WEBVTT
00:00:01.919 --> 00:00:07.440
There's been a lot of talk lately about these AI models that are scary good.
00:00:07.679 --> 00:00:20.399
And even more chatter about how many of the brightest minds in AI and all the tech AI CEOs feel that we've already reached artificial general intelligence and are racing toward superintelligence.
00:00:20.640 --> 00:00:23.679
But what does that matter to you or your company?
00:00:23.920 --> 00:00:27.920
Because right now, those are kind of just theoretical issues.
00:00:28.079 --> 00:00:33.439
It's kind of like when you see an email on a Friday and you look at it and you're like, yeah, that's a Monday problem.
00:00:33.679 --> 00:00:40.799
But do you know what's scarier than some private AI model that's scary good or artificial superintelligence?
00:00:41.039 --> 00:00:42.719
The AI capability gap.
00:00:42.799 --> 00:00:44.320
That's the scariest of all.
00:00:44.640 --> 00:00:45.280
Why?
00:00:45.679 --> 00:00:47.600
Because it's very real.
00:00:47.920 --> 00:00:56.399
And unlike that Friday afternoon email that you're maybe going to tackle on Monday, the AI capability gap is a today problem.
00:00:56.640 --> 00:01:02.240
And if you don't address it head on, it's going to slowly stall your company's growth.
00:01:02.320 --> 00:01:10.319
And if your company isn't growing, the AI capability gap will undoubtedly stop your company dead in its tracks.
00:01:10.480 --> 00:01:12.560
And this is not an exaggeration.
00:01:12.719 --> 00:01:15.920
And the timing here is specifically urgent.
00:01:16.079 --> 00:01:26.000
So we're going to hit rewind today and explain the capability gap in AI and why AI is racing ahead of your company.
00:01:26.159 --> 00:01:30.480
And we're going to show you how to slow it down a bit and catch it.
00:01:30.719 --> 00:01:32.000
All right, you ready?
00:01:32.159 --> 00:01:36.560
This is part of our start here series on everyday AI.
00:01:36.799 --> 00:01:38.400
Let's get into it.
00:01:38.640 --> 00:01:43.120
So if you are new here, well, let's just get to the big picture.
00:01:43.519 --> 00:01:47.840
AI capability right now is outpacing business adoption.
00:01:47.920 --> 00:01:50.319
And it has changed recently.
00:01:50.400 --> 00:01:53.680
And the pace is too fast for any of us.
00:01:53.840 --> 00:01:55.439
Right now, Frontier AI.
00:01:55.519 --> 00:01:59.280
And when we talk about actual capabilities, this is not an exaggeration.
00:01:59.439 --> 00:02:08.639
Frontier AI models, when used correctly, and that's the big asterisk here, they now match or exceed human professionals on most defined knowledge work tasks.
00:02:08.800 --> 00:02:19.840
And there's a massive gap that persists between what AI could, in theory, automate today and what organizations are actually using it for, which is often just topical.
00:02:19.919 --> 00:02:20.319
Uh, right.
00:02:20.400 --> 00:02:28.879
There's a McKinsey study that said only about 6% of organizations are generating meaningful profits despite widespread AI adoption.
00:02:29.039 --> 00:02:32.960
And the bottleneck is not actually the AI models.
00:02:33.199 --> 00:02:40.400
You could have made that argument maybe mid-2025, but not anymore in 2026.
00:02:40.800 --> 00:02:47.039
It's actually about organizational adoption, workflow design, and training and education.
00:02:47.199 --> 00:02:49.360
So here's what we're going to tackle on today's show.
00:02:49.439 --> 00:02:52.479
So stick with me for the next, I'm going to try to make this one 25 minutes.
00:02:52.560 --> 00:02:52.960
We'll see.
00:02:53.120 --> 00:02:56.319
All right, stick with me for the next 22-ish minutes.
00:02:56.400 --> 00:03:02.000
And you're going to learn the benchmark proving that AI outperforms human experts, and it's not even close.
00:03:02.159 --> 00:03:09.680
You're going to know more about Anthropic's new-ish study, revealing most knowledge workers barely touch AI's true automation potential.
00:03:09.840 --> 00:03:14.800
You're going to learn why the AI capability gap only started to show itself over the past few months.
00:03:14.879 --> 00:03:23.520
It's actually kind of new, and what the top 6% of AI performing companies actually do that everyone else completely ignores.
00:03:23.759 --> 00:03:25.759
All right, welcome to Everyday AI.
00:03:25.840 --> 00:03:27.759
And this is our Start Here series.
00:03:27.919 --> 00:03:33.840
This is the essential podcast series to both learn the AI basics and for AI experts to double down.
00:03:34.000 --> 00:03:34.400
All right.
00:03:34.479 --> 00:03:40.719
I started this thing because after 750 plus episodes, everyone always said, Jordan, great podcast, right?
00:03:40.800 --> 00:03:42.400
When they found it, they're like, where do I start?
00:03:42.560 --> 00:03:43.039
I have no clue.
00:03:43.199 --> 00:03:44.800
And I'm like, I also have no clue.
00:03:44.960 --> 00:03:46.879
So that's why I started the Start Here series.
00:03:46.960 --> 00:03:50.400
Uh, it's best if you are brand new here, start in order.
00:03:50.479 --> 00:03:50.800
All right.
00:03:50.879 --> 00:03:55.120
These are faster uh, you know, podcasts, usually about 25 to 35 minutes.
00:03:55.280 --> 00:03:57.759
But if you listen on 2x, they're even half that, right?
00:03:57.840 --> 00:04:02.240
Uh, but this is a great way to listen from one to now, volume 19.
00:04:02.319 --> 00:04:04.960
And then also make sure to go to startherseries.com.
00:04:05.039 --> 00:04:08.800
That's going to give you free access to our exclusive inner circle community.
00:04:08.960 --> 00:04:31.199
You can't find access right now any gap report card.
00:04:31.439 --> 00:04:35.040
All right, so uh a little bit more on that at the end.
00:04:37.279 --> 00:04:40.480
But trust me, you are going to want to repost today's episode.
00:04:40.639 --> 00:04:41.360
I'm just saying.
00:04:41.600 --> 00:04:55.600
All right, and if you missed our last start here series in volume 18, and this was episode 750, we went over the vibe coding boom, why vibe coding isn't going away, and how it's both good and bad.
00:04:55.680 --> 00:04:57.199
So make sure to go check that one out.
00:04:57.279 --> 00:05:05.600
But now let's talk about how you can actually manage the AI capability gap and why AI is more than ready and companies are not.
00:05:05.759 --> 00:05:08.480
All right, and I'm not the only one talking about this.
00:05:08.639 --> 00:05:17.439
I've been talking about this now since I think late 2025, but a lot of super smart people are starting to dive in deep on the AI capability gap.
00:05:17.600 --> 00:05:19.519
I really like what Jack Clark said.
00:05:19.759 --> 00:05:24.480
He is uh the anthropic um co-founder of anthropic.
00:05:24.560 --> 00:05:27.839
And here's what he said in a post a couple of weeks ago.
00:05:28.000 --> 00:05:30.720
He said, most of AI progress has this flavor.
00:05:30.879 --> 00:05:39.920
If you have a bit of intellectual curiosity and some time, you can very quickly shock yourself with how amazingly capable modern AI systems are.
00:05:40.160 --> 00:05:44.319
But you need to have that magic combination of time and curiosity.
00:05:44.480 --> 00:05:59.120
And otherwise, you're going to consume AI like most people do, as a passive viewer of some unremarkable synthetic slop content, or at best, just asking your LLM of choice how to roast a turkey and keep it moist.
00:05:59.360 --> 00:06:03.279
Or Tony Box lights spinning, but not playing music.
00:06:03.519 --> 00:06:04.720
What do I do?
00:06:04.879 --> 00:06:08.240
And all the amazing advancements are mostly hidden from you.
00:06:08.319 --> 00:06:14.959
All right, so that's what Jack Clark said in a uh kind of a viral uh post that he had a couple of weeks ago um on X.
00:06:15.839 --> 00:06:18.480
And I think this is very telling.
00:06:18.959 --> 00:06:22.319
And the two things that he talked about, I think are very true.
00:06:22.480 --> 00:06:30.160
All right, to start, well, first to even realize the AI capability gap, let alone close it, you have to be extremely curious.
00:06:30.399 --> 00:06:48.240
And the reason why I say that is because if you work how you've been working for the last few decades, you're not going to discover, or your or your team, your organization is not truly going to discover that capability gap because AI is not meant for the average knowledge worker, right?
00:06:48.399 --> 00:06:59.040
That's why I talk about all the time how I hate, absolutely hate the concept of upskilling, because AI is not something that you sprinkle on top of a seasoned knowledge worker, right?
00:06:59.120 --> 00:07:03.839
Like myself, I've been working full time for 20-ish years, right?
00:07:04.160 --> 00:07:05.920
I can't just sprinkle AI on the top.
00:07:06.000 --> 00:07:07.600
I have to, I have to unlearn, right?
00:07:07.680 --> 00:07:13.920
So you do have to have the time and you have to be curious on, hey, what does my role look like if I completely start it over?
00:07:14.000 --> 00:07:17.360
And if I kind of forgot everything that got me to the point I am in my career.
00:07:17.439 --> 00:07:21.360
And that's what you have to do, and you also have to have a lot of time and you have to devote a lot of time to it.
00:07:21.519 --> 00:07:29.279
Uh, Olivia Moore from A16Z uh talked about, she said, uh, OpenAI dropped a state of enterprise reports across a million customers.
00:07:29.439 --> 00:07:36.639
This was a couple of months ago, but her response, she said, the gulf between AI power users and everyone else is wide.
00:07:36.879 --> 00:07:45.839
The 95th percentile user spends six sends six times more messages than the median with coding, writing, and analysis showing the biggest gaps.
00:07:46.000 --> 00:07:52.399
So, yeah, those people that are power users, well, because they actually understand the capability gap, they're using AI all the time.
00:07:52.560 --> 00:08:00.160
This is why myself, and I'm not saying this is like some weird flex, I'm saying this because I'm part of this uh, you know, power user group.
00:08:00.319 --> 00:08:06.399
When I say that I use AI for 10 to 12 hours a day, right, that's how much I'm working.
00:08:06.480 --> 00:08:08.959
Uh, you know, it's not like I'm working 16 hours a day.
