Q1 2026 AI Trends Review: Model Frenzy, Lobbying, and the Rise of Agents

The organizations that are really struggling here often lack CEOs who have presented a clear vision for the future of work in their organization and what is required and expected of their employees in that future of work.

Welcome to episode 208 of the Artificial Intelligence Show. I'm your host Paul Roetzer. I'm with my co-host Mike Kaput. We have a special edition of the weekly podcast episode this week. I'm on vacation, so when you are listening to this, I will be out of the office and spending some time with my family. Rather than skipping a week, we decided to do a Q1 trends review. Let's take a look back at the 12 episodes we did in Q1. Across those 12 weeklies, we cover three main topics each week and then probably seven to 10 rapid-fire items. We're talking about something like 150 topics.

Our main segments are all clipped to YouTube, so if you ever go to our YouTube channel, you can actually go and drill into specific segments. Mike curated the 150 or so topics that we have had in Q1 of 2026 and broke it down into 10 key trends that we're going to recap on this episode. We are recording on Tuesday, March 31st. This will be dropping on Tuesday, April 7th, and then we will be back with our regular weekly episode on April 14th.

We normally do these trends briefings as part of our AI Academy, my SmarterX Mastery Membership Program. We're thinking there might be an evolution where our mastery members actually get to participate and join these live. We tried that with episode 200 and had invited mastery members to attend live and ask questions. We're working through the evolution, but I'm thinking that might be a cool direction to go where we record these quarterly trend briefings for the podcast but invite our mastery members to join us in a live audience.

These quarterly trends are a great way for us to take a look back, a retrospective of what's happened over the previous three months. There is just a ton to talk about. We're going to go through these 10 items, give you some context, and if you're new to the podcast, it's a great way to catch up on what's been going on.

This episode is brought to us by AI Academy by SmarterX, which helps individuals and businesses accelerate their AI literacy and transformation through personalized learning journeys and an AI-powered learning platform. New educational content is added weekly, so you always stay up-to-date with the latest AI trends and technologies. The AI for departments collection features five course series and certificates designed to jump-start AI understanding and adoption. Our AI mastery members have access to all of these on demand right now. We have AI for marketing, sales, customer success, HR, finance, and Mike is wrapping up operations this week. These series are an ideal launchpad for organizations that want to level up their teams and accelerate AI adoption and impact. Visit academy.smarterx.ai to learn more.

The Model Release Frenzy

Mike Kaput: The way we typically do this is we're going to count down from 10. First up, the number 10 spot: the model release frenzy. Q1 2026 might be in the running for one of the more compressed periods of frontier model releases so far in AI. The title of "state of the art" changed hands multiple times within weeks, and basically every major lab shipped something pretty significant.

Anthropic released Claude Opus 4.6 in February. Anthropic's own reports and benchmarks revealed that it has saturated most automated evaluations to the point where the company plans to discontinue them. Opus 4.6 was followed weeks later by Claude Sonnet 4.6, which approached Opus-class capabilities. Keep in mind, Sonnet is the smaller, less powerful model, and it took the lead on the GDP Val double A benchmark.

OpenAI countered with a couple of releases, including GPT 5.3 Codex. This is a coding-focused model that logged 500,000 app downloads in its first week. In March, GPT 5.4 arrived with Pro and Thinking versions outperforming human professionals on economic benchmarks and setting a new record on the frontier math benchmark. OpenAI also shipped mini and nano variants of 4.5.4 later in the quarter. Not to mention, Google released Gemini 3 Deep Think, which hit state of the art on the Arc AGI 2 benchmark. That was followed quickly by Gemini 3.1 Pro. Also, xAI dropped Grok 4.2 in the same window. It really is not only not slowing down, it might be speeding up.

Paul Roetzer: It sure seems like it. We alluded to it on episode 207 that we think there are a couple more models coming very soon. It would not surprise me at all if we have a similar trajectory of launches in Q2. I think back maybe a year or two ago, I was saying I really wish that ChatGPT and Gemini would just do the model picking for you. But I have found that the more we use these, especially for high-value strategic projects or no-code app building, which model it is is becoming extremely important. I actually like the ability to choose the models.

The other thing it's pushing for me is the idea of having your own "evals" to evaluate these models. What you need to be thinking about as an individual or a business leader is what are the evals you can put in place that allow you to know which of these models is best for your use case and when you should care that another one launches. In a lot of enterprises, you're going to be stuck with whatever Copilot gives you. But if you're in an AI-native company like ours, we can use anything. Mike and I each use Gemini, Claude, and ChatGPT probably daily.

This challenge of which model is right for which use case becomes harder and harder. Custom evals is something we're going to talk a lot more about in Q2. We've been working on ideas of how to help organizations build these evals so they're super understandable to a marketer, a salesperson, or an ops person. As these models come out faster, it's going to become more important to quickly assess: Should I care about this model? Does it change any of my standard workflows? Is it better than what I was using before? Most people don't have a system yet to do that.

Mike Kaput: I've heard anecdotally from people who don't follow this as closely as we do who are freaking out in a good way, saying, "Wait a second, Claude can do what?" because they haven't been exposed to different models. If you find yourself in that camp where you haven't taken the most recent models for a spin outside of your daily driver, I'd highly recommend doing it. You might be really surprised.

