 19 minutes.
00:18:33.920 --> 00:18:37.599
You could have set the hundreds of agents up in 19 minutes.
00:18:37.680 --> 00:18:38.960
I kid you not, right?
00:18:39.200 --> 00:18:41.839
Um, one click, very easy.
00:18:42.240 --> 00:18:57.200
But poorly implemented AI can crush productivity because you just end up spending more time correcting your errors and managing expectations that maybe went awry, and then running multiple parallel backups.
00:18:57.440 --> 00:19:09.920
So I like to say this if you had a bad workflow with AI and you upskilled or reskilled that workflow, right?
00:19:11.680 --> 00:19:16.799
You you can't just throw makeup on an ugly process and think it's gonna be pretty, it's still ugly.
00:19:17.680 --> 00:19:20.240
Now it's just like, you know, got some makeup on it.
00:19:20.880 --> 00:19:23.200
Got a little shine that it doesn't deserve.
00:19:24.160 --> 00:19:29.200
All this does is it recreates workflows that weren't working.
00:19:29.440 --> 00:19:50.640
And if anything, it just creates this augmentation debt that ultimately tanks productivity because now you just have to go back, you know, and instead of getting more things done in a better way, now you're just getting more things that need fixed faster with potentially more airs because you're not putting the right people in the right processes, right?
00:19:50.799 --> 00:19:59.119
You're putting anyone in old processes, you have to rebuild that to be AI native.
00:20:00.960 --> 00:20:03.440
So, how do you do this?
00:20:06.480 --> 00:20:08.400
It's a mindset shift.
00:20:09.039 --> 00:20:19.839
Like I said, you're not the one if your team wants to excel and outrun the competition in 2026 and beyond.
00:20:22.400 --> 00:20:26.880
You can't just use AI, you can't leverage AI.
00:20:28.079 --> 00:20:29.839
You have to orchestrate it.
00:20:30.240 --> 00:20:30.720
Right.
00:20:30.880 --> 00:20:36.319
What's funny is most most times, this is one of the times I you know, I'm looking at my other screen here.
00:20:36.480 --> 00:20:42.559
Um, you know, I actually don't have agents running right now, uh, where normally I would, right?
00:20:42.880 --> 00:20:52.240
For the most part, we have to start thinking of ourselves as orchestrators, or I like to say tastemakers sometimes, right?
00:20:52.720 --> 00:20:55.599
You need to provide that context, right?
00:20:55.759 --> 00:21:03.519
Context is gonna be one of those, you know, context engineering uh is gonna be one of those buzzwords of 2026, right?
00:21:03.680 --> 00:21:07.839
In 2024, I called it uh first company data, right?
00:21:08.000 --> 00:21:12.160
Um, but I still think that that's more realistic in what you need.
00:21:12.319 --> 00:21:13.599
Context can be anything, right?
00:21:13.759 --> 00:21:15.680
Context needs more context to be defined.
00:21:15.839 --> 00:21:25.839
But you know, context engineering, uh to put simply, you know, when I said, hey, in 2023, you were prompting and then you were iterating, and then you were providing more context, right?
00:21:26.000 --> 00:21:32.000
That's just more data, more data, more direction to a model before it goes off and do its uh does its thing.
00:21:32.240 --> 00:21:41.200
Because, you know, before when the models were just, you know, non-reasoning models, non-thinking models, they would just spit something back pretty quickly, right?
00:21:41.359 --> 00:21:44.720
Now they might go work for five minutes, right?
00:21:44.880 --> 00:21:47.039
Uh, 10 minutes longer.
00:21:47.680 --> 00:21:51.759
Uh, I had one run the other day on the front end, not on the back end, right?
00:21:51.920 --> 00:21:56.720
On the back end, if you're using it via the API, it's pretty easy to get these things working for hours.
00:21:56.799 --> 00:21:59.759
On the front end, you know, yeah, you might have a model work for five, 10, 15.
00:22:00.160 --> 00:22:00.960
20 minutes.
00:22:01.039 --> 00:22:08.640
Uh, I had an actual model, not a deep research run, uh, run for I think 92 minutes uh the other day.
00:22:08.880 --> 00:22:09.359
Right.
00:22:09.519 --> 00:22:12.079
You have to give it the context.
00:22:12.240 --> 00:22:15.680
And then at that point, you're orchestrating, right?
00:22:16.000 --> 00:22:21.599
Um I'm I'm looking at different models, what they bring to me, different agents, what they bring to me.
00:22:21.680 --> 00:22:23.920
And I'm saying this is good, this is good, this isn't.
00:22:24.079 --> 00:22:25.599
Let's rebuild that skill, right?