00:08:09.199 --> 00:08:19.920
I am only working in AI every single day, no matter where I am, I'm using AI from beginning to end in literally every single step in between.
00:08:20.079 --> 00:08:22.319
And it has completely reshaped my workflow.
00:08:22.480 --> 00:08:27.600
Uh uh Kevin Roos, uh New York Times columnist, said this, and I really like how he put this.
00:08:27.759 --> 00:08:34.399
He said, I follow AI adoption pretty closely, and I have never seen such a yawning inside-outside gap.
00:08:34.559 --> 00:08:48.320
People in SF San Francisco are putting multi-agent cloud storms in charge of their lives, consulting chatbots before every decision, wire heading to a degree only sci-fi writers dared to imagine.
00:08:48.559 --> 00:08:55.039
People everywhere are still trying to get approval to use co-pilot in Teams if they're using AI at all.
00:08:55.200 --> 00:09:06.320
It's possible that the early adopter bubble I'm in has always been this intense, but there seems to be a cultural kick takeoff happening in addition to the technical one, not ideal.
00:09:06.639 --> 00:09:16.559
And I'll say from my experience, I think there's five main reasons that the AI capability gap has really been exposed in 2026.
00:09:16.720 --> 00:09:21.120
And number one, the models of what they can actually do, right?
00:09:21.279 --> 00:09:23.200
I'm just gonna roll through these things really quickly.
00:09:23.360 --> 00:09:25.759
So, number one, what AI models can actually do.
00:09:25.919 --> 00:09:29.360
Number two, what business leaders think AI can do.
00:09:29.679 --> 00:09:31.840
Number three, AI literacy.
00:09:31.919 --> 00:09:33.840
Uh, number four, human skill set.
00:09:33.919 --> 00:09:35.279
And number five, AI access.
00:09:35.360 --> 00:09:38.240
Now, let me break those five things down a little bit more in depth.
00:09:38.399 --> 00:09:41.120
So, number one, what AI models can actually do.
00:09:41.360 --> 00:09:42.639
A year ago, right?
00:09:42.720 --> 00:09:48.399
So I don't know when you're listening to this Start Here series episode, you know, could be in April 2026.
00:09:48.480 --> 00:09:50.480
You might be listening to it at the end of 2026.
00:09:50.559 --> 00:09:51.039
I don't know.
00:09:51.120 --> 00:09:51.679
Uh, right.
00:09:51.759 --> 00:09:59.679
But a year ago, in the beginning of 2025, um, you could, for the most part, understand what AI models are capable of.
00:09:59.919 --> 00:10:02.480
Today, you absolutely can't.
00:10:02.639 --> 00:10:06.000
And it is literally getting to the point of science fiction, right?
00:10:06.080 --> 00:10:11.440
All of these math problems and science problems that have been plaguing researchers for decades are now being solved.
00:10:11.600 --> 00:10:13.039
I look at them and I have no clue, right?
00:10:13.120 --> 00:10:15.840
I read the papers and I'm like, okay, I don't have any clue.
00:10:16.000 --> 00:10:18.080
I can follow and understand it a year ago.
00:10:18.240 --> 00:10:23.440
What AI models can actually do today is mind-boggling.
00:10:23.600 --> 00:10:26.639
Number two, what business leaders think AI can do.
00:10:26.799 --> 00:10:29.759
Um, I get to talk to a lot of smart people.
00:10:31.759 --> 00:10:33.440
It's really right.
00:10:33.519 --> 00:10:40.080
I've I've had hundreds of guests on this show, but I also maybe a handful that I've talked to.
00:10:40.320 --> 00:10:45.759
And I'm like, yes, this business leader fully understands what AI can do for the most part.
00:10:45.840 --> 00:10:53.519
And that's not their fault necessarily, because most people in their role, they have to be an expert at one application, right?
00:10:53.679 --> 00:10:58.080
You can't be an like very few business leaders know what AI can do.
00:10:58.240 --> 00:11:03.519
I feel fairly confident that I'm in that category, but only because this is all I do.
00:11:03.679 --> 00:11:08.159
If I had another job, right, and I was just an AI champion at a company, you can't.
00:11:08.320 --> 00:11:09.759
Like you literally can't.
00:11:09.919 --> 00:11:12.480
You used to be able to, you can't anymore.
00:11:12.639 --> 00:11:14.399
Because literally every single day, right?
00:11:14.480 --> 00:11:21.600
Whether you're talking about uh Google, Microsoft, Open AI, Anthropic, now Meta's back in the conversation, perplexity, et cetera, right?
00:11:21.759 --> 00:11:28.960
You literally can't have an actual job where you have to uh produce something for a company and still understand what AI can do.
00:11:29.039 --> 00:11:29.519
Not anymore.
00:11:29.600 --> 00:11:30.879
You used to be able to juggle that.
00:11:30.960 --> 00:11:34.320
Now, unless you're literally working 20 plus hours a day, you can't do it.
00:11:34.480 --> 00:11:36.559
All right, number three, AI literacy.
00:11:36.879 --> 00:11:43.279
And this kind of goes uh, you know, hand in hand with uh, you know, number two, what business leaders understanding.
00:11:43.360 --> 00:11:56.080
So there's a difference between um understanding AI's capabilities and then AI literacy, because you have to actually be able to speak the language foundationally, which again, most people skip over that, right?
00:11:56.159 --> 00:12:03.519
They go straight to the bells and the whistles, they go straight to clicking the button, not knowing what happens, you know, from A to X.
00:12:03.759 --> 00:12:07.279
They only want to get to Y in the result that brings on Z.
00:12:07.600 --> 00:12:09.759
Uh number four is human skill set.
00:12:10.080 --> 00:12:12.080
All right, this is different than literacy.
00:12:12.240 --> 00:12:16.720
You have to have the foundational understanding, but then you also have to have these skills, right?
00:12:16.799 --> 00:12:21.200
I know a lot about basketball, but my basketball skill set is not great anymore, right?
00:12:21.360 --> 00:12:23.200
I think I peaked in eighth grade.
00:12:23.279 --> 00:12:26.080
Uh, but you have to have that human skill set.
00:12:26.159 --> 00:12:28.320
Uh, and then last but not least is AI access.
00:12:28.399 --> 00:12:33.039
So that's access to the tools, to the technology in your role.
00:12:33.279 --> 00:12:37.360
Okay, so what can AI models actually do?
00:12:37.600 --> 00:12:37.919
Right.
00:12:38.080 --> 00:12:52.240
And I do want to spend a little bit more time on that because that will better explain this gap that I think started manageable and now it's very hard to manage that capability gap because it has turned into a friggin' canyon.
00:12:52.480 --> 00:12:58.559
So the reason being is because AI models are better than expert humans.
00:12:58.799 --> 00:13:09.440
All right, I've talked about this uh this evaluation a couple of times, but the more I follow these AI benchmarks and all of these other things, uh, there's there's dozens of them.
00:13:09.600 --> 00:13:11.600
I think there's about five that matter, right?
00:13:11.679 --> 00:13:15.600
And one of them, and I think probably the most important one, is GDP Val.
00:13:15.919 --> 00:13:17.840
So this is a benchmark from OpenAI.
00:13:17.919 --> 00:13:24.000
And the reason why I think it's the most valuable benchmark to look at is because it's about creating business value.
00:13:24.159 --> 00:13:34.080
So, more or less, uh, this benchmark evaluates AI models against real professional deliverables across 44 different high GDP occupations.
00:13:34.159 --> 00:13:47.120
So these are things that you and I may do, you know, going in and doing research on a fast-moving market segment and then creating an artifact like a, you know, a pitch deck or creating a spreadsheet, something like that.
00:13:47.279 --> 00:14:04.080
These are real-world domain-specific tasks where both a human and an AI model from start to finish uh completed a task and then submitted their final version to a panel of experts who are experts in that field.
00:14:04.240 --> 00:14:04.559
All right.
00:14:04.639 --> 00:14:12.080
And expert judges then compared the unlabeled AI and human outputs in a blind head-to-head pairwise comparison.
00:14:12.240 --> 00:14:23.840
And what we've seen now is, as an example, the best model for actually creating front-to-back economically valuable outputs like a human would is the OpenAI's GPT-5.
00:14:24.720 --> 00:14:30.799
And it matches or exceeds industry professionals in 83% of these evaluations.
00:14:31.279 --> 00:14:31.919
Right.
00:14:33.360 --> 00:14:38.559
I remember when some of these first GDP valve numbers came out, and I'm like, oh, that's that's pretty impressive, right?
00:14:38.639 --> 00:14:42.320
When we were in the 30, 40%, but now it's it's undeniable.
00:14:42.399 --> 00:14:49.200
And I do assume that by the end of the year, that number is gonna be like in the mid-90s, right?
00:14:49.840 --> 00:14:54.320
My like one of my predictions is it was gonna get to 80%, and we're already at 80%.
00:14:54.639 --> 00:14:56.080
It was just a huge jump.
00:14:56.159 --> 00:15:03.200
Uh, that's why I I keep talking about the AI capabilities are just running too fast for most business leaders to understand and keep up.
00:15:03.360 --> 00:15:04.480
And it's nearly doubled.
00:15:04.559 --> 00:15:07.759
That GDP bow score has nearly doubled in five months.
00:15:07.919 --> 00:15:26.080
So when I talk about the last, you know, since essentially late 2025, uh, I'm gonna talk about why, but it has been, we've seen more developments in the past five months than we have in the five years prior, at least when it comes to large language models, and it's not even close.
00:15:26.320 --> 00:15:27.039
Um, right.
00:15:27.200 --> 00:15:32.960
So the in October, the best AI model scored on that was a 47%.
00:15:33.519 --> 00:15:39.840
So it hasn't quite doubled, but it's nearly doubled in about five or six months.
00:15:40.080 --> 00:15:44.559
And right now, AI handles most defined professional tasks yet.
00:15:44.799 --> 00:15:48.159
Most organizations barely use models in this way.
00:15:48.240 --> 00:15:49.840
It's like, did you know, right?
00:15:49.919 --> 00:15:56.159
As an example, your uh, you know, GPT and Claude models can do this.
00:15:56.320 --> 00:15:58.720
They can create artifacts by default.
00:15:58.799 --> 00:16:00.559
You don't have to have anything special turned on.
00:16:00.720 --> 00:16:07.679
It can just create spreadsheets, presentations, word docs, etc., all personalized based on your company's info.