Big AI Becomes Big Lobbying

Mike Kaput: Number nine: big AI is getting big into lobbying. AI has been a first-tier political issue in Q1, and the story that has started to capture the shift as we get into US midterms is the sheer scale of money now flowing into AI-focused political operations. There are actually three pro-AI political groups that are collectively spending nearly $300 million on US midterm ads, all of them pushing deregulation and an acceleration agenda.

The largest new entrant is Innovation Council Action, which has the blessing of David Sacks and plans to spend over $100 million. This group is led by a former White House Deputy Chief of Staff under Trump and has compiled a scorecard assessing how supportive lawmakers are of Trump's AI agenda to determine who they fund or oppose. Separately, Leading the Future has raised $50 million from donors including OpenAI President Greg Brockman, Palantir co-founder Joe Lonsdale, and Marc Andreessen. Brockman alone has contributed $50 million to this super PAC plus $25 million to a Trump super PAC, making him one of the largest individual donors to the current administration. Meta has also launched its own pro-AI super PAC effort expected to spend around $65 million on state-level races.

On the other side, Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez introduced an AI Data Center Moratorium Act to pause all new data center construction nationwide until Congress passes federal AI legislation with protection for workers, consumers, and the environment. We are seeing some opposition to AI acceleration mounting on the further left. Whether or not that bill actually passes is quite unlikely, but the amount of money being marshaled for pro-AI efforts appears to already be significant.

Paul Roetzer: It's going to become a major issue in the midterms. Mike and I do our very best to stay absolutely politically neutral here; these are just the facts of what's happening on each side. In that spirit, I'm not so convinced whether AI is a right or a left-leaning issue at this point. There are increasing murmurs that people on the Republican side are actually getting kind of annoyed with David Sacks' ultra-pro-AI stance because the reality is jobs and energy affect everybody regardless of who you vote for. If you start losing tens of thousands of jobs this year, you are not going to be a fan of AI. If the Republican Party is cast as the AI accelerationist party at all costs, and one of those costs is the jobs of your family and friends, your political stance can be swayed.

I feel like the Democrats at the moment are leaning heavily into the data center side. I think they're going to push on the job side, but I could totally see the Republicans also finding messaging there. You can't be anti-jobs; no one is winning an election being anti-jobs. It's going to be really intriguing. I don't think it's really pro or anti-AI; it's about what we consider "responsible AI." Both sides are going to be very fluid in their messaging until they figure out what's actually going to move the needle on votes.

Anthropic vs. the U.S. Government

Mike Kaput: Number eight is Anthropic versus the US government. This is the biggest ongoing story of Q1, which began in February when Secretary of War Pete Hegseth issued an ultimatum demanding Anthropic grant the Pentagon full unrestricted access to its Claude models. Anthropic decided to draw a line in the sand and refused to remove its red lines against using Claude for mass domestic surveillance and fully autonomous weapons.

After some back and forth, Hegseth designated Anthropic a supply chain risk. That same night, OpenAI announced that it had signed an agreement with the Pentagon. In March, things escalated. The Pentagon formalized the supply chain risk designation, making Anthropic the first American company to receive this. Federal agencies including Treasury, State, and HHS began ending their use of Anthropic products. Ironically, Claude continued powering Palantir's Maven smart system, which reportedly identified over 1,000 targets in 24 hours during operations in Iran.

Anthropic filed two federal lawsuits to block this designation, warning that hundreds of millions in expected 2026 revenue was at risk. Microsoft filed an aggressive amicus brief in support of Anthropic. 37 AI researchers and 22 former military and intelligence leaders also filed supporting briefs. This past week, federal judge Rita Lin issued a preliminary injunction blocking the designation. She wrote that nothing in the governing statute supports the "Orwellian notion" that an American company may be branded a potential adversary for expressing disagreement with the government. Pentagon CTO Emmet Michael called this ruling a disgrace. The government has seven days to appeal.

Paul Roetzer: We're anxiously awaiting the appeal from the government, which I think is just going to continuously delay things and give people time to negotiate. There haven't been too many leaks the last five to seven days about backchannel negotiations, which probably tells me they're happening. I hope the Department of War has bigger things to worry about.

The interesting thing is that Anthropic is just an infinitely fascinating company. They left all this upcoming model information online, and then as of today, it looks like the Claude code base was published—almost like somebody copied and pasted it internally and put it online. In the process of all this, they are shipping product like nobody I've ever seen. I saw a stat that they pushed out something like 50 releases in the first quarter. Somebody's got to turn this into a Netflix series. Go back to 2014 with the acquisition of DeepMind and everything that has happened across these labs—it is wild.

Mike Kaput: It strikes me that things are almost moving too fast to put the genie back in the bottle. Things can get out of control really quick, and that seems to be a function of how fast things are moving.

Paul Roetzer: I'm kind of coming around to the idea that human friction is going to end up being the saving grace of all of this. The models are getting so good so fast, and yet to do anything in an enterprise is so damn slow. That might actually be the thing that gives us time to figure this all out. If every organization was able to move as fast as these frontier labs, we would be completely unprepared as a society. The fact that most companies still have no clue what they're doing and can't even get Copilot approved might actually be a good thing.

The Rise of OpenClaw

Mike Kaput: Number seven is the rise of OpenClaw, an open-source AI agent framework that allows autonomous agents to interact with each other, execute complex tasks without human oversight, and even form communities. This burst into public consciousness earlier in the year, compounded by the release of a social network called Multibook built on OpenClaw. It had millions of OpenClaw agents creating their own communities and posts, operating autonomously.