00:22:25.680 --> 00:22:29.599
If I'm in Claude, I'm sitting, let's I'm saying let's update this skill, right?
00:22:29.680 --> 00:22:33.519
If I'm in uh Chad GPT, I'm going and updating that GPT, right?
00:22:33.599 --> 00:22:41.920
You always have to be improving the processes and not just trying to sprinkle some AI on an old process that's broken.
00:22:42.079 --> 00:22:46.400
And that's why we're talking about these things like Asian supervisors, op orchestrators.
00:22:46.640 --> 00:22:50.480
This is a fundamental shift in how work is getting done.
00:22:50.640 --> 00:22:59.200
This is the difference in the human AI collaboration and why I think you need to rethink your working relationship with AI.
00:22:59.599 --> 00:23:07.039
Because if you're still using it like you were in, you know, late 2022 or 2023, right?
00:23:07.200 --> 00:23:13.839
If you're going out of your way, we tackled this earlier in our start here series about treating AI like an operating system.
00:23:15.839 --> 00:23:19.920
So one, that part covers the tech, right?
00:23:20.240 --> 00:23:22.480
Treating AI like an operating system.
00:23:23.039 --> 00:23:28.960
This part, volume four here, while we're tackling the human AI collaboration, that's the mindset shift.
00:23:29.039 --> 00:23:38.799
That's going from an operator, I'm the one pushing the buttons, to nope, I'm orchestrating an agent that's technically going out there and pushing the buttons and coming back.
00:23:38.880 --> 00:23:41.519
And then I'm telling it how to improve, right?
00:23:41.759 --> 00:23:44.640
I am building that expertise, even if you don't know.
00:23:44.799 --> 00:23:54.079
Hey, what is that one thing that you know you're definitely smarter than these AI models that are genius level on offline IQ tests, right?
00:23:54.240 --> 00:24:00.400
You might be saying, okay, what could I be smarter than an AI model that knows everything if you give it the right context?
00:24:00.559 --> 00:24:03.440
Well, you don't know until you look at its chain of thought.
00:24:03.599 --> 00:24:13.680
You don't know until you know you've put in, you know, five, 10, 15, 30 hours on a project, having multiple AI models go and do something that you know how to do front to back.
00:24:13.839 --> 00:24:16.319
That's how you carve out your expertise.
00:24:16.480 --> 00:24:22.880
And then you really have to then deploy that and duplicate it across your team or your organization.
00:24:23.279 --> 00:24:34.000
That is how you go about shifting your mindset to go from an operator to an orchestrator, because then that output compounds, right?
00:24:34.240 --> 00:24:37.279
Your ability to generate revenue compounds.
00:24:37.519 --> 00:24:44.559
Your ability to do things that you didn't have time to do last year, all of a sudden that frees up, right?
00:24:44.640 --> 00:24:49.359
And this is where the most advanced businesses are shifting toward right now.
00:24:50.240 --> 00:25:03.279
So the last thing I want to talk about last two things is where our expertise actually belongs.
00:25:04.000 --> 00:25:04.319
Right?
00:25:04.640 --> 00:25:06.319
Where should we be spending our time?
00:25:06.799 --> 00:25:14.319
Well, you have to go do what I just said, you have to make that transition successfully from operator to orchestrator.
00:25:14.480 --> 00:25:18.160
But then I want you to think about kind of this jagged frontier of capabilities, right?
00:25:18.240 --> 00:25:26.400
Because there's things that humans are good at, and then there's things that we're terrible at.
00:25:26.480 --> 00:25:29.039
And those things are changing all the time, right?
00:25:29.279 --> 00:25:30.799
And the same thing with AI models.
00:25:30.880 --> 00:25:34.160
There's things that AI models are amazing at that we never thought would be possible.
00:25:34.319 --> 00:25:36.160
Then there's things that they kind of fail at.
00:25:36.319 --> 00:25:38.480
So right now, this is today, right?
00:25:38.559 --> 00:25:43.039
So if you're listening to this in January, February 2026, probably still safe.
00:25:43.119 --> 00:25:47.519
If you're listening to it in, you know, July, September, this could be different.
00:25:47.920 --> 00:25:58.960
But right now, humans win at high context empathy, ambiguous decisions with incomplete data, accountability, and novel judgment, right?
00:25:59.200 --> 00:26:05.759
Reading between the lines where there's not structured data or there's not company context to fill those cracks.
00:26:05.920 --> 00:26:09.200
Right now, AI wins at pretty much everything else, right?
00:26:09.440 --> 00:26:13.680
Data synthesis, first drafts, pattern recognition, repetitive cognition, right?
00:26:13.759 --> 00:26:15.119
All of those things.