00:16:07.919 --> 00:16:15.440
All right, next, I want to talk very briefly about the anthropic labor data study that came out uh about two months ago.
00:16:15.679 --> 00:16:17.679
I did go over this in more depth.
00:16:17.759 --> 00:16:21.440
So if you're interested in this, go listen to episode 730.
00:16:21.840 --> 00:16:36.480
Here's essentially uh similarly, uh, whereas the GDP benchmark looks at actual economic output, uh, this anthropic labor data study looks a little bit more at the capability gap.
00:16:36.639 --> 00:16:50.879
So that's why I think these two uh kind of this benchmark, OpenAI benchmark, and the anthropic labor daddy labor data study in tandem really tell a powerful story of where we're at and why this capability gap even exists.
00:16:51.039 --> 00:16:56.240
So in this study, Anthropic mapped over 20,000 work tasks, right?
00:16:56.320 --> 00:17:02.159
So they actually used uh some public uh some public jobs data from the federal government.
00:17:02.320 --> 00:17:11.839
So they mapped over 20,000 work tasks across 800 occupations to millions of real AI conversations with their Claude chatbot.
00:17:11.920 --> 00:17:16.960
These were anonymized, but essentially they said, okay, what are people using Claude for?
00:17:17.119 --> 00:17:26.160
And they're matching millions of these chats to 20,000 work tasks across these 800 occupations uh from this federal data.
00:17:26.400 --> 00:17:45.920
And what they found in theory was that computer and math roles could solve 94% of uh sorry, uh today's most capable AI models could automate up to 94% of computer and math tasks, right?
00:17:46.240 --> 00:17:50.480
But they were only seeing 33% usage, right?
00:17:50.559 --> 00:17:57.279
Because people assume, oh, of computer and math, yeah, everyone's using AI for that because everyone knows that's the lowest hanging fruit, uh, right?
00:17:57.440 --> 00:18:00.160
Anything in coding, software development, uh, et cetera.
00:18:00.319 --> 00:18:13.359
Yet even in the area where people assumed, oh, yeah, literally everyone, if you're working in anything computer, right, uh software engineering, anything dev, anything with math, of course, you're using AI.
00:18:13.440 --> 00:18:15.200
And they said, well, actually not, right?
00:18:15.279 --> 00:18:23.839
And even office, admin, business, financial, and legal roles all revealed the same deep uh kind of adoption shortfall, right?
00:18:23.920 --> 00:18:30.319
So uh for our live stream audience, you can always see the video version of this on our website at your everydayai.com.
00:18:30.480 --> 00:18:37.119
So I have the theoretical capability and observe usage by occupational category, kind of map that anthropic put together.
00:18:37.200 --> 00:18:38.319
Uh very fascinating.
00:18:38.400 --> 00:18:50.640
But all this shows is in certain categories, right, such as management, business and finance, computer and math, um, you know, architecture and engineering, legal is another high one, arts and media, right?
00:18:50.799 --> 00:19:02.240
Kate, like AI's capabilities, when mapped to literal, the actual um work tasks that the federal government uses to define these roles.
00:19:02.799 --> 00:19:08.079
I mean, most of them have coverage in the 80 to 90 percent.
00:19:08.319 --> 00:19:12.799
Yet the observed, or sorry, the um, that's the theoretical coverage.
00:19:12.880 --> 00:19:18.400
So in theory, today's most powerful models could do anywhere from 80 to 90 percent of the actual work, right?
00:19:18.559 --> 00:19:23.839
Computer and math was the high one where they had an observed coverage of a little more than 30%.
00:19:24.319 --> 00:19:29.200
But these other areas are not even 20%, right?
00:19:29.359 --> 00:19:33.599
They're for the most part, you know, in the 20, some of them below 20.
00:19:33.759 --> 00:19:48.480
So you have this huge gap where in most of these general cases, right, like management, you know, in theory, today's AI models, if you understand the capabilities, they can automate 85 plus percent.
00:19:48.640 --> 00:19:52.079
Yet you don't even have a 20% coverage.
00:19:52.240 --> 00:19:55.119
That is a crazy gap, right?
00:19:55.200 --> 00:20:07.200
Where essentially you have a magic wand that if you know how to use the magic wand and you know the magic words, and you say, poof, work be gone, poof, the work is gone, but people don't know because it's impossible to keep up with.
00:20:07.440 --> 00:20:20.480
And some other things that they found uh on this study, which is actually the workers that are most impacted, they said that workers most exposed to AI right now earn 47% more and hold graduate degrees at nearly a four times rate.
00:20:20.640 --> 00:20:37.839
So essentially, I think people early on assumed that AI would displace or would uh potentially have the capabilities to displace workers who were maybe a little more junior or you know, weren't as high up the kind of quote unquote knowledge work totem pole, so to speak.
00:20:38.000 --> 00:20:40.480
And what they found was the exact opposite, right?
00:20:40.559 --> 00:20:52.480
It is those uh people who have those higher degrees, uh, and then conversely, people with physical and manual occupations register the lowest exposure according to anthropics study.
00:20:53.119 --> 00:20:55.759
So, how did we get here so quickly?
00:20:56.000 --> 00:20:56.480
Right?
00:20:56.799 --> 00:21:06.640
How do we get, as an example from GDP Val, where literally five and a half months ago it was even, right?
00:21:06.720 --> 00:21:07.599
It was a little less.
00:21:07.680 --> 00:21:11.759
It was about a 47% tie or win rate against humans, right?
00:21:11.839 --> 00:21:13.039
It was a it was a coin flip.
00:21:13.440 --> 00:21:22.240
And if the off-the-shelf model that you can pay$20 a month for was as good as a you know a human with a decade of experience who specializes in that.
00:21:22.400 --> 00:21:27.759
To now, humans aren't gonna be able to compete in a couple of months, right?
00:21:27.920 --> 00:21:35.200
By blind benchmarks, almost every single time within probably six months, humans are always gonna prefer the AI model.
00:21:35.279 --> 00:21:36.880
How do we get here so quickly?
00:21:37.119 --> 00:21:37.759
Right?
00:21:38.079 --> 00:21:43.759
Two years ago, the AI models were not good at all, right?
00:21:43.920 --> 00:21:46.000
They're they're actually pretty bad, right?
00:21:46.079 --> 00:21:48.640
Especially obviously we have today's comparison to draw back on.
00:21:48.799 --> 00:21:52.079
But one of the biggest reasons, I think, is recursive self improvement.
00:21:52.240 --> 00:21:54.000
Stick with me here if you're not super technical.
00:21:54.160 --> 00:21:59.599
All right, so recursive self improvement or RSI, it's a concept within artificial general intelligence.
00:22:00.240 --> 00:22:10.720
Essentially, an AI system improves its own source code, architecture, or training data, leading to a more capable model, which then improves itself further, creating a self-reinforcing loop.
00:22:10.880 --> 00:22:11.119
All right.
00:22:11.200 --> 00:22:20.160
So why am I talking about this kind of strange niche concept called recursive self-improvement on something about managing the AI capability gap?
00:22:20.400 --> 00:22:33.119
Well, because at the end of 2025, the AI companies started to either directly admit or to uh kind of allude to the fact that they were all now using recursive self-improvement uh on their models, right?
00:22:33.200 --> 00:22:39.200
So now their quote unquote big models were good enough that it could start writing its own code.
00:22:39.279 --> 00:22:41.519
It could start improving itself, right?
00:22:41.680 --> 00:22:56.559
Uh, I think um, you know, anthropic with their clawed code, probably one of the most consequential uh products of the AI lifecycle outside of uh ChatGPT, you could make the argument that clawed code is one of the most important outside of ChatGPT, right?
00:22:56.960 --> 00:23:02.400
The lead at Anthropic's Claude Code says, yeah, we don't write code anymore.
00:23:02.799 --> 00:23:06.480
You know, clawed codes, clawed code, right?
00:23:06.880 --> 00:23:12.480
And we've seen the same uh inferred by different researchers at OpenAI as well.
00:23:12.640 --> 00:23:25.680
Uh, you know, Google has been a little less direct, but they've still alluded to the fact that a lot of their models today, not just the smaller versions that are distilled from the biggest versions, but the biggest versions are being improved by themselves, right?
00:23:25.759 --> 00:23:34.000
And these new big scary models, right, like Anthropics Mythos or, you know, OpenAIs, whatever it's called, you know, SPUD or Glacier, right?
00:23:34.079 --> 00:23:38.160
If you're listening to this in six months, these names, these code names don't matter anymore.
00:23:38.400 --> 00:23:52.079
But one of the reasons that maybe the general public isn't getting their hands on them is well, they're too compute intensive, but they are using these models internally to improve the best consumer models and to also ship new products.
00:23:52.240 --> 00:24:00.720
And that's why Anthropic, as an example, their ship rate in in February in March was straight up off the charts, right?
00:24:00.799 --> 00:24:11.680
And then we found out later, well, one of the reasons was because they were using this mythos model to put out uh a lot of this, a lot of these new features that we're all using now.
00:24:11.839 --> 00:24:13.440
And and this is why, right?
00:24:13.599 --> 00:24:19.519
Because a year ago, it would have taken a team, even using AI, it would have taken teams way longer.
00:24:19.599 --> 00:24:27.519
But now that you have this kind of recursive self-improvement or models improving themselves and building new features that uh use those models, right?
00:24:27.599 --> 00:24:30.400
This is why it's hard now to keep up.
00:24:31.359 --> 00:24:41.920
So this started, like I said, in 2025, and ever since, large language model capabilities have far outpaced enterprise training and learning and development.
00:24:42.240 --> 00:24:54.000
You know, even the companies that wanted to do it right, I think they could they could do it, you know, in uh, you know, quarter two, quarter three of last year, but now you can't, right?
00:24:54.160 --> 00:25:03.119
Unless you have an entire segment of people, um, you know, maybe 5% of your total workforce that literally has no deliverables.
00:25:03.279 --> 00:25:04.400
And all they do, right?
00:25:04.480 --> 00:25:13.119
I've said this all along, your team needs a bunch of me's where all they do all day is they just scope different AI models.
00:25:13.200 --> 00:25:16.400
They play with AI releases, they sandbox thing.
00:25:16.559 --> 00:25:24.240
They're not building anyone for anything, they're just building solutions for what they think the company needs, and then they're training those people.