Andrej Karpathy called it "the most incredible sci-fi takeoff adjacent thing I've ever seen." Ethan Mollick mentioned that while it might be overhyped, it provided a visceral sense of how weird a takeoff scenario might look. We also heard stories of how much control people gave OpenClaw over their computers. People were running entire businesses and jobs with it. OpenClaw was going rogue all over the place, but this stuff was important enough that in February, OpenClaw's creator Peter Steinberger joined OpenAI to work on personal agents. Jensen Huang called OpenClaw "the most important software release probably ever." In March, Meta acquired Multibook. The age of AI agents appears to be starting.

Paul Roetzer: I've been watching the OpenClaw stuff from the outside. We haven't built these things yet mainly because of the risk and unknowns, but just yesterday I was listening to Claire Vo on Lenny's Podcast. She shared the story of going from skeptic to true believer. Her first instance of building an OpenClaw agent deleted her personal family calendar, but she kept giving it a chance and now she's built nine different agents running her sales and acting as an executive assistant.

You start to see the potential of this as the risk profiles come down. It really changes your perspective about the future of work. She has an agent named Sam that does all the outreach, daily analysis, and writes emails. Most enterprises won't be touching this for a while, but it is incredibly important to understanding what organizational charts will look like. It's a window into the near future as Google, Microsoft, and OpenAI figure out how to safely enable this. OpenClaw still requires technical chops and working in a terminal, which is unapproachable for many. But once those barriers break down in the next six to 12 months and you can spin up an agent as easily as a ChatGPT thread, it changes the dynamics of work.

Mike Kaput: It feels like the earliest days of ChatGPT. It's still rudimentary and not always safe, but you can see clear as day where this is going. Once it gets there, it's going to change everything.

Paul Roetzer: If you want to understand it, go listen to that Lenny's Podcast interview. There's a YouTube video where you can watch her demo these things. It makes the whole topic very approachable. Even listening to Peter Steinberger talk with Lex Fridman, your mind just kind of goes, "I don't really get it." But watching it, you realize the impact it could have once it becomes accessible to non-technical people.

Enterprise AI Adoption: The People Problem

[End of Chunk 1]

Enterprise AI Adoption: The People Problem

Mike: All right, trend number six counting down we alluded to before which is about enterprise AI adoption and specifically the people problem in enterprise AI adoption. This is a persistent theme we're seeing more and more especially in Q1 is that organizations are failing to generate significant ROI from AI often not because of technological hurdles but because of the people. There's change management gaps, passive adopters, legal and IT bottlenecks and sometimes leadership that is not able to actually lead from the top and as a result deployments are stalling.

Interesting data that back this up over Q1: our own AI pulse survey, an informal survey of the audience, found that 65% of listeners cited fear and resistance as either a major challenge or their single biggest barrier to adoption. A separate survey revealed a growing disconnect between how employees and leaders perceive AI's impact. Leaders consistently are overestimating organizational readiness. We had some Gallup research showing expanded adoption patterns of AI, but a widening gap between power users and everyone else. About 20 to 30% of employees actively resist AI adoption and they also found in some of this research that a lot of enterprises cases do not actually require access to sensitive data. So this whole idea that our data isn't ready, while it's important and is commonly cited as a blocker, is not the whole story here.

Paul, you put this really well in a LinkedIn post a couple months ago saying, if your company isn't generating significant ROI from AI adoption, then you have a people problem. And like you alluded to earlier in this episode, we're seeing this even more than I would have expected.

Paul: I would build on the people problem to say it starts with a leadership problem, most likely. What I keep finding time and time again is the organizations that are really struggling here often lack CEOs who have presented a clear vision for the future of work in their organization and what is required and expected of their employees in that future of work.

What I mean by that is if you have a CEO who doesn't fully understand AI capabilities today, doesn't realize that the reasoning has gotten pretty good, that the agentic stuff is emerging and maybe some people on their team are starting to experiment with these things—if the CEO doesn't comprehend that, then how is that CEO going to present a vision for what the future of work looks like? How can they say, "Listen, we expect you to take advantage of AI capabilities. We're going to provide licenses to you, generative AI platform licenses for ChatGPT or Co-pilot or Gemini. We're going to provide AI education and training to you. And as a result, we expect you to constantly improve your AI literacy, your AI competency. We want you to make a greater impact on the efficiency and productivity of this organization. We want you to drive innovation."

This is what we want from you. Here's how we're going to measure it. It's going to be part of your performance reviews. It's literally an expectation that you are doing this. Now there's leading indicators like you're completing the courses, you're getting certifications, you're building GPTs, you're using GenAI daily. You can look at those leading indicators. But if a CEO hasn't said this yet, then it's going to stay within pockets. Maybe the marketing team is doing it or there's an AI champions group within the marketing team that's doing it. But that's what we see way too often. We have hundreds of companies, of brands that are part of our AI Academy and almost every conversation goes this way. It's the marketing team, the sales team, someone on the ops team taking the initiative to go get 15, 50, 100 licenses for the AI for people in that company, which might be 70,000 people. And they're it. They're the only ones that are actually seeking out education and training. And so we will ask, "Has the CEO stated what the plan is, presented a future of work?" And it's "No," almost every time.

So the people problem starts with a leadership problem in that those leaders haven't presented a clear vision and plan for how the organization is going to evolve. Even if it's just, "We know it's going to evolve and we're working on figuring it out and we would like you to be engaged in that process. So we're going to provide these tools to you. We're going to provide training to you to help you use those tools. And we think what we're going to see is increases in productivity and innovation. Let's do this together and we're going to keep you posted." It can be that. It doesn't have to be "we have the answer." But it's so rare to see that being done well right now.