00:26:15.920 --> 00:26:24.640
Those are the danger zones for if those are your competitive motes, or you think that's your competitive advantage right now, right?
00:26:24.720 --> 00:26:38.559
If you think uh one of those things, data synthesis, first drafts, pattern recognition, if that's what your department, your career, your company is built on right now, you you you've got to find your pivot.
00:26:40.960 --> 00:26:46.000
Because right now, the number one predictor of human AI success isn't just the model you use.
00:26:46.400 --> 00:26:53.200
It's the quality of the context and kind of those procedures that you create and you provide.
00:26:53.440 --> 00:26:59.599
Because if garbage context goes in, a worse outcome, a worse outcome comes out.
00:27:01.119 --> 00:27:03.839
So here's some quick takeaway advice.
00:27:04.640 --> 00:27:05.119
Right?
00:27:05.519 --> 00:27:11.519
I can't sit here and give you uh advice for how to set up your agent orchestration.
00:27:11.599 --> 00:27:17.039
What I can tell you, your individual using different platforms, right?
00:27:17.519 --> 00:27:22.960
One of the biggest mistakes is skipping over repeatable and scalable context.
00:27:23.279 --> 00:27:26.400
So stop starting from zero, right?
00:27:26.880 --> 00:27:31.839
And you need to take this, start this at the individual level, but then take this to your team.
00:27:32.079 --> 00:27:34.240
Stop starting at zero every single prompt.
00:27:34.319 --> 00:27:37.359
You have to start building uh context faults.
00:27:38.000 --> 00:27:38.400
Right.
00:27:38.559 --> 00:27:40.319
You can think of those as skills, right?
00:27:40.400 --> 00:27:44.559
You can have a markdown file with all your skills with your company knowledge, right?
00:27:44.720 --> 00:27:54.319
If you use uh Claude skills, you know, you're probably familiar with these markdown files, but start creating these skill files based on the tasks that you and your team repeatedly do.
00:27:54.480 --> 00:27:56.960
This is something I'm constantly updating mine, right?
00:27:57.039 --> 00:28:02.960
If you're using, as an example, Claude, you can update them in um kind of in chat.
00:28:03.119 --> 00:28:07.200
If you're using uh the GPT builder in ChatGPT, you can do it there as well.
00:28:07.359 --> 00:28:07.839
Uh right.
00:28:07.920 --> 00:28:17.599
But you need to be building and reusing these uh, you know, custom GPTs inside of uh inside of ChatGPT, the Claude projects and skills, Google Gems, whatever.
00:28:18.000 --> 00:28:21.759
You need to have that rag for your personal use, right?
00:28:21.839 --> 00:28:31.839
We think of when we think about retrieval augmented generation, we usually think about these you know vector databases that would cost millions of dollars three, four, five years ago to build.
00:28:32.000 --> 00:28:33.599
And we think about these very complex things.
00:28:33.759 --> 00:28:38.480
I want you to think about your personal uh rag, your personal retrieval augmented generation.
00:28:38.640 --> 00:28:39.359
What are those things?
00:28:39.440 --> 00:28:41.680
I update mine all the time, right?
00:28:41.920 --> 00:28:46.319
I have my, you know, as a small business owner, these are the things that I'm thinking about.
00:28:46.480 --> 00:28:51.920
These are uh, you know, my important facts, stats about my role, about what I'm trying to drive.
00:28:52.079 --> 00:28:55.440
Here's my KPIs, and I'm constantly, you know, updating these things.
00:28:55.519 --> 00:29:02.480
But you need to have your personal context, your team context, uh, your company context, your competitive landscape context.
00:29:02.640 --> 00:29:08.400
You have to have these things that are reusable because if you're just starting at zero, you're wasting time.
00:29:09.039 --> 00:29:11.680
That's being a button pusher, right?
00:29:12.000 --> 00:29:17.039
Instead of agents that don't need to have the button pushed.
00:29:17.119 --> 00:29:19.119
They're already doing it, they're already delivering it.
00:29:19.279 --> 00:29:23.359
They have all that and they're able to use it in a repeatable and scalable way.
00:29:24.240 --> 00:29:30.079
And then last but not least, you need to elevate your champions.
00:29:30.480 --> 00:29:32.240
Okay, here's what I mean by that.
00:29:32.480 --> 00:29:40.720
Uh uh funny, I had uh Chris Caldwell, the CEO from Concentrics on uh a couple of weeks ago.
00:29:40.880 --> 00:29:44.240
And he kind of said, Hey, you don't want 100 Jordans running around, do you?
00:29:44.400 --> 00:29:45.200
And you don't.