00:25:24.400 --> 00:25:26.079
But companies don't have that, right?
00:25:26.160 --> 00:25:30.000
And that's why, at least right now, it is nearly impossible to keep up.
00:25:30.160 --> 00:25:34.240
And most companies, though, claim AI adoption, but they can't actually prove real results.
00:25:34.319 --> 00:25:36.000
And it's getting even harder and harder, right?
00:25:36.079 --> 00:25:47.680
That McKinsey study that I talked about in uh 2025 said that 88% of organizations are using AI, but only 6% are generating meaningful business profits.
00:25:47.839 --> 00:25:55.039
And right now, I think leadership competence is running far ahead of what frontline practitioners report actually seeing on the ground.
00:25:55.200 --> 00:25:56.160
That's the thing, right?
00:25:56.319 --> 00:26:17.519
So a lot of times the people who are in charge of AI at certain companies, now, because it's getting easier to build, I think maybe their hands are on keyboard or their eyes are on monitoring agents a little bit more and they are getting removed from what the frontline practitioners are actually experiencing.
00:26:17.680 --> 00:26:27.440
So not only is it pretty hard to manage that gap just from a technical perspective, but from a change management in a people management perspective, it's getting even harder.
00:26:28.319 --> 00:26:35.839
And I think that's why some AI capability gaps are closing, but others are still remaining stuck.
00:26:36.000 --> 00:26:42.960
So, as an example, right, AI is helping AI-enabled humans close some gaps, but many still exist and are getting worse.
00:26:43.119 --> 00:26:45.519
So, as an example, coding performance, right?
00:26:45.680 --> 00:26:55.440
This and in a lot of different benchmarks, this surge from single digits to 90% unstructured benchmarks in three years in terms of what the AI itself was capable to do.
00:26:55.599 --> 00:27:03.920
And that helps obviously the humans that use this as part of their daily daily workflow close a big chunk of that gap individually, right?
00:27:04.000 --> 00:27:13.440
But you now have this, you know, PhD level uh models with you know science and standardized math capabilities that are rapidly approaching the ceiling of the current tests.
00:27:13.599 --> 00:27:21.839
So you aren't even necessarily able to know by the benchmarks what the capability gap is because these benchmarks are becoming saturated.
00:27:21.920 --> 00:27:28.480
So we even the AI community doesn't even fully understand these models' capabilities because the benchmarks, right?
00:27:28.559 --> 00:27:32.319
A lot of them have been stuck in the 90 to 95 percentile.
00:27:32.559 --> 00:27:34.720
You know, a lot of them are at 98, 99.
00:27:35.039 --> 00:27:37.920
Uh so I think the benchmarks themselves are getting saturated.
00:27:38.000 --> 00:27:44.480
So even the people building the models aren't even fully aware or understand what these models are actually capable of.
00:27:45.440 --> 00:27:49.920
So let's talk about how you can actually manage this gap and start to close it.
00:27:50.160 --> 00:28:03.039
All right, right now the top performers, right, the top companies are investing more than 20% of their digital budgets to fundamentally rework their processes.
00:28:03.200 --> 00:28:03.599
All right.
00:28:03.759 --> 00:28:06.319
Let me repeat that 20% of their digital budgets.
00:28:06.400 --> 00:28:14.319
So whatever your digital budget budget is, I'm guessing most companies are probably at 1%, maybe, maybe 5%.
00:28:15.200 --> 00:28:24.400
Very few, only the top performers are investing that 20% of their digital budget to rework their day-to-day knowledge processes, right?
00:28:24.559 --> 00:28:32.000
Because you need to learn to separate workflows into risk tiers with verification and human approval matched to each level.
00:28:32.240 --> 00:28:39.519
And organizations that wait for AI to become ready, they're gonna find that their competitors have already captured the advantage.
00:28:39.680 --> 00:28:49.599
Because in the same way, right, just to draw a little parallel here, what anthropic, and I keep saying they've won 2026 so far.
00:28:49.759 --> 00:28:51.759
Well, it's because they were first, right?
00:28:51.839 --> 00:28:59.119
They were first to um at least internally uh close that capability gap and put it to work, right?
00:28:59.200 --> 00:29:02.640
And that's why they've been able to outpace their competitors.
00:29:02.799 --> 00:29:05.119
Um, I don't know how long it'll last.
00:29:05.200 --> 00:29:09.119
We'll see, because I think the other the other labs have closed uh that gap.
00:29:09.200 --> 00:29:10.799
And I think we'll start to see that soon.
00:29:10.880 --> 00:29:13.759
But think of that within your own organization, right?
00:29:13.920 --> 00:29:20.880
The first to close the gap is going to be able to accelerate at a pace that we haven't seen before.
00:29:21.039 --> 00:29:21.279
Right.
00:29:21.359 --> 00:29:37.519
That's why you see unfortunately a lot of these big companies, you know, block as an example, cutting 40% of their workforce, and they seem fairly confident that they're going to be able to actually grow revenue because of the way they they've completely reworked their organization.
00:29:37.680 --> 00:29:46.240
Actually, Jack Dorsey had a very fascinating essay that we shared about in our in our newsletter about how they're essentially flipping the work pyramid on its head.
00:29:47.039 --> 00:29:51.759
So here's what I want to have you focus on.
00:29:52.079 --> 00:29:52.960
One metric.
00:29:53.359 --> 00:29:53.680
Okay.
00:29:54.480 --> 00:29:59.599
I want to make this very digestible for you and hopefully actionable as we close out today's show.
00:29:59.839 --> 00:30:08.559
I want you to track the percentage of AI-assisted workflow steps right now that are accepted without any rework or incident.
00:30:08.960 --> 00:30:10.559
And here's why.
00:30:10.960 --> 00:30:19.759
Because you probably, most organizations, they kind of get their AI plan or their AI training for the year, right?
00:30:19.920 --> 00:30:22.720
Or some companies, unfortunately, it's kind of like one time.
00:30:23.200 --> 00:30:27.599
Maybe the companies that are really investing heavily, you might get it once a quarter.
00:30:27.759 --> 00:30:31.519
But when was the last time that you used the most capable thinking models?
00:30:31.680 --> 00:30:33.839
I'm talking about GPT-5.4 Pro.
00:30:34.160 --> 00:30:37.039
I'm talking about Opus 4.6 with extended reasoning.
00:30:37.200 --> 00:30:41.839
I'm talking about Gemini 3.1 Pro with uh a higher thinking budget.
00:30:42.000 --> 00:30:42.799
Um, right?
00:30:42.960 --> 00:30:51.039
When was the last time that you used those and tracked the percentage of your workflows that get accepted without rework or incident?
00:30:51.119 --> 00:30:54.559
And then I want you to break that metric down by risk tier, right?
00:30:54.640 --> 00:30:59.519
So the low risk, you know, drafting versus the high stakes legal or financial outputs.
00:30:59.680 --> 00:31:37.359
So organizations that measure and understand that operational reliability for the high percentage of AI-assisted workflows that can be accepted that are lower risk, you will be surprised that depending on what your team does, depending on what your company does, depending on what sector you're in, you'd be surprised without too much investment, aside from just reverse engineering your current day-to-day processes, you'd be surprised to say that about 30 to maybe 60% of a lot of the work that many of us do, right?
00:31:37.440 --> 00:31:39.839
I'm not talking about, you know, people in specialized role.
00:31:39.920 --> 00:31:46.079
I'm saying if your organization has a thousand employees, you know, if you look at that work collectively, you you'd be surprised.
00:31:46.160 --> 00:31:59.519
I would say 30 to 60%, if you rescope everything, would fall in that can essentially be automated without rework or incident and is in a lower risk or a medium risk tier.
00:31:59.839 --> 00:32:04.880
And if you're not measuring that on an ongoing basis, you can't scale.
00:32:05.039 --> 00:32:08.799
But that's step one to managing the AI capability gap.
00:32:08.880 --> 00:32:12.400
And you have to be doing this right at least monthly.
00:32:13.519 --> 00:32:17.759
Again, a year ago, you could get away with quarterly, but the way models, right?
00:32:18.000 --> 00:32:23.359
If you're using a model from last quarter, good luck, right?
00:32:23.759 --> 00:32:28.240
That's showing up to a to an F1 race in a bicycle.
00:32:28.400 --> 00:32:28.880
Good luck.
00:32:28.960 --> 00:32:29.839
You're gonna get smoked.
00:32:29.920 --> 00:32:31.119
You don't stand a chance.
00:32:31.279 --> 00:32:35.599
So you can no longer take these year-long pilots, these quarter-long plans.
00:32:35.839 --> 00:32:42.000
You have to be agile to actually understand number one, the capability gap, but to begin to manage it.
00:32:42.160 --> 00:32:42.480
All right.
00:32:42.640 --> 00:32:45.440
I hope this was helpful in our start here series.
00:32:45.599 --> 00:32:49.039
Like I said, make sure to repost today's episode on LinkedIn.
00:32:49.119 --> 00:33:02.799
Here's why we put together the AI Capability Gap Report card to help you know where your organization stands when it comes to number one, understanding this gap, and number two, how you can actually tackle it.
00:33:02.880 --> 00:33:15.920
So, in this report card, it's a great guide for you and your team to go through it together to understand the latest capabilities of all the models and how they break down for different types of work.
00:33:16.079 --> 00:33:16.400
All right.
00:33:16.480 --> 00:33:21.599
So if you didn't know, yes, if you're listening on the podcast, this is actually live streamed on LinkedIn.
00:33:21.680 --> 00:33:29.359
Um, so in the podcast show notes, we always put a link to today's uh uh LinkedIn um show.
00:33:29.519 --> 00:33:35.440
So just go click repost, and I will send that capability gap report card your way.
00:33:35.599 --> 00:33:36.640
All right, thank you for tuning in.
00:33:36.720 --> 00:33:40.000
If you haven't already, please go to starthearseries.com.
00:33:40.160 --> 00:33:43.119
That's going to give you free access to the inner circle community.
00:33:43.279 --> 00:33:51.519
And you can go listen to all of the Start Here series in order in the playlist that we have, and you can go read and listen to all of the shows there.
00:33:51.599 --> 00:33:52.559
So thank you for tuning in.
00:33:52.640 --> 00:33:55.680
We hope to see you back tomorrow and every day for more everyday AI.
00:33:55.759 --> 00:33:56.799
Thanks, y'all.