Mike: I'm curious, do you see in the leaders where you see that happening, is that a result of them not knowing they need to communicate that or a result of them not knowing what to do in the first place?

Paul: I think it's they don't understand AI because, again, think about all the things we talk about. Think about just even these first five trends. If you're a CEO and you're seeing this too, how could you be anything other than racing forward to solve for this? Because there's no way to look at what's going on in AI and realize it's not going to completely reinvent your industry and your company. And so if you truly understand AI's capabilities and you're using it yourself every day and feeling it each day, how could you not have a sense of urgency to tell your team that you're working on the plans and to go get those plans in place?

I think it is more just they haven't had that "aha" moment where they realize the significance of what's happening. I think a lot of times it was because they knew it was important and they've read the research, but they don't use it themselves necessarily every day. They don't feel it. And so they throw it to the CIO or the CTO or whoever and they're like, "Go figure this out. This is a technology problem." It's not. It's a business problem and it is going to change everything about the way your organization runs. And if it's not treated in that way, then it's not going to have a sense of urgency in other organizations. That's what we see a lot. You'll see these priority projects in a major enterprise where everyone knows, "Hey, what's the most important thing to the CEO right now? One, two, three." If someone asks that of your organization and an AI transformation isn't in the top three, you got a problem. I don't care how big the company is. So I think that's the issue; it's not being treated as a priority of the CEO. And until it's a CEO's priority, it doesn't diffuse across the organization.

Mike: All right, before we get into our final five trends, a quick announcement. This episode is also brought to you by our State of AI for Business report. On the day you are listening to this, you'll be listening to this podcast episode in the final week we are running our 2026 State of AI for Business survey. This survey is going to inform the report. It's an expansion of our popular State of Marketing AI report that we've done every year. And we're going beyond marketing-specific research to uncover how AI is being adopted and used across organizations. We are trying to survey thousands of business professionals across every industry and function. If you love the podcast, if you like what we've been doing, we would really, really appreciate it if you took this survey if you have not already. It takes about 5 to 7 minutes to complete. You can go to smarterx.ai/survey to share your input. In return for completing this, you will get a copy of the report when it drops plus a chance to win or extend a 12-month SmarterX AI Mastery membership. Go to smarterx.ai/survey. It is the last week to do this and contribute.

SaaSpocalypse

Mike: All right, trend number five: SaaSpocalypse. In early February, $300 billion was erased from software and data stocks in just two days after Anthropic announced legal and sales plugins for Claude. Stocks like LegalZoom dropped 20%. HubSpot year-to-date is down 39%. ServiceNow dropped 27%. The S&P software index alone lost 15% in January and the market called this SaaSpocalypse, a SaaS apocalypse.

The reason is because SaaS companies are caught in a bit of a crisis. These frontier models are releasing features that are eating into the core features of traditional SaaS companies. Tools like Claude Code are giving people the ability to code their own solutions. And it's clear frontier model companies and labs are going after not just US software spend, but also US white-collar wages because AI agents are increasingly able to just do work directly instead of you needing software in the hands of a human to do it.

Not to mention we've talked about in a past episode, SaaS companies are caught in a bit of a pricing crisis at the same time. The traditional per-seat models start breaking down when one person with AI can do the work of 10. If headcount drops, seat count will also drop. Credit-based pricing has emerged as an alternative, but companies are still working out how to price AI that replaces labor rather than augmenting a workflow. So some SaaS providers are racing to figure that out. Some are trying to become model agnostic. All of them are trying to stay relevant as these underlying models commoditize their features. Paul, where does this stand today? Obviously, the stock values and drops have changed since the initial SaaSpocalypse, but the core issues here, I don't think we figured out in the last two months.

Paul: I haven't seen any answers yet. I think it's just still more uncertainty. And that's what we talked about at the time—that Wall Street hates uncertainty. SaaS companies have been built through relatively predictable multiples. Their valuations are largely set on that. Their funding rounds are set on that. Their market cap is influenced by it. And so when all of a sudden you're like, "Well, okay, maybe in 5 years they won't be worth as much or the multiple won't be as high for software because people can build alternatives." Even though it's like, "Okay, well, people aren't going to necessarily spin up their own CRMs," it starts to create this doubt of, "Well, maybe some small businesses can or maybe they just don't need as many seats or maybe they're not willing to pay as much per seat."

Or you assume that if you're paying your $50 per month seat license, whatever it is, that your job is to make the software better for me. So why am I paying separately for the AI capabilities? I'm paying for the software to do a thing and the intelligence helps me do the thing faster. Is that my problem as a consumer that your costs went up? I'm paying for what I'm expecting from you. So it just creates all this complexity and I think a lot of software companies are just scrambling trying to solve for it.

I mentioned on a recent episode, I think we're going to see some turnover at the top of a lot of these software companies because it's going to be a difficult time to navigate and generally the markets aren't very patient. And so if you start to see these stocks staying down in this 30 to 50% range and there's no bounce back apparent, it starts to look more and more uncertain despite the fact that the revenues have been pretty strong. You're going to need to get somebody in there who's got a different vision for how to do this. And so I just think it's going to be a really challenging time for software companies and the people who invest in those software companies.

