Transcription
So, there's this rumor going around Silicon Valley that we are in the last 300 days of work. Now, I want to tell you where this rumor came from and what I think about it, but so as to not bury the lead, we're just going to say that no, we're not necessarily truly in the last 300 days of work. But the next 300 days are going to be very interesting. And I do think that we are going to hit a few tipping points or inflection points. So, let's get into it.
So, this is the source of the rumor. Kevin Roose wrote on Twitter just 2 days ago at the time of recording, "Overheard at an AI lab. How were you spending the last 300 days of work?" Now, of course, there's not much context here, and this sounds like your typical hype. However, Kevin is a New York Times tech columnist, co-host of the Hardfork Podcast, and author of three books writing about AGI with 172,000 followers. So you might say, okay, well obviously he's got some vested interest in building hype. He's writing a book about AGI and all of that is perfectly valid. So you know, some people say, "Oh, well Dave, you shouldn't do ad hominem." But I think the messenger matters, particularly on narratives. However, if we take him at face value and say, "Yes, maybe he did actually hear this." He is he is positioned both geographically because he's in San Francisco and professionally to have maybe overheard that. Did he hear it from OpenAI? Did he hear it from Google? Did he hear it from Anthropic or somewhere else? We don't know. He hasn't, uh, you know, expounded upon this revelation anymore. But there is some signal here. So let's unpack what that could actually mean.
From my perspective, the primary thing that this means is that if if it is true, so just let's just hang a lampshade on that and pre-qualify everything that if he did in fact hear this from someone credible at a credible AI lab. Okay, cool. If that happened, then what it could mean is that the people inside these frontier AI labs are seeing trends towards maybe not full AGI, but basically the saturation of things like GDP val, which is a benchmark that measures how economically effective or impactful an AI could be. And of course, all of you would be right to point out that a benchmark is not reality. A benchmark is an approximation of reality, and they're often misleading or unhelpful, particularly the fact that they get saturated so quickly. But we haven't seen AI completely eat the world.
With that being said, I do have some examples of how AI is proliferating very quickly. My wife is a marketing and social media specialist, uh, amongst other things. She does lots of customer work for her company. And over the last few months, she has switched to where most of her job is just using Claude or Gemini and directing those tools. So those tools have reached a tipping point where they are able to do a ton of the grunt work. Not obviously the prestige high-end polish work of a high-end professional, but they're able to do a lot of the grunt work of sales, marketing, and that sort of thing. Um, and also customer research. But it's not just those raw tools that she's using. There are more and more integrations for these tools, but also there are tools that she uses that have AI integrated in the back end. So things like Canva and other things are adding more and more AI. So basically everything is copilot now, where she was using Slack, and Slack has, if you're not familiar, Slack is like a Teams chat thing. It's sort of like Microsoft Teams. So it is a business-oriented, uh, workspace for, uh, asynchronous communication, asynchronous business communication, and it has a lot of built-in hooks and automations and other ways that you can extend it. Well, all of those are accessible via co-pilots now. And when I say copilot, I don't mean specifically GitHub copilot or Microsoft copilot. What I mean is the particular model where you have an AI chat window writing shotgun with the main thing, and you basically just tell it what you want it to do. And so over just the last couple weeks, she has seen multiple tools that she uses on a daily basis integrate these kinds of abilities.
Now, there's multiple adaptation curves happening here. So number one, there is the underlying model itself. So when you hear OpenAI or Anthropic or, you know, the frontier lab saying, "Ah, we're going to saturate work in in 300 days," in terms of raw capability, that might be true. Obviously, we're in the middle of 2026, which is the year of the agent. So agents are everywhere. People are spending literally millions of dollars per month on tokens running agents at some of these frontier labs. If you haven't heard the news, then, uh, basically it's like they put token leaderboards, and the assumption is you should be using agents to spend millions, if not billions, of tokens every day, every week, every month, so that you are learning how to build better agents. That is what's happening at the frontier. However, what's happening at the frontier is very different from the skill level of front-end users like my wife, and also the downstream integrations for other companies, because Canva, just as an example, is not a frontier AI company. So it takes them time to learn. So there's the learning curve that happens on their level, and then there's the learning curve that happens on the front-end users.
Now, my wife works at a very small company. Smaller companies tend to be more agile when they have good leadership, and her company does have good leadership because the CEO of her company explicitly makes it a like a policy point that they should be using AI as much as possible, and they have a weekly stand-up meeting on every Friday to say, "What have we learned about AI this week? What can AI do for us this week?" And they make it so that everyone has to share, and so that it's a it's a very, uh, deliberately, like, learning-oriented culture, and they know that AI is the next big thing, and they've made it a first-class priority in their business. So there are multiple layers of adoption curves that have to happen for things to get saturated.