00:00:01.919 --> 00:00:07.440
There's been a lot of talk lately about these AI models that are scary good.
00:00:07.679 --> 00:00:20.399
And even more chatter about how many of the brightest minds in AI and all the tech AI CEOs feel that we've already reached artificial general intelligence and are racing toward superintelligence.
00:00:20.640 --> 00:00:23.679
But what does that matter to you or your company?
00:00:23.920 --> 00:00:27.920
Because right now, those are kind of just theoretical issues.
00:00:28.079 --> 00:00:33.439
It's kind of like when you see an email on a Friday and you look at it and you're like, yeah, that's a Monday problem.
00:00:33.679 --> 00:00:40.799
But do you know what's scarier than some private AI model that's scary good or artificial superintelligence?
00:00:41.039 --> 00:00:42.719
The AI capability gap.
00:00:42.799 --> 00:00:44.320
That's the scariest of all.
00:00:44.640 --> 00:00:45.280
Why?
00:00:45.679 --> 00:00:47.600
Because it's very real.
00:00:47.920 --> 00:00:56.399
And unlike that Friday afternoon email that you're maybe going to tackle on Monday, the AI capability gap is a today problem.
00:00:56.640 --> 00:01:02.240
And if you don't address it head on, it's going to slowly stall your company's growth.
00:01:02.320 --> 00:01:10.319
And if your company isn't growing, the AI capability gap will undoubtedly stop your company dead in its tracks.
00:01:10.480 --> 00:01:12.560
And this is not an exaggeration.
00:01:12.719 --> 00:01:15.920
And the timing here is specifically urgent.
00:01:16.079 --> 00:01:26.000
So we're going to hit rewind today and explain the capability gap in AI and why AI is racing ahead of your company.
00:01:26.159 --> 00:01:30.480
And we're going to show you how to slow it down a bit and catch it.
00:01:30.719 --> 00:01:32.000
All right, you ready?
00:01:32.159 --> 00:01:36.560
This is part of our start here series on everyday AI.
00:01:36.799 --> 00:01:38.400
Let's get into it.
00:01:38.640 --> 00:01:43.120
So if you are new here, well, let's just get to the big picture.
00:01:43.519 --> 00:01:47.840
AI capability right now is outpacing business adoption.
00:01:47.920 --> 00:01:50.319
And it has changed recently.
00:01:50.400 --> 00:01:53.680
And the pace is too fast for any of us.
00:01:53.840 --> 00:01:55.439
Right now, Frontier AI.
00:01:55.519 --> 00:01:59.280
And when we talk about actual capabilities, this is not an exaggeration.
00:01:59.439 --> 00:02:08.639
Frontier AI models, when used correctly, and that's the big asterisk here, they now match or exceed human professionals on most defined knowledge work tasks.
00:02:08.800 --> 00:02:19.840
And there's a massive gap that persists between what AI could, in theory, automate today and what organizations are actually using it for, which is often just topical.
00:02:19.919 --> 00:02:20.319
Uh, right.
00:02:20.400 --> 00:02:28.879
There's a McKinsey study that said only about 6% of organizations are generating meaningful profits despite widespread AI adoption.
00:02:29.039 --> 00:02:32.960
And the bottleneck is not actually the AI models.
00:02:33.199 --> 00:02:40.400
You could have made that argument maybe mid-2025, but not anymore in 2026.
00:02:40.800 --> 00:02:47.039
It's actually about organizational adoption, workflow design, and training and education.
00:02:47.199 --> 00:02:49.360
So here's what we're going to tackle on today's show.
00:02:49.439 --> 00:02:52.479
So stick with me for the next, I'm going to try to make this one 25 minutes.
00:02:52.560 --> 00:02:52.960
We'll see.
00:02:53.120 --> 00:02:56.319
All right, stick with me for the next 22-ish minutes.
00:02:56.400 --> 00:03:02.000
And you're going to learn the benchmark proving that AI outperforms human experts, and it's not even close.
00:03:02.159 --> 00:03:09.680
You're going to know more about Anthropic's new-ish study, revealing most knowledge workers barely touch AI's true automation potential.
00:03:09.840 --> 00:03:14.800
You're going to learn why the AI capability gap only started to show itself over the past few months.
00:03:14.879 --> 00:03:23.520
It's actually kind of new, and what the top 6% of AI performing companies actually do that everyone else completely ignores.
00:03:23.759 --> 00:03:25.759
All right, welcome to Everyday AI.
00:03:25.840 --> 00:03:27.759
And this is our Start Here series.
00:03:27.919 --> 00:03:33.840
This is the essential podcast series to both learn the AI basics and for AI experts to double down.
00:03:34.000 --> 00:03:34.400
All right.
00:03:34.479 --> 00:03:40.719
I started this thing because after 750 plus episodes, everyone always said, Jordan, great podcast, right?
00:03:40.800 --> 00:03:42.400
When they found it, they're like, where do I start?
00:03:42.560 --> 00:03:43.039
I have no clue.
00:03:43.199 --> 00:03:44.800
And I'm like, I also have no clue.
00:03:44.960 --> 00:03:46.879
So that's why I started the Start Here series.
00:03:46.960 --> 00:03:50.400
Uh, it's best if you are brand new here, start in order.
00:03:50.479 --> 00:03:50.800
All right.
00:03:50.879 --> 00:03:55.120
These are faster uh, you know, podcasts, usually about 25 to 35 minutes.
00:03:55.280 --> 00:03:57.759
But if you listen on 2x, they're even half that, right?
00:03:57.840 --> 00:04:02.240
Uh, but this is a great way to listen from one to now, volume 19.
00:04:02.319 --> 00:04:04.960
And then also make sure to go to startherseries.com.
00:04:05.039 --> 00:04:08.800
That's going to give you free access to our exclusive inner circle community.
00:04:08.960 --> 00:04:31.199
You can't find access right now any gap report card.
00:04:31.439 --> 00:04:35.040
All right, so uh a little bit more on that at the end.
00:04:37.279 --> 00:04:40.480
But trust me, you are going to want to repost today's episode.
00:04:40.639 --> 00:04:41.360
I'm just saying.
00:04:41.600 --> 00:04:55.600
All right, and if you missed our last start here series in volume 18, and this was episode 750, we went over the vibe coding boom, why vibe coding isn't going away, and how it's both good and bad.
00:04:55.680 --> 00:04:57.199
So make sure to go check that one out.
00:04:57.279 --> 00:05:05.600
But now let's talk about how you can actually manage the AI capability gap and why AI is more than ready and companies are not.
00:05:05.759 --> 00:05:08.480
All right, and I'm not the only one talking about this.
00:05:08.639 --> 00:05:17.439
I've been talking about this now since I think late 2025, but a lot of super smart people are starting to dive in deep on the AI capability gap.
00:05:17.600 --> 00:05:19.519
I really like what Jack Clark said.
00:05:19.759 --> 00:05:24.480
He is uh the anthropic um co-founder of anthropic.
00:05:24.560 --> 00:05:27.839
And here's what he said in a post a couple of weeks ago.
00:05:28.000 --> 00:05:30.720
He said, most of AI progress has this flavor.
00:05:30.879 --> 00:05:39.920
If you have a bit of intellectual curiosity and some time, you can very quickly shock yourself with how amazingly capable modern AI systems are.
00:05:40.160 --> 00:05:44.319
But you need to have that magic combination of time and curiosity.
00:05:44.480 --> 00:05:59.120
And otherwise, you're going to consume AI like most people do, as a passive viewer of some unremarkable synthetic slop content, or at best, just asking your LLM of choice how to roast a turkey and keep it moist.
00:05:59.360 --> 00:06:03.279
Or Tony Box lights spinning, but not playing music.
00:06:03.519 --> 00:06:04.720
What do I do?
00:06:04.879 --> 00:06:08.240
And all the amazing advancements are mostly hidden from you.
00:06:08.319 --> 00:06:14.959
All right, so that's what Jack Clark said in a uh kind of a viral uh post that he had a couple of weeks ago um on X.
00:06:15.839 --> 00:06:18.480
And I think this is very telling.
00:06:18.959 --> 00:06:22.319
And the two things that he talked about, I think are very true.
00:06:22.480 --> 00:06:30.160
All right, to start, well, first to even realize the AI capability gap, let alone close it, you have to be extremely curious.
00:06:30.399 --> 00:06:48.240
And the reason why I say that is because if you work how you've been working for the last few decades, you're not going to discover, or your or your team, your organization is not truly going to discover that capability gap because AI is not meant for the average knowledge worker, right?
00:06:48.399 --> 00:06:59.040
That's why I talk about all the time how I hate, absolutely hate the concept of upskilling, because AI is not something that you sprinkle on top of a seasoned knowledge worker, right?
00:06:59.120 --> 00:07:03.839
Like myself, I've been working full time for 20-ish years, right?
00:07:04.160 --> 00:07:05.920
I can't just sprinkle AI on the top.
00:07:06.000 --> 00:07:07.600
I have to, I have to unlearn, right?
00:07:07.680 --> 00:07:13.920
So you do have to have the time and you have to be curious on, hey, what does my role look like if I completely start it over?
00:07:14.000 --> 00:07:17.360
And if I kind of forgot everything that got me to the point I am in my career.
00:07:17.439 --> 00:07:21.360
And that's what you have to do, and you also have to have a lot of time and you have to devote a lot of time to it.
00:07:21.519 --> 00:07:29.279
Uh, Olivia Moore from A16Z uh talked about, she said, uh, OpenAI dropped a state of enterprise reports across a million customers.
00:07:29.439 --> 00:07:36.639
This was a couple of months ago, but her response, she said, the gulf between AI power users and everyone else is wide.
00:07:36.879 --> 00:07:45.839
The 95th percentile user spends six sends six times more messages than the median with coding, writing, and analysis showing the biggest gaps.
00:07:46.000 --> 00:07:52.399
So, yeah, those people that are power users, well, because they actually understand the capability gap, they're using AI all the time.
00:07:52.560 --> 00:08:00.160
This is why myself, and I'm not saying this is like some weird flex, I'm saying this because I'm part of this uh, you know, power user group.
00:08:00.319 --> 00:08:06.399
When I say that I use AI for 10 to 12 hours a day, right, that's how much I'm working.
00:08:06.480 --> 00:08:08.959
Uh, you know, it's not like I'm working 16 hours a day.