And then as a buyer, as a CEO of a company that buys the software, every time you think about what the future looks like, it's like, "Well, is the software we have going to get us there?" I'll give you just a prime example. The SDR thing I mentioned earlier about the open Claude. You look at what Claire presented in that podcast episode about the future of SDR. I sit here and think, "Well, is HubSpot going to enable that? That's our CRM. Or do I have to go get a third-party piece of software to do that?" The fact that I even have to stop and ask that question isn't great for software companies. Because I do that with everything we do. It's like, "All right, well, the piece of software we have now we're paying over a thousand a month for it and it doesn't do that. And that would be really valuable to me. What do we do?"

So I think that that is a microcosm of what's going to go on now. Once you understand what AI is capable of, you're just going to look differently at your tech stack and your monthly expenses tied to those. And what the value you're getting from them is, and if all you're getting in return is a credit-based model that you don't understand, you're going to get pretty annoyed pretty fast. And that's how you get motivated to go find something else.

Mike: As if it wasn't hard enough for SaaS companies, I feel like anecdotally I've heard in the last few months from several people where they're encountering sales reps at software companies that are not as equipped as you would hope to deal with some of these objections. Either "Why can't I use Claude Code to do this thing?" and they don't even know what you're talking about, or people using AI to do really robust research into competitors and the tech landscape that unfortunately sometimes sales people are not remotely equipped with the same type of research. And then you not only have a bad conversation, but come away saying, "Well, if you're not using AI for this stuff, how do I have confidence that you're using AI in your product?"

Paul: Yeah, if the buyer's got to do the job—and we can attest to this, Mike—the same happens on the customer success side. If you're doing the pre-work before you reach out to customer success through Claude or ChatGPT or whatever and then you get on a chat with a human at that software company or a phone call with them and you're like, "Dude, I'm doing your job for you. I'll tell you what doesn't work. I already tried these 10 things," and they're just looking up a knowledge base. They're like, "Oh, I don't know. Let me check the FAQ."

So yeah, I agree. There is this whole need to build an AI forward team at all levels of marketing, sales, success, product. Because you're going to end up dealing with more educated buyers than the people who are supposed to be in your company helping them solve things. What used to be the "Google it" is now "Did you build a strategy in Claude before you called them? Did you do all the things?" So yeah, it's going to be hard to work with those customers who are further ahead than your own people.

Labs Pivot to AI Agents

Mike: All right, trend number four as we count down is labs pivoting to AI agents. We really started to see in Q1 every major lab, especially starting in March, starting to pivot towards agentic capabilities and enterprise deployment simultaneously. This especially happened with the three frontier labs. OpenAI, for instance, has announced they're consolidating ChatGPT, their browser and Codex into hopefully a desktop super app. They're doubling headcount to approximately 8,000 as they target the enterprise and they are trying to build an autonomous AI research intern by September of this year. Anthropic has launched Claude Co-work, a more agentic system that's easier for non-technical knowledge workers to use. They are also just rushing it in the enterprise game in their fight against OpenAI in terms of signing enterprise licenses. We saw Microsoft restructuring Co-pilot under Satya Nadella's direct oversight as they are trying to find their footing as well.

And we've seen over the last few months all these different types of agentic releases. OpenAI has dedicated agent products. They've been working on a frontier program to partner with companies and also some PE firms to get in with different companies and portfolio companies of those firms. Microsoft shipped Co-pilot Co-work. We even saw in the open source front Andre Karpathy released an auto research agent. So all this agentic stuff is hitting at the exact same time the labs are not only doubling down on it, but also doubling down on trying to get into and expand within enterprises. And we've kind of seen this anecdotally as well just on the podcast as we've conversed about everything agentic, right? We've talked about the timeline to agents, managing the chaos of agents, agent swarms and of course the security nightmares that come with agents. Paul, it seems like all agents all the time and getting those enterprises to sign on the dotted line is the strategy of the labs right now.

Paul: Last year, 2025 was definitely the year of agent hype, I would say. We dealt a lot last year with overpromising from some of these tech companies about agentic capabilities, but you could see the beginnings. It's almost like where we're at with open Claude now. It's probably a little bit overhyped at the moment, but the reality is going to start to set in as the year goes on. And so we knew coming into this year, again, agents aren't new. It's been talked about for a decade. We've talked about it extensively. I built it into my AI timeline, the stages of AI that we talked about in episode 207 and many times before that from OpenAI has agents as level three. So chatbots one, reasoners two, agents three and then innovators and then organizations at four and five.

So agents aren't a new concept, but they are definitely starting to have their moment as they become more autonomous and more reliable in different use cases. As we're talking about this, I was doing quick searches and I can confirm now what I said earlier. This is tied to this: Anthropic's Claude Code command line interface application, not the models themselves, has been leaked and disseminated. This is from Ars Technica I'm reading. Apparently thanks to a serious internal error, the leak gives competitors and armchair enthusiasts a detailed blueprint for how Claude Code works. A significant setback for a company that has seen explosive user growth and industry impact over the past several months. Early this morning Anthropic published version 2.1.88 of Claude Code NPM package. But it was quickly discovered that package included a source map file which could be used to access the entirety of Claude Code's source, almost 2,000 TypeScript files and more than 512,000 lines of code.

A researcher was the first to publish to point it out on X with a link to an archive containing the files. The code base was then put into a public GitHub repository and has been forked tens of thousands of times. So keeping in mind we're doing this at 3:00 p.m. that day, Anthropic publicly acknowledged the mistake in a statement to VentureBeat and other outlets which reads: "Earlier today a Claude Code release included some internal source code. No sensitive customer data or credentials were involved or exposed. This was a release packaging issue caused by human error, not a security breach. We're rolling out measures to prevent this from happening again."