Now, you might be saying, like, "Okay, Dave, you're the post-labor economics guy. Aren't you worried that your wife is going to lose her job?" Not necessarily. Now, here's the thing that is happening at her workplace, as well as literally every other small business and medium-sized business that I've talked with or heard about, and that is that they are slowing hiring down. Uh, this is out there in the news as well, because if you look at the data that shows that, uh, new grad hiring is down 9 to 13% amongst companies that use generative AI more, and then the new grad unemployment rate is a little bit higher than it has been. It's only about 5.6%, 6%, which doesn't sound that bad, but when you look at the broader unemployment rate at less than, I think, 4.8%. So, new grads are still hurting. Also, new grads are far more likely to be underemployed, which is another sign that the ladder is being pulled up. They're calling it the junior crisis. I have a blog post that I'm working on about this. Anyways, back to the point at hand.
So, we are seeing the last 300 days of work from one vantage point. Now, one thing that we need to point out is that Silicon Valley is often delusional. And I don't mean that in the in that they are delusional about the actual, like, what is the model physically capable of? What they're delusional about is how it's going to diffuse through the rest of society, the difficulty of how you take these things up, and that sort of thing. Now, my in my past life, I worked at very large enterprises. Some of them were cutting edge. The biggest one that I worked at was Cisco Systems, which is a Silicon Valley networking, uh, company. I've also worked for old brick-and-mortar companies. Uh, Advanced Auto Parts was my last corporate job, and they are one of the slowest adopters of technology. Why? Because cars haven't really fundamentally changed in like almost a hundred years, and the the the logistics business, cuz it's fundamentally a logistics business, is mostly about, you know, time, distance, warehouses. Technology obviously helps, it's indispensable for that kind of thing, but it is not for first and foremost a software company. So I've seen the full range of how these kinds of things get diffused out, and the thing is is there's going to be legacy code, there's legacy systems, there is all kinds of stuff that AI I just doesn't really hit yet.
And so Theo, uh, Theo Von, I think that's his name. Big-time YouTuber, um, also, uh, entrepreneur. He and I had a very brief exchange. Let me show it to you. Okay, it was Theo Gigi. Sorry, not Theo, different people. Anyways, Theo, uh, about a week ago wrote, "A lot of people are building with the assumption that the code bases we work in today will still matter next year. I'm not sure if that's the case." Now, this is a, you know, entrepreneur. I think he's in Silicon Valley. Anyways, software-first kind of guy. And I said, "Legacy code will stick around for a long time. Yes, the models are going to get better at refactoring and updating them. Whoops, sorry. But yeah, don't underestimate corporate inertia and the 'if it ain't broke, don't fix it' mentality that's out there." Theo says, "I think those corporations are much easier to replace now. I also think the incentive for them to rewrite has gone up. Cost of all software is rapidly approaching zero. Corporations either respond or die."
Now, before I read my rebuttal to that, I will say yes, in the long run. However, there's a lot of inertia, and that inertia can carry companies for a very long time. So my response was, "In some cases, yes, but I'm thinking about the auto parts store that I used to serve as an infrastructure engineer at. Software can only do so much. I mean, yes, the business is dependent upon software, but the core business activity is logistics, which yes, software helps with, but it's interesting because once you have an optimal fit, marginal improvements of software at the edges doesn't really move the needle. With that said, sorry, with that being said, I 100% agree with you because the technical debt at those legacy companies needs to be erased. One reason I no longer work for private corporations." So we are in agreement. I'm not going to say "violent agreement" because one, I think that's a stupid turn of phrase, but also we're looking at it from slightly different perspectives. I'm looking at it from the perspective of someone who's managed multi-petabyte data centers, and yes, you know, when you talk to a software developer and they say, "We're going to automate everything." To him, he's thinking automating code and automating pull requests and automating, uh, you know, CI/CD pipelines and that sort of thing. But when you look at the broader technology stack, you can't automate power. You can't automate, you know, the bare metal. You have to have someone physically plug in the servers, run the cables, and all kinds of other stuff. Now, a lot of that work gets poo-pooed on because it doesn't require as much, uh, you know, intelligence. It doesn't require as much education or certification. Um, but there is also the middleware, there's the infrastructure, and there's everything else, the backups, identity management. There's so many layers to this stuff that is resistant to automation.