00:08:09.199 --> 00:08:19.920
I am only working in AI every single day, no matter where I am, I'm using AI from beginning to end in literally every single step in between.
00:08:20.079 --> 00:08:22.319
And it has completely reshaped my workflow.
00:08:22.480 --> 00:08:27.600
Uh uh Kevin Roos, uh New York Times columnist, said this, and I really like how he put this.
00:08:27.759 --> 00:08:34.399
He said, I follow AI adoption pretty closely, and I have never seen such a yawning inside-outside gap.
00:08:34.559 --> 00:08:48.320
People in SF San Francisco are putting multi-agent cloud storms in charge of their lives, consulting chatbots before every decision, wire heading to a degree only sci-fi writers dared to imagine.
00:08:48.559 --> 00:08:55.039
People everywhere are still trying to get approval to use co-pilot in Teams if they're using AI at all.
00:08:55.200 --> 00:09:06.320
It's possible that the early adopter bubble I'm in has always been this intense, but there seems to be a cultural kick takeoff happening in addition to the technical one, not ideal.
00:09:06.639 --> 00:09:16.559
And I'll say from my experience, I think there's five main reasons that the AI capability gap has really been exposed in 2026.
00:09:16.720 --> 00:09:21.120
And number one, the models of what they can actually do, right?
00:09:21.279 --> 00:09:23.200
I'm just gonna roll through these things really quickly.
00:09:23.360 --> 00:09:25.759
So, number one, what AI models can actually do.
00:09:25.919 --> 00:09:29.360
Number two, what business leaders think AI can do.
00:09:29.679 --> 00:09:31.840
Number three, AI literacy.
00:09:31.919 --> 00:09:33.840
Uh, number four, human skill set.
00:09:33.919 --> 00:09:35.279
And number five, AI access.
00:09:35.360 --> 00:09:38.240
Now, let me break those five things down a little bit more in depth.
00:09:38.399 --> 00:09:41.120
So, number one, what AI models can actually do.
00:09:41.360 --> 00:09:42.639
A year ago, right?
00:09:42.720 --> 00:09:48.399
So I don't know when you're listening to this Start Here series episode, you know, could be in April 2026.
00:09:48.480 --> 00:09:50.480
You might be listening to it at the end of 2026.
00:09:50.559 --> 00:09:51.039
I don't know.
00:09:51.120 --> 00:09:51.679
Uh, right.
00:09:51.759 --> 00:09:59.679
But a year ago, in the beginning of 2025, um, you could, for the most part, understand what AI models are capable of.
00:09:59.919 --> 00:10:02.480
Today, you absolutely can't.
00:10:02.639 --> 00:10:06.000
And it is literally getting to the point of science fiction, right?
00:10:06.080 --> 00:10:11.440
All of these math problems and science problems that have been plaguing researchers for decades are now being solved.
00:10:11.600 --> 00:10:13.039
I look at them and I have no clue, right?
00:10:13.120 --> 00:10:15.840
I read the papers and I'm like, okay, I don't have any clue.
00:10:16.000 --> 00:10:18.080
I can follow and understand it a year ago.
00:10:18.240 --> 00:10:23.440
What AI models can actually do today is mind-boggling.
00:10:23.600 --> 00:10:26.639
Number two, what business leaders think AI can do.
00:10:26.799 --> 00:10:29.759
Um, I get to talk to a lot of smart people.
00:10:31.759 --> 00:10:33.440
It's really right.
00:10:33.519 --> 00:10:40.080
I've I've had hundreds of guests on this show, but I also maybe a handful that I've talked to.
00:10:40.320 --> 00:10:45.759
And I'm like, yes, this business leader fully understands what AI can do for the most part.
00:10:45.840 --> 00:10:53.519
And that's not their fault necessarily, because most people in their role, they have to be an expert at one application, right?
00:10:53.679 --> 00:10:58.080
You can't be an like very few business leaders know what AI can do.
00:10:58.240 --> 00:11:03.519
I feel fairly confident that I'm in that category, but only because this is all I do.
00:11:03.679 --> 00:11:08.159
If I had another job, right, and I was just an AI champion at a company, you can't.
00:11:08.320 --> 00:11:09.759
Like you literally can't.
00:11:09.919 --> 00:11:12.480
You used to be able to, you can't anymore.
00:11:12.639 --> 00:11:14.399
Because literally every single day, right?
00:11:14.480 --> 00:11:21.600
Whether you're talking about uh Google, Microsoft, Open AI, Anthropic, now Meta's back in the conversation, perplexity, et cetera, right?
00:11:21.759 --> 00:11:28.960
You literally can't have an actual job where you have to uh produce something for a company and still understand what AI can do.
00:11:29.039 --> 00:11:29.519
Not anymore.
00:11:29.600 --> 00:11:30.879
You used to be able to juggle that.
00:11:30.960 --> 00:11:34.320
Now, unless you're literally working 20 plus hours a day, you can't do it.
00:11:34.480 --> 00:11:36.559
All right, number three, AI literacy.
00:11:36.879 --> 00:11:43.279
And this kind of goes uh, you know, hand in hand with uh, you know, number two, what business leaders understanding.
00:11:43.360 --> 00:11:56.080
So there's a difference between um understanding AI's capabilities and then AI literacy, because you have to actually be able to speak the language foundationally, which again, most people skip over that, right?
00:11:56.159 --> 00:12:03.519
They go straight to the bells and the whistles, they go straight to clicking the button, not knowing what happens, you know, from A to X.
00:12:03.759 --> 00:12:07.279
They only want to get to Y in the result that brings on Z.
00:12:07.600 --> 00:12:09.759
Uh number four is human skill set.
00:12:10.080 --> 00:12:12.080
All right, this is different than literacy.
00:12:12.240 --> 00:12:16.720
You have to have the foundational understanding, but then you also have to have these skills, right?
00:12:16.799 --> 00:12:21.200
I know a lot about basketball, but my basketball skill set is not great anymore, right?
00:12:21.360 --> 00:12:23.200
I think I peaked in eighth grade.
00:12:23.279 --> 00:12:26.080
Uh, but you have to have that human skill set.
00:12:26.159 --> 00:12:28.320
Uh, and then last but not least is AI access.
00:12:28.399 --> 00:12:33.039
So that's access to the tools, to the technology in your role.
00:12:33.279 --> 00:12:37.360
Okay, so what can AI models actually do?
00:12:37.600 --> 00:12:37.919
Right.
00:12:38.080 --> 00:12:52.240
And I do want to spend a little bit more time on that because that will better explain this gap that I think started manageable and now it's very hard to manage that capability gap because it has turned into a friggin' canyon.
00:12:52.480 --> 00:12:58.559
So the reason being is because AI models are better than expert humans.
00:12:58.799 --> 00:13:09.440
All right, I've talked about this uh this evaluation a couple of times, but the more I follow these AI benchmarks and all of these other things, uh, there's there's dozens of them.
00:13:09.600 --> 00:13:11.600
I think there's about five that matter, right?
00:13:11.679 --> 00:13:15.600
And one of them, and I think probably the most important one, is GDP Val.
00:13:15.919 --> 00:13:17.840
So this is a benchmark from OpenAI.
00:13:17.919 --> 00:13:24.000
And the reason why I think it's the most valuable benchmark to look at is because it's about creating business value.
00:13:24.159 --> 00:13:34.080
So, more or less, uh, this benchmark evaluates AI models against real professional deliverables across 44 different high GDP occupations.
00:13:34.159 --> 00:13:47.120
So these are things that you and I may do, you know, going in and doing research on a fast-moving market segment and then creating an artifact like a, you know, a pitch deck or creating a spreadsheet, something like that.
00:13:47.279 --> 00:14:04.080
These are real-world domain-specific tasks where both a human and an AI model from start to finish uh completed a task and then submitted their final version to a panel of experts who are experts in that field.
00:14:04.240 --> 00:14:04.559
All right.
00:14:04.639 --> 00:14:12.080
And expert judges then compared the unlabeled AI and human outputs in a blind head-to-head pairwise comparison.
00:14:12.240 --> 00:14:23.840
And what we've seen now is, as an example, the best model for actually creating front-to-back economically valuable outputs like a human would is the OpenAI's GPT-5.
00:14:24.720 --> 00:14:30.799
And it matches or exceeds industry professionals in 83% of these evaluations.
00:14:31.279 --> 00:14:31.919
Right.
00:14:33.360 --> 00:14:38.559
I remember when some of these first GDP valve numbers came out, and I'm like, oh, that's that's pretty impressive, right?
00:14:38.639 --> 00:14:42.320
When we were in the 30, 40%, but now it's it's undeniable.
00:14:42.399 --> 00:14:49.200
And I do assume that by the end of the year, that number is gonna be like in the mid-90s, right?
00:14:49.840 --> 00:14:54.320
My like one of my predictions is it was gonna get to 80%, and we're already at 80%.
00:14:54.639 --> 00:14:56.080
It was just a huge jump.
00:14:56.159 --> 00:15:03.200
Uh, that's why I I keep talking about the AI capabilities are just running too fast for most business leaders to understand and keep up.
00:15:03.360 --> 00:15:04.480
And it's nearly doubled.
00:15:04.559 --> 00:15:07.759
That GDP bow score has nearly doubled in five months.
00:15:07.919 --> 00:15:26.080
So when I talk about the last, you know, since essentially late 2025, uh, I'm gonna talk about why, but it has been, we've seen more developments in the past five months than we have in the five years prior, at least when it comes to large language models, and it's not even close.
00:15:26.320 --> 00:15:27.039
Um, right.
00:15:27.200 --> 00:15:32.960
So the in October, the best AI model scored on that was a 47%.
00:15:33.519 --> 00:15:39.840
So it hasn't quite doubled, but it's nearly doubled in about five or six months.
00:15:40.080 --> 00:15:44.559
And right now, AI handles most defined professional tasks yet.
00:15:44.799 --> 00:15:48.159
Most organizations barely use models in this way.
00:15:48.240 --> 00:15:49.840
It's like, did you know, right?
00:15:49.919 --> 00:15:56.159
As an example, your uh, you know, GPT and Claude models can do this.
00:15:56.320 --> 00:15:58.720
They can create artifacts by default.
00:15:58.799 --> 00:16:00.559
You don't have to have anything special turned on.
00:16:00.720 --> 00:16:07.679
It can just create spreadsheets, presentations, word docs, etc., all personalized based on your company's info.