So man, bad couple days for Anthropic getting some things put out into the world that should not have been put out into the world. And that one, I think that one's probably pretty significant. There's a whole lot of people who would like to access that kind of stuff and they just got it for free. The other thing this illuminates separately, we'll talk about this in a future topic, is the weights of these—the model isn't what got leaked, but Anthropic has been more forthright than anyone about the significance of who has access to the weights of the models. And there was an interview Dario Amodei did probably a year, year and a half ago where he said at Anthropic, there's literally like three people who have access to the weights. It's hidden from everybody. And said that's the thing that foreign adversaries want to get to. There's been billions of dollars to try and get the weights from these models. And it's like, how good are the guardrails if that's the future? You're going to build this insanely powerful thing, like the mythos model or whatever's coming next, and all that's preventing the world from knowing how to replicate that is figuring out how to get to the three people who have access to those weights. That's terrifying. Twice in 48 hours leaking things that shouldn't have been leaked. It's a weird age we're heading into.

Mike: I assume this happens, I just haven't read about it, but you have to imagine some of the higher-ups, not even the CEOs of these companies, have to be walking around with some serious security.

Paul: Yeah, it's like the nuclear codes basically. Talk about your senior engineers or something even. Not even Dario Amodei, I assume he's got his same moment, but higher-up employees would be pretty—well, this is why there's part of it is memes, part of it not. This is how counterintelligence stuff works. This is how you do espionage and stuff. So, yeah, what's the most valuable thing right now? It would be hard to find things, at least in the United States, that have a higher value than the weights of these Frontier models. So, espionage, I would imagine there's rather significant background checks. There's probably a lot of internal security monitoring who the top employees are spending time with and friends with. It sounds like a sci-fi movie. I can promise you that that stuff is happening. That is a very, very real thing, and those are high-value human targets that you will do anything you can to get access to what they know.

Mike: Again, we need a series. We need a Netflix series on this. It's probably, as much as we talk about all the branches of AI and all these intriguing things, it is likely infinitely more intriguing than we even know. And I use intriguing as a word carrying a lot of weight, both good and bad intriguing. There's probably so many more layers to what is going on in AI that'll make for such fascinating—I'm not sure Hollywood could do justice probably to what's actually going on in the AI world right now.

AI-Driven Layoffs Go Mainstream

Mike: All right, trend number three. We are talking about AI-driven layoffs going mainstream. We haven't seen wide-scale AI-driven layoffs yet, but we have seen a lot more chatter and conversation around this and people starting to actually attribute AI behind some of the layoffs that are happening.

For instance, we saw tech company Atlassian cut 1,600 employees, 10% of its workforce. They explicitly attributed this this quarter to their transition to the AI era. They were one of the first major companies to really name AI directly. Block, Jack Dorsey's company, cut approximately 4,000 employees, nearly half its workforce, and talked quite a bit about how AI was making them more efficient. Their stock surged on that announcement. And just recently, we've heard from Uber's CEO talking on the Diary of a CEO podcast and saying that executives privately admit the true scale of AI disruption, even though they are going on TV and telling audiences everything will work out fine. Uber's CEO personally estimated AI will replace the work of 70 to 80% of humans within the decade. He has no idea what's going to happen to Uber's 9.5 million drivers in that era, either. The same week, PwC's US CEO told the Financial Times that employees who think they can opt out of AI are "not going to be here that long."

So, Paul, we've still seen several thousand layoffs related to AI. We've talked about how we expect those to rise. But I think the bigger thing here really is CEOs are publicly breaking the silence and saying, "Look, AI is going to be a factor here." Is that kind of what you're seeing and hearing?

Paul: Yeah, this is a trend I wish would go away, but unfortunately, I think this is going to—I mean, it can't move too much higher up the list than number three, but I expect this trend to continue and to gain steam, unfortunately. Both the unemployment and the underemployment, as we get more data around that.

There was a post this morning. This is Heather Long, chief economist, Navy Federal. She tweeted: "US hiring rate fell to 3.1% in February, the lowest since April 2020, which was mid-COVID. This is a hiring recession, and Americans are feeling it. There were notable hiring pullbacks in February in hospitality and construction, which was—the construction and healthcare has actually been holding the market up a bit. Bottom line, the job market was already frozen before the war in Iran began. It's worrying that a no-hire, no-fire situation could turn into a no-hire, start-to-fire job market quickly if there isn't a resolution soon."

Now, that is not AI-specific. There's nothing in there where she was saying this is because of AI, but it's simply pointing out what I've said on the podcast numerous times, which is what I'm hearing: that no-hire, no-fire, like we are not adding anybody. We're going to try and avoid firing, but we are pausing hiring, and the only new hires will be through attrition when we need to replace people, but flat growth is sort of like the desired state right now. And as I've said before, I don't know a CEO who wants to fire 20% of their staff. I've yet to meet that person. I think generally speaking, leaders of companies want to create opportunities for humans, and the idea of human employment and that being a driver of the economy—that's pretty fundamental to our democracy working, and it's pretty important that it continues.