Now, with that being said, I am aware that there is a concept called software-defined data center. And basically, you've only heard that if you're a virtualization and and infrastructure cloud nerd like I was. And so, yes, you can automate more and more of these things. And I'm not going to be the one to say it is never going to be automated because can I imagine a future in which all of this is software-defined? Absolutely. However, my point is that it takes a lot longer than 300 days to get there. Furthermore, it takes a lot longer than 300 days for, uh, not not even just the raw capability to have a product that could hypothetically do that. It takes a lot more than 300 days for a big company, a big big iron company or a brick-and-mortar company to actually go through that. It's 18 months minimum just for planning, uh, for for larger Fortune 500 enterprises. And then every step of the way, you're going to be arguing with the CTO. You're going to be arguing with the CI, the SISO, the CISO, the, um, chief information security officer. You're going to be arguing with the CEO. Is this the best use of these funds right now? Because here's the thing is when your top-line business model is not delivering more software, it's delivering things like parts or cars, the software is still just a point of friction. And what we're agreeing on here is that yes, better software can reduce friction, but when you have some, you know, some basement-dwelling devs or other tech guys that are like, "We need to improve everything right now." The CEO says, "What is the marginal improvement that we're going to get? Are we going to deliver parts faster or cheaper if we spend, you know, a million dollars or a billion dollars refactoring our entire data center?" No, it can wait.
And so this is one of the things where yes, I am super ultra optimistic, and many of you have pointed this out for years on my YouTube channel, that in the long run, we tend to underestimate how transformative these kinds of disruptive innovations can be and overestimate it in the short run. And that's if I had to boil it down, that's what I'm saying that this Silicon Valley rumor is, and also what Theo is saying. They're overestimating the short-term disruptive potential, um, but not underestimating the long-term disruptive potential. So, we're all on the same page there.
But I do want to provide a little bit of, uh, sanity check to say yes, within 300 days, I fully expect that agentic AI could probably take over most of the job that I used to do. Why? Because it was a KVM job, keyboard, video, mouse. If it's something that you can do entirely in front of a computer without a webcam or without actually being physically present, because by the time I left the corporate rat race, my job was entirely digital. It was entirely cybernetic. I was not going into data centers physically anymore. They had other people to do that for me with the rare exception of when there was something that either I was the best person to do it or there was just no one else available. But the point is that yes, we will get there. However, the adoption and the maturation of those capabilities and users getting used to it are all very different things. And then just at the end of the day, you're going to have corporate compliance officers, you're going to have the legal department, you're going to have the HR department at all these big companies. They're all going to be saying, "No, wait. Don't do it."
And also, I have actually gone to give talks and give education at some of these kinds of more brick-and-mortar and logistics companies, and 95% of the people in those companies are like terrified of AI. They're petrified of it. They don't want to touch it. They don't want to look at it. They don't want to think about it because it, from from most workers' perspective, it's just another thing that they're going to get judged by. It might be a threat. It if it if it messes up and they don't have any training, they're not familiar with it, they don't want to touch something and break it and get in trouble. So like what, what I'm saying is that most people do not understand how like paralyzed most Fortune 500 companies are, uh, when it comes to disruptive things, because disruption means lost revenue. It means, uh, what we call resume-generating events, RGE. So like there's so much friction, and that's not to say like that's not a that's not even a bad thing either, because whenever you see a story of like, "Oh yeah, Claude Code or whatever deleted a prod database," that's a multi-million dollar setback there. So every time those stories happen, a CEO or CTO says, "This technology isn't ready," and they just put it out of their mind. That's what most CTOs do. That's what most CEOs and chief information security officers do. They say that technology is still a toy. I don't care. It is not fit for purpose.
If you've been following along, you probably know that my book, Labor Zero, is about ready. We had a very successful Kickstarter, raised over $45,000. Most of that is going to the team. It's going to the editor. It's going to be going to the audio engineer. Next, I have finished with the manuscript. We have, uh, I used the community actually to decide on a subtitle. Um, the book cover is getting designed, and I'm going to be doing the narration myself. So, if you notice that I sound a little bit different in this video, that's because I've been practicing my narrator's voice, uh, and that sort of thing. I've been, um, doing drills and that sort of stuff. I've also been practicing with tea to see which kind of tea is the best for clearing out my voice. Now, I'm recovering from a chest cold right now, so if I sound a little weird, that's why.
But anyways, with that being said, my YouTube channel did get demonetized. Um, and that really sucks. That removed about a quarter of my income. However, I do still have Patreon. Um, so if you would like to support me, you can sign up on Patreon. There's also Substack, which, uh, allows you to sign up and support me there. Either way is fine. Um, some people have been, like, I've had a steady stream of messages of people who don't like Patreon over the over the years because they take their cut. And while that is true, Patreon only takes 8%, whereas YouTube took 50%. So, you know, I was losing half of half of what I was of the support I was getting from YouTube to YouTube anyways. So, now I'm not getting anything, and neither are they. I guess YouTube is probably still getting still getting their cut. But anyways, so with that being said, that's all. Thanks for watching. Let me know what you think, and cheers. Have a good one.