00:16:07.919 --> 00:16:15.440
All right, next, I want to talk very briefly about the anthropic labor data study that came out uh about two months ago.
00:16:15.679 --> 00:16:17.679
I did go over this in more depth.
00:16:17.759 --> 00:16:21.440
So if you're interested in this, go listen to episode 730.
00:16:21.840 --> 00:16:36.480
Here's essentially uh similarly, uh, whereas the GDP benchmark looks at actual economic output, uh, this anthropic labor data study looks a little bit more at the capability gap.
00:16:36.639 --> 00:16:50.879
So that's why I think these two uh kind of this benchmark, OpenAI benchmark, and the anthropic labor daddy labor data study in tandem really tell a powerful story of where we're at and why this capability gap even exists.
00:16:51.039 --> 00:16:56.240
So in this study, Anthropic mapped over 20,000 work tasks, right?
00:16:56.320 --> 00:17:02.159
So they actually used uh some public uh some public jobs data from the federal government.
00:17:02.320 --> 00:17:11.839
So they mapped over 20,000 work tasks across 800 occupations to millions of real AI conversations with their Claude chatbot.
00:17:11.920 --> 00:17:16.960
These were anonymized, but essentially they said, okay, what are people using Claude for?
00:17:17.119 --> 00:17:26.160
And they're matching millions of these chats to 20,000 work tasks across these 800 occupations uh from this federal data.
00:17:26.400 --> 00:17:45.920
And what they found in theory was that computer and math roles could solve 94% of uh sorry, uh today's most capable AI models could automate up to 94% of computer and math tasks, right?
00:17:46.240 --> 00:17:50.480
But they were only seeing 33% usage, right?
00:17:50.559 --> 00:17:57.279
Because people assume, oh, of computer and math, yeah, everyone's using AI for that because everyone knows that's the lowest hanging fruit, uh, right?
00:17:57.440 --> 00:18:00.160
Anything in coding, software development, uh, et cetera.
00:18:00.319 --> 00:18:13.359
Yet even in the area where people assumed, oh, yeah, literally everyone, if you're working in anything computer, right, uh software engineering, anything dev, anything with math, of course, you're using AI.
00:18:13.440 --> 00:18:15.200
And they said, well, actually not, right?
00:18:15.279 --> 00:18:23.839
And even office, admin, business, financial, and legal roles all revealed the same deep uh kind of adoption shortfall, right?
00:18:23.920 --> 00:18:30.319
So uh for our live stream audience, you can always see the video version of this on our website at your everydayai.com.
00:18:30.480 --> 00:18:37.119
So I have the theoretical capability and observe usage by occupational category, kind of map that anthropic put together.
00:18:37.200 --> 00:18:38.319
Uh very fascinating.
00:18:38.400 --> 00:18:50.640
But all this shows is in certain categories, right, such as management, business and finance, computer and math, um, you know, architecture and engineering, legal is another high one, arts and media, right?
00:18:50.799 --> 00:19:02.240
Kate, like AI's capabilities, when mapped to literal, the actual um work tasks that the federal government uses to define these roles.
00:19:02.799 --> 00:19:08.079
I mean, most of them have coverage in the 80 to 90 percent.
00:19:08.319 --> 00:19:12.799
Yet the observed, or sorry, the um, that's the theoretical coverage.
00:19:12.880 --> 00:19:18.400
So in theory, today's most powerful models could do anywhere from 80 to 90 percent of the actual work, right?
00:19:18.559 --> 00:19:23.839
Computer and math was the high one where they had an observed coverage of a little more than 30%.
00:19:24.319 --> 00:19:29.200
But these other areas are not even 20%, right?
00:19:29.359 --> 00:19:33.599
They're for the most part, you know, in the 20, some of them below 20.
00:19:33.759 --> 00:19:48.480
So you have this huge gap where in most of these general cases, right, like management, you know, in theory, today's AI models, if you understand the capabilities, they can automate 85 plus percent.
00:19:48.640 --> 00:19:52.079
Yet you don't even have a 20% coverage.
00:19:52.240 --> 00:19:55.119
That is a crazy gap, right?
00:19:55.200 --> 00:20:07.200
Where essentially you have a magic wand that if you know how to use the magic wand and you know the magic words, and you say, poof, work be gone, poof, the work is gone, but people don't know because it's impossible to keep up with.
00:20:07.440 --> 00:20:20.480
And some other things that they found uh on this study, which is actually the workers that are most impacted, they said that workers most exposed to AI right now earn 47% more and hold graduate degrees at nearly a four times rate.
00:20:20.640 --> 00:20:37.839
So essentially, I think people early on assumed that AI would displace or would uh potentially have the capabilities to displace workers who were maybe a little more junior or you know, weren't as high up the kind of quote unquote knowledge work totem pole, so to speak.
00:20:38.000 --> 00:20:40.480
And what they found was the exact opposite, right?
00:20:40.559 --> 00:20:52.480
It is those uh people who have those higher degrees, uh, and then conversely, people with physical and manual occupations register the lowest exposure according to anthropics study.
00:20:53.119 --> 00:20:55.759
So, how did we get here so quickly?
00:20:56.000 --> 00:20:56.480
Right?
00:20:56.799 --> 00:21:06.640
How do we get, as an example from GDP Val, where literally five and a half months ago it was even, right?
00:21:06.720 --> 00:21:07.599
It was a little less.
00:21:07.680 --> 00:21:11.759
It was about a 47% tie or win rate against humans, right?
00:21:11.839 --> 00:21:13.039
It was a it was a coin flip.
00:21:13.440 --> 00:21:22.240
And if the off-the-shelf model that you can pay$20 a month for was as good as a you know a human with a decade of experience who specializes in that.
00:21:22.400 --> 00:21:27.759
To now, humans aren't gonna be able to compete in a couple of months, right?
00:21:27.920 --> 00:21:35.200
By blind benchmarks, almost every single time within probably six months, humans are always gonna prefer the AI model.
00:21:35.279 --> 00:21:36.880
How do we get here so quickly?
00:21:37.119 --> 00:21:37.759
Right?
00:21:38.079 --> 00:21:43.759
Two years ago, the AI models were not good at all, right?
00:21:43.920 --> 00:21:46.000
They're they're actually pretty bad, right?
00:21:46.079 --> 00:21:48.640
Especially obviously we have today's comparison to draw back on.
00:21:48.799 --> 00:21:52.079
But one of the biggest reasons, I think, is recursive self improvement.
00:21:52.240 --> 00:21:54.000
Stick with me here if you're not super technical.
00:21:54.160 --> 00:21:59.599
All right, so recursive self improvement or RSI, it's a concept within artificial general intelligence.
00:22:00.240 --> 00:22:10.720
Essentially, an AI system improves its own source code, architecture, or training data, leading to a more capable model, which then improves itself further, creating a self-reinforcing loop.
00:22:10.880 --> 00:22:11.119
All right.
00:22:11.200 --> 00:22:20.160
So why am I talking about this kind of strange niche concept called recursive self-improvement on something about managing the AI capability gap?
00:22:20.400 --> 00:22:33.119
Well, because at the end of 2025, the AI companies started to either directly admit or to uh kind of allude to the fact that they were all now using recursive self-improvement uh on their models, right?
00:22:33.200 --> 00:22:39.200
So now their quote unquote big models were good enough that it could start writing its own code.
00:22:39.279 --> 00:22:41.519
It could start improving itself, right?
00:22:41.680 --> 00:22:56.559
Uh, I think um, you know, anthropic with their clawed code, probably one of the most consequential uh products of the AI lifecycle outside of uh ChatGPT, you could make the argument that clawed code is one of the most important outside of ChatGPT, right?
00:22:56.960 --> 00:23:02.400
The lead at Anthropic's Claude Code says, yeah, we don't write code anymore.
00:23:02.799 --> 00:23:06.480
You know, clawed codes, clawed code, right?
00:23:06.880 --> 00:23:12.480
And we've seen the same uh inferred by different researchers at OpenAI as well.
00:23:12.640 --> 00:23:25.680
Uh, you know, Google has been a little less direct, but they've still alluded to the fact that a lot of their models today, not just the smaller versions that are distilled from the biggest versions, but the biggest versions are being improved by themselves, right?
00:23:25.759 --> 00:23:34.000
And these new big scary models, right, like Anthropics Mythos or, you know, OpenAIs, whatever it's called, you know, SPUD or Glacier, right?
00:23:34.079 --> 00:23:38.160
If you're listening to this in six months, these names, these code names don't matter anymore.
00:23:38.400 --> 00:23:52.079
But one of the reasons that maybe the general public isn't getting their hands on them is well, they're too compute intensive, but they are using these models internally to improve the best consumer models and to also ship new products.
00:23:52.240 --> 00:24:00.720
And that's why Anthropic, as an example, their ship rate in in February in March was straight up off the charts, right?
00:24:00.799 --> 00:24:11.680
And then we found out later, well, one of the reasons was because they were using this mythos model to put out uh a lot of this, a lot of these new features that we're all using now.
00:24:11.839 --> 00:24:13.440
And and this is why, right?
00:24:13.599 --> 00:24:19.519
Because a year ago, it would have taken a team, even using AI, it would have taken teams way longer.
00:24:19.599 --> 00:24:27.519
But now that you have this kind of recursive self-improvement or models improving themselves and building new features that uh use those models, right?
00:24:27.599 --> 00:24:30.400
This is why it's hard now to keep up.
00:24:31.359 --> 00:24:41.920
So this started, like I said, in 2025, and ever since, large language model capabilities have far outpaced enterprise training and learning and development.
00:24:42.240 --> 00:24:54.000
You know, even the companies that wanted to do it right, I think they could they could do it, you know, in uh, you know, quarter two, quarter three of last year, but now you can't, right?
00:24:54.160 --> 00:25:03.119
Unless you have an entire segment of people, um, you know, maybe 5% of your total workforce that literally has no deliverables.
00:25:03.279 --> 00:25:04.400
And all they do, right?
00:25:04.480 --> 00:25:13.119
I've said this all along, your team needs a bunch of me's where all they do all day is they just scope different AI models.
00:25:13.200 --> 00:25:16.400
They play with AI releases, they sandbox thing.
00:25:16.559 --> 00:25:24.240
They're not building anyone for anything, they're just building solutions for what they think the company needs, and then they're training those people.