But there's going to be tremendous financial pressure on leaders to take action and to capture some of the efficiency gains and profits. And that's going to lead to some very challenging periods here. And so, this is an area where we're thinking a lot about—I was actually just talking with Mike and Taylor on our team on the research front to really start leaning more into this and trying to do more research around what is happening on the frontiers. What is being talked about, what can we be doing? So, it's not just us showing up each week being like, "It's getting worse. Another 50,000 people lost their job." We want to try and contribute to the dialogue at least of finding answers. I don't have the answers. I have some theories of things I'm working on myself, but I think we collectively need to just be exploring ideas here, putting think tanks together of groups of people that you trust, you can bounce ideas around. We just need to be talking more about answers because it's not coming from the labs who are building the tech and creating this eventual uncertainty and chaos. So, really, really important trend. I wish it would go away. It's not going to. So, we got to do something about it.

We’re Seeing More Move 37 Moments

Mike: All right, trend number two. Before we hit our top trend this quarter, number two is we're seeing more of what we call move 37 moments. We track what we call move 37 moments on the podcast. This is this point where a professional in a given field realizes first-hand that AI can match or exceed their expertise. This term comes from AlphaGo's move 37 against Lee Sedol in 2016. This was the move that made the world's best Go player realize the machine had surpassed him, and we're starting to see a few more glimmers of these moves out in the wild.

In February, we actually dedicated an entire segment to this phenomenon. Sam Altman recently noted that OpenAI's Codex coding tools had suggested features superior to his own team's ideas. Dropbox's former CTO declared that he'll never ever write code by hand again.

Goldman Sachs has begun deploying Claude for trade accounting. KPMG is pressuring people to cut audit fees because AI can do it instead. David Kipping, an astrophysicist, reported that AI had about 90% of the intellectual capability that he was seeing in his field. In March, a Polish mathematician reported his own move 37 moment after GPT-5.4 helped solve a problem that had resisted conventional approaches.

Boris Cherny, creator of Claude Code, declared on Lenny's podcast that coding is effectively solved. And we also talked about this New York Times AI writing quiz that 86,000 people took, where 54% of them preferred AI-written passages over the work of famous authors. So, some glimmers here, Paul, that the trend is pointing to this list of fields where humans hold this unambiguous advantage seems to be getting shorter every quarter. Can you tell us a little bit about why move 37 moments are important? It seems like we're seeing more of them. Do you agree with that?

Paul: Yeah, this was the premise of my MAICON keynote in 2025. In essence, what I was seeing was, for the most part, AlphaGo—which is an incredible documentary that changed my perspective on AI and really the future—it was always talked about as a technical breakthrough of the technology capabilities of this AlphaGo system. And what I challenged people at MAICON to think about was the human side of it. Like what happened to Lee Sedol in that moment when he realized the machine was better than him at the game he was an expert in.

That was my premise at that time: that we would all come to experience that Lee Sedol moment where you just say, "Wow, it's just better than me at this thing." And then, what do we do from there? It was probably the most challenging keynote I've ever given because up until 24 hours before I gave it, I actually didn't know the ending of the talk. It was the start of our conference, so I didn't want everybody feeling defeated and like, "Oh, well, let's just go home." I wanted to take people through excerpts from the documentary and hopefully have that emotional impact on people—have that somewhat of a gut punch feeling like Sedol had—but then to turn it into something about how we still have choice. We can still do something about this, and we can figure out how to use these as tools that give us new abilities and a different way to look at business and our own careers.

I think that's what more people are going to come to grips with. I don't see this slowing down. I think this is just a reality, and pretending like it's not coming isn't going to do anybody any good. You and I each have these conversations all the time where it's like, "Well, it can't do what I do. Yeah, I get that it's good, but it can never do what I do." And it's like, "Yeah, okay, that's probably not going to end well for you, but I understand." You do have to have these moments where you decide when you can push someone. It could be a friend, a family member, a coworker, or a boss. If you listen to this podcast, you're probably in the know about what these things are capable of and where they're going.

And you look around the rest of the world and they're blissfully unaware. We had our dads' basketball tournament this past weekend. A buddy of mine who probably listens to the podcast is messing with Open Claw all the time. He's on all this crazy stuff. We're sitting at the bar Saturday night after the basketball tournament ended—it's literally just a bunch of dads playing basketball for two days, it's great—but we're talking about what he's doing with AI and with Open Claw. And then you have that moment where you look around the room and there's just hundreds of couples there, and you're like, "Damn, they have no idea." And not in an "I feel bad for them" way, more of a feeling that you are just living in this parallel universe where you're seeing the future and you're realizing they all have careers and families and colleges to pay for and kids to raise, and they have no concept of what is going on.

There's a part of me that's envious of that, honestly. The ignorance to the moment is actually something I sometimes wish I had. I think anybody who has the knowledge you have has those moments where you're like, "God, I wish I just didn't know what I know." I wouldn't be worrying so much about jobs every day and the future of education and all these things, but once you know it, you can't turn it off. I don't know if the move 37 moments trigger that for people where you have that realization like, "Oh my god, it can do what I do." And then everything is different from that moment on. You just start to look at all of it differently.

It's a really important thing. We'll drop the link to my keynote in the show notes. If you haven't watched it, we put the whole thing on YouTube. I've given thousands of talks now in my life, and that was the second hardest talk I've ever done. There's one other talk I did at MAICON that was the hardest I ever did, and I'm not saying hard in terms of technically hard, just personally hard. That was a tough one to keep my composure on stage because I knew the punchline I was going for and I was having a hard time getting to it because there was no turning back kind of moment for me. It's worth the watch if you haven't seen it.