00:29:45.440 --> 00:29:47.119
But I'll tell you this.
00:29:49.519 --> 00:29:56.880
And I don't want this to come off in the wrong way, but you need people like me on your team, and you need a lot of them.
00:29:57.279 --> 00:30:05.200
Here's what I need: you need people whose mainly their only job is to keep up with AI every single day, right?
00:30:05.680 --> 00:30:11.200
Every large organization needs dozens of people who are in my position.
00:30:11.440 --> 00:30:14.720
All they do, they they read about AI every day.
00:30:14.799 --> 00:30:20.400
Uh, they're scoping out new projects, uh, measuring, building modular backups, right?
00:30:20.480 --> 00:30:26.960
Because you don't want all of your uh you know systems in one model, the model changes and your whole company comes to comes to a halt.
00:30:27.039 --> 00:30:31.279
Uh, and then you need to constantly have those people training the non-champions.
00:30:31.519 --> 00:30:35.039
So you need to find where your time savings opportunities are.
00:30:35.200 --> 00:30:41.440
You need to scope those, measure those, and then deploy the ones that can easily gain back time the quickest.
00:30:41.599 --> 00:30:49.359
And then you apply those hour saves into creating those, you know, go create your dozen Jordans uh on your team, right?
00:30:49.519 --> 00:30:51.839
Your domain experts, right?
00:30:52.079 --> 00:31:04.799
It might feel weird to start automating some of their work or automating some of their roles, but then you find those people that are actually able to automate parts of their job and then teach others, build scalable systems.
00:31:04.960 --> 00:31:09.519
There's still new lines of revenue uh to build in your company, right?
00:31:09.680 --> 00:31:28.559
Uh, people, when they look at AI, yes, I do ultimately think AI will um cause a net negative in the job market, but there's millions, tens of millions of new jobs that we have no clue that are gonna exist in three years.
00:31:28.880 --> 00:31:36.799
And you need those people, those champions on your team, you need to elevate them, challenge them, and you need to deploy them, right?
00:31:36.960 --> 00:31:50.160
They need to be listening to this show every day and other, you know, AI podcasts and other, you know, YouTube people on AI reading newsletters, and you need to be scoping, breaking, and training people every single day.
00:31:50.319 --> 00:31:53.519
But the largest organizations aren't even doing that.
00:31:53.920 --> 00:31:56.880
So stop looking for cool AI tricks.
00:31:57.119 --> 00:32:00.240
Focus on automating the dull stuff first, right?
00:32:00.559 --> 00:32:03.279
Get rid of shiny AI object syndrome.
00:32:03.519 --> 00:32:11.920
Go, the invoices, the summarization, the filing, all those boring AI things beat the flashy AI every single time.
00:32:12.079 --> 00:32:24.720
Because as AI starts to handle more and more digital interaction, face-to-face and high empathy relationships are going to become your company's differentiators, right?
00:32:24.880 --> 00:32:28.400
But you can't have that if you're still the one pushing buttons.
00:32:28.640 --> 00:32:29.359
You can't.
00:32:29.599 --> 00:32:32.240
The human premium is rising.
00:32:32.640 --> 00:32:34.240
So you have to use it wisely.
00:32:34.319 --> 00:32:58.720
But you can only do that if you go back and follow the steps that we just laid out in re-examining the human AI collaboration, going through the best practices of shifting away from being the operator, from being the button, the button pusher, the chat GPT prompter into being the orchestrator, the tastemaker, and the champion that's pushing your organization to do the same top to bottom.
00:32:58.880 --> 00:32:59.119
All right.
00:32:59.200 --> 00:33:02.400
I hope this Start Here series was helpful.
00:33:02.559 --> 00:33:04.720
Volume four done in the books.
00:33:04.799 --> 00:33:09.359
Uh, so if this was helpful, remember please go to startheseries.com.
00:33:09.519 --> 00:33:14.160
That's gonna give you access for free to our inner circle community.
00:33:14.319 --> 00:33:18.000
And you're gonna see all of the start here uh episodes right there.
00:33:18.079 --> 00:33:22.640
We're gonna be throwing in and adding more and more additional resources to help you with your journey.
00:33:22.720 --> 00:33:27.359
So whether you're just starting out, I hope this episode helped you better understand some things.
00:33:27.519 --> 00:33:35.599
If you're an expert doing this every day, I hope this challenged you to look at AI a little bit differently and to push even your own human AI collaboration.
00:33:35.759 --> 00:33:37.200
So, thank you for tuning in.
00:33:37.279 --> 00:33:40.799
I hope to see you back tomorrow and every day for more everyday AI.
00:33:40.960 --> 00:33:42.079
Thanks, y'all.