00:25:24.400 --> 00:25:26.079
But companies don't have that, right?
00:25:26.160 --> 00:25:30.000
And that's why, at least right now, it is nearly impossible to keep up.
00:25:30.160 --> 00:25:34.240
And most companies, though, claim AI adoption, but they can't actually prove real results.
00:25:34.319 --> 00:25:36.000
And it's getting even harder and harder, right?
00:25:36.079 --> 00:25:47.680
That McKinsey study that I talked about in uh 2025 said that 88% of organizations are using AI, but only 6% are generating meaningful business profits.
00:25:47.839 --> 00:25:55.039
And right now, I think leadership competence is running far ahead of what frontline practitioners report actually seeing on the ground.
00:25:55.200 --> 00:25:56.160
That's the thing, right?
00:25:56.319 --> 00:26:17.519
So a lot of times the people who are in charge of AI at certain companies, now, because it's getting easier to build, I think maybe their hands are on keyboard or their eyes are on monitoring agents a little bit more and they are getting removed from what the frontline practitioners are actually experiencing.
00:26:17.680 --> 00:26:27.440
So not only is it pretty hard to manage that gap just from a technical perspective, but from a change management in a people management perspective, it's getting even harder.
00:26:28.319 --> 00:26:35.839
And I think that's why some AI capability gaps are closing, but others are still remaining stuck.
00:26:36.000 --> 00:26:42.960
So, as an example, right, AI is helping AI-enabled humans close some gaps, but many still exist and are getting worse.
00:26:43.119 --> 00:26:45.519
So, as an example, coding performance, right?
00:26:45.680 --> 00:26:55.440
This and in a lot of different benchmarks, this surge from single digits to 90% unstructured benchmarks in three years in terms of what the AI itself was capable to do.
00:26:55.599 --> 00:27:03.920
And that helps obviously the humans that use this as part of their daily daily workflow close a big chunk of that gap individually, right?
00:27:04.000 --> 00:27:13.440
But you now have this, you know, PhD level uh models with you know science and standardized math capabilities that are rapidly approaching the ceiling of the current tests.
00:27:13.599 --> 00:27:21.839
So you aren't even necessarily able to know by the benchmarks what the capability gap is because these benchmarks are becoming saturated.
00:27:21.920 --> 00:27:28.480
So we even the AI community doesn't even fully understand these models' capabilities because the benchmarks, right?
00:27:28.559 --> 00:27:32.319
A lot of them have been stuck in the 90 to 95 percentile.
00:27:32.559 --> 00:27:34.720
You know, a lot of them are at 98, 99.
00:27:35.039 --> 00:27:37.920
Uh so I think the benchmarks themselves are getting saturated.
00:27:38.000 --> 00:27:44.480
So even the people building the models aren't even fully aware or understand what these models are actually capable of.
00:27:45.440 --> 00:27:49.920
So let's talk about how you can actually manage this gap and start to close it.
00:27:50.160 --> 00:28:03.039
All right, right now the top performers, right, the top companies are investing more than 20% of their digital budgets to fundamentally rework their processes.
00:28:03.200 --> 00:28:03.599
All right.
00:28:03.759 --> 00:28:06.319
Let me repeat that 20% of their digital budgets.
00:28:06.400 --> 00:28:14.319
So whatever your digital budget budget is, I'm guessing most companies are probably at 1%, maybe, maybe 5%.
00:28:15.200 --> 00:28:24.400
Very few, only the top performers are investing that 20% of their digital budget to rework their day-to-day knowledge processes, right?
00:28:24.559 --> 00:28:32.000
Because you need to learn to separate workflows into risk tiers with verification and human approval matched to each level.
00:28:32.240 --> 00:28:39.519
And organizations that wait for AI to become ready, they're gonna find that their competitors have already captured the advantage.
00:28:39.680 --> 00:28:49.599
Because in the same way, right, just to draw a little parallel here, what anthropic, and I keep saying they've won 2026 so far.
00:28:49.759 --> 00:28:51.759
Well, it's because they were first, right?
00:28:51.839 --> 00:28:59.119
They were first to um at least internally uh close that capability gap and put it to work, right?
00:28:59.200 --> 00:29:02.640
And that's why they've been able to outpace their competitors.
00:29:02.799 --> 00:29:05.119
Um, I don't know how long it'll last.
00:29:05.200 --> 00:29:09.119
We'll see, because I think the other the other labs have closed uh that gap.
00:29:09.200 --> 00:29:10.799
And I think we'll start to see that soon.
00:29:10.880 --> 00:29:13.759
But think of that within your own organization, right?
00:29:13.920 --> 00:29:20.880
The first to close the gap is going to be able to accelerate at a pace that we haven't seen before.
00:29:21.039 --> 00:29:21.279
Right.
00:29:21.359 --> 00:29:37.519
That's why you see unfortunately a lot of these big companies, you know, block as an example, cutting 40% of their workforce, and they seem fairly confident that they're going to be able to actually grow revenue because of the way they they've completely reworked their organization.
00:29:37.680 --> 00:29:46.240
Actually, Jack Dorsey had a very fascinating essay that we shared about in our in our newsletter about how they're essentially flipping the work pyramid on its head.
00:29:47.039 --> 00:29:51.759
So here's what I want to have you focus on.
00:29:52.079 --> 00:29:52.960
One metric.
00:29:53.359 --> 00:29:53.680
Okay.
00:29:54.480 --> 00:29:59.599
I want to make this very digestible for you and hopefully actionable as we close out today's show.
00:29:59.839 --> 00:30:08.559
I want you to track the percentage of AI-assisted workflow steps right now that are accepted without any rework or incident.
00:30:08.960 --> 00:30:10.559
And here's why.
00:30:10.960 --> 00:30:19.759
Because you probably, most organizations, they kind of get their AI plan or their AI training for the year, right?
00:30:19.920 --> 00:30:22.720
Or some companies, unfortunately, it's kind of like one time.
00:30:23.200 --> 00:30:27.599
Maybe the companies that are really investing heavily, you might get it once a quarter.
00:30:27.759 --> 00:30:31.519
But when was the last time that you used the most capable thinking models?
00:30:31.680 --> 00:30:33.839
I'm talking about GPT-5.4 Pro.
00:30:34.160 --> 00:30:37.039
I'm talking about Opus 4.6 with extended reasoning.
00:30:37.200 --> 00:30:41.839
I'm talking about Gemini 3.1 Pro with uh a higher thinking budget.
00:30:42.000 --> 00:30:42.799
Um, right?
00:30:42.960 --> 00:30:51.039
When was the last time that you used those and tracked the percentage of your workflows that get accepted without rework or incident?
00:30:51.119 --> 00:30:54.559
And then I want you to break that metric down by risk tier, right?
00:30:54.640 --> 00:30:59.519
So the low risk, you know, drafting versus the high stakes legal or financial outputs.
00:30:59.680 --> 00:31:37.359
So organizations that measure and understand that operational reliability for the high percentage of AI-assisted workflows that can be accepted that are lower risk, you will be surprised that depending on what your team does, depending on what your company does, depending on what sector you're in, you'd be surprised without too much investment, aside from just reverse engineering your current day-to-day processes, you'd be surprised to say that about 30 to maybe 60% of a lot of the work that many of us do, right?
00:31:37.440 --> 00:31:39.839
I'm not talking about, you know, people in specialized role.
00:31:39.920 --> 00:31:46.079
I'm saying if your organization has a thousand employees, you know, if you look at that work collectively, you you'd be surprised.
00:31:46.160 --> 00:31:59.519
I would say 30 to 60%, if you rescope everything, would fall in that can essentially be automated without rework or incident and is in a lower risk or a medium risk tier.
00:31:59.839 --> 00:32:04.880
And if you're not measuring that on an ongoing basis, you can't scale.
00:32:05.039 --> 00:32:08.799
But that's step one to managing the AI capability gap.
00:32:08.880 --> 00:32:12.400
And you have to be doing this right at least monthly.
00:32:13.519 --> 00:32:17.759
Again, a year ago, you could get away with quarterly, but the way models, right?
00:32:18.000 --> 00:32:23.359
If you're using a model from last quarter, good luck, right?
00:32:23.759 --> 00:32:28.240
That's showing up to a to an F1 race in a bicycle.
00:32:28.400 --> 00:32:28.880
Good luck.
00:32:28.960 --> 00:32:29.839
You're gonna get smoked.
00:32:29.920 --> 00:32:31.119
You don't stand a chance.
00:32:31.279 --> 00:32:35.599
So you can no longer take these year-long pilots, these quarter-long plans.
00:32:35.839 --> 00:32:42.000
You have to be agile to actually understand number one, the capability gap, but to begin to manage it.
00:32:42.160 --> 00:32:42.480
All right.
00:32:42.640 --> 00:32:45.440
I hope this was helpful in our start here series.
00:32:45.599 --> 00:32:49.039
Like I said, make sure to repost today's episode on LinkedIn.
00:32:49.119 --> 00:33:02.799
Here's why we put together the AI Capability Gap Report card to help you know where your organization stands when it comes to number one, understanding this gap, and number two, how you can actually tackle it.
00:33:02.880 --> 00:33:15.920
So, in this report card, it's a great guide for you and your team to go through it together to understand the latest capabilities of all the models and how they break down for different types of work.
00:33:16.079 --> 00:33:16.400
All right.
00:33:16.480 --> 00:33:21.599
So if you didn't know, yes, if you're listening on the podcast, this is actually live streamed on LinkedIn.
00:33:21.680 --> 00:33:29.359
Um, so in the podcast show notes, we always put a link to today's uh uh LinkedIn um show.
00:33:29.519 --> 00:33:35.440
So just go click repost, and I will send that capability gap report card your way.
00:33:35.599 --> 00:33:36.640
All right, thank you for tuning in.
00:33:36.720 --> 00:33:40.000
If you haven't already, please go to starthearseries.com.
00:33:40.160 --> 00:33:43.119
That's going to give you free access to the inner circle community.
00:33:43.279 --> 00:33:51.519
And you can go listen to all of the Start Here series in order in the playlist that we have, and you can go read and listen to all of the shows there.
00:33:51.599 --> 00:33:52.559
So thank you for tuning in.
00:33:52.640 --> 00:33:55.680
We hope to see you back tomorrow and every day for more everyday AI.
00:33:55.759 --> 00:33:56.799
Thanks, y'all.