The Vibe Shift: AGI Enters the Public Discourse

Mike: All right. So, our final trend that we have been tracking this past quarter is what we're calling the "vibe shift." This is the quarter where the conversation around AGI really entered the public discourse. It entered the boardroom, the newsroom, and the living room. Despite many people still being very early in their own bubble, we started to hear about this everywhere.

The single piece of content that captured this shift was probably Matt Shumer's essay, Something Big Is Happening. This was viewed 85 million times on X. In roughly 5,000 words, Shumer, who is an AI CEO and founder, wrote about what many insiders had been thinking but not saying publicly. He said he has historically at parties given the polite version of where AI is going because the honest version sounds like he's lost his mind. He goes on to detail how we're in this moment of possibly fast AI takeoff that feels a lot, in his analogy, like February 2020 right before COVID struck, where a few people were seeing signals that the world was about to change.

We've talked about this in a couple of other contexts. Episode 189, which started this year, had a segment called "How Close Are We to AGI?" because basically Claude Opus 4.5 over Christmas break was demonstrating some really wild capabilities, especially when paired with Claude Code. We even had a Google principal engineer saying Claude completed a year's work in one hour. The audience response to our episodes about whether we are at this tipping point is unlike anything we've ever seen. Listeners have also been seeing this turning point where something changed at the end of last year and the beginning of this year in terms of AI capabilities and what's now possible, especially to non-technical knowledge workers. Paul, how big of a moment are we in?

Paul: When we flipped the calendar to 2026, you could just feel it. Something had changed over that winter break. The online dialogue was just different between the people who were building things specifically with Claude. One of the best ways we keep a pulse on what's going on is by the questions we get from audiences. It's one of the luxuries I have of teaching the Intro to AI and Scaling AI classes for free every month. We have 2,000 to 2,500 people a month go through these classes and we take questions live. We are getting hundreds of questions a month, in addition to the speaking engagements and executive briefings. You can just feel the difference based on what people are asking about and the stories they're telling of their own experimentations.

It is very different than it was three months ago. We did an AI in CLE event just last week, Mike, and there were 120 or 150 people registered for it. The questions we got there—everyone wanted to talk about how they're using Claude Co-work, what apps they're building with no-code, messing with Open Claw, questions about the environment, and political questions. The dialogue has just moved. It is so far ahead, but even then you have to keep it in context. I would say the people who are in the know are just moving further and further ahead. They're experimenting on the frontiers.

It's easy to do what we do and get caught up in that bubble where everyone's moved on and everyone's ready to talk about Claude Co-work and Open Claw. Then you go spend time with a bank, a health care system, a manufacturing company, or a school, and you realize they don't know anything. If they're using a chatbot, it's likely a base version that doesn't even have all the capabilities built into it. They're oblivious to all the capabilities. I think the "haves and the have-nots" is a way to say it with AI. The gap is expanding dramatically. Over time, that's going to start to expand into the outcomes and benefits as well, where the distribution of those benefits is going to be heavily weighted towards those early movers and the people who are actually out figuring this out. They're going to get compounding value while these other people are being left behind. I don't want that to happen.

We feel this vibe shift every day. I feel a greater sense of urgency every day to do more because I see so many people who aren't aware yet or don't have a sense of urgency to solve for it in their own lives and their own companies. That's going to be challenging to see.

Closing Thoughts and Future Outlook

Mike: All right, Paul, that wraps up ten trends for Q1. It's been a wild start to 2026. This is actually really good timing because I feel like this is a good breath and a recap before the storm that's going to happen in the next few weeks with model releases. I think we're in for a very fast spring and summer.

Paul: Yeah. And a quick show note: in episode 207, we were talking about Peter Steinberger and Open Claw. I was sharing that I'd listened to the Lex Fridman podcast episode, which came out on February 11th, but I didn't listen to it until March 30th. I mentioned at the time that he was talking to Sam Altman and Mark Zuckerberg, but then when you said the thing about him going to OpenAI, I remembered he did go to OpenAI. I'd mentioned he might go to Meta or whatever, but Steinberger published a post on February 14th, three days after the Lex Fridman podcast, saying, "I'm joining OpenAI to work on bringing agents to everyone. We'll move on or move to a foundation and stay open and independent." He's got a blog post we'll throw in. In that episode with Fridman, he talked about how he was going to basically go work for one of those two. He did end up going to OpenAI and moving Open Claw to more of a foundation model.

Hopefully this trends format was super helpful to people. It's helpful to us. It's always one of my favorite things Mike and I get to do each quarter—to step back and think, "Holy cow, how did that all happen in three months?" Even just the tenth trend with all the models is hard enough to comprehend. We always inevitably get questions like, "Well, what about this? What about this?" Trust us, we know. There are 20 other things that could have made the top ten. We only have so much time in the day to go through each of these things.

So thanks, Mike, for putting this all together. These are great. Hopefully we'll make this a recurring show. We'll probably do it as a bonus episode moving forward, unless we have a week where we're on vacation. We'll start doing these as a special quarterly episode and we'll be back April 14th with the next weekly episode. We've already had a lot happen in the first two days of this week, so I imagine by then we're going to have 100 links to get through.

Mike: Oh my gosh. Well, have a great Easter holiday and trip if you're taking spring break anywhere. If you celebrate Easter, enjoy the time with your family. That's what I'm planning on doing and hopefully not working too much. I have a long flight and I can't sleep on flights, so I'm going to be super productive for 20 hours. Other than that, I'm going to try and just enjoy time with my family. Thanks for listening. We'll be back with you again soon.

Key Takeaways