Your CEO Has AI Psychosis and Doesn't Know It

Well, that's why I think all these conversations about taste and judgment have come back, because in a world where you can build anything, I agree with the core premise. Deciding what to build and what to work on is really, really, really important. And ironically, I think in a world where building is cheap, planning is actually far more valuable. Everyone's like, oh, it's so cheap to build. I'm like, yeah, but once you've built something, fixing it doesn't feel cheap.

It's the week of May 11th. Welcome to the Learning Corner, a Precursor Ventures podcast. Welcome to the Learning Corner by Precursor Ventures. I'm Mia Farnam, and I'm joined by our GP and managing partner Charles Hudson, where we will be talking through the articles and good reads that caught our team's attention over the past week. Our first piece is a post on X from the account Signal Ventures that has been circulating across platform, and it's a direct take on networking culture, specifically for startups and venture. The argument is that traditional networking, the conference circuit, the warm intro economy, the grab coffee culture. Is mostly negative selection. The logic is that the people actually worth knowing are too busy doing the work to be formable, and the people who are consistently available to be networked with are often available precisely because they don't have that much else going on. The post makes the case for a completely different orientation: do the work, publish it loudly enough that the right people find you rather than chasing them. One-way broadcast over two-way schmoozing, and specifically calls out X as one of the most important platforms right now to be doing this. This one is super interesting because venture is one of those industries most historically synonymous with networking as a core function of the job. You've got warm intros, you have LP conversations, founder referrals. Obviously this is different when you're a founder, but how much of this did you resonate with, Charles, and agree with? And then is there a balance that you can actually strike here?

Yeah, I definitely have talked to some of our portfolio company founders. I mean, I think the— once upon a time there was this mind virus going through YC, which was like, you should never do anything other than work on your startup. You should never go to an event. You should never— I'm paraphrasing how it was described to me. This idea that anything that was like networking on your startup was a waste of your time. And I do think that if you're a new founder, it can be really attractive because you get invited to all these events and there's so many things you could go to. They all feel productive. You're like, oh, I'm going to go. But I always ask people, what, what's your goal in going to this event? Are you trying to meet a customer? Are you trying to build a relationship? And I read the article and I was like, well, this doesn't seem like an argument against networking. It seems to be an argument that networking on X is more effective than networking in the real world. I think it depends on who your audience is. If your audience is X-pilled VCs, there's probably no better use of your time than posting and sharing and engaging on X because that's where your audience is. If you're trying to become an enterprise software thought leader, I might argue your time is maybe better spent on LinkedIn because your audience is there. And so I think when it comes to networking, the, the useful part of this article is some events are just low value and it's worth questioning like, why are you attending? And it's also the reason I try to keep the programming that we do for our portfolio light. At least at the founder and CEO level, we don't do a lot of events because I think people are busy. And I think the best portfolio company CEOs I work with, they're always evaluating time requests. Is this time that I could spend at this event? The irony is everyone's like, you should go to no events and network, but you should spend a ton of time with your customers. Well, where do the customers come from?

Where do they hang out?

Where do they hang out? Where do they come from? Is it possible that spending time in person with your customers might actually yield more customers? Of course it is. Is it also possible to go into some fourth-tier networking event in your city is a complete and total waste of time. Yes. But I do think what this post highlights is there are certain platforms that are far more important to certain people than others. So I do think X is a platform that is very important for a lot of VCs and a lot of founders. And I think about someone like Nicole Wishoff. You would argue that she basically X/tweeted her way into the venture business and she's got two funds under management and she's got a big following on X. And I think it's cool that there are platforms where people can start from scratch and build followings. And I think I know other people who are really good at going to the 4 or 5 events in their industry that really matter and crushing those events and building tons of relationships. And so I think networking done right is one of these things where you never know when you're going to be in the room with somebody that you really need to meet. Like, I met one of our most important LP relationships walking out of an event that had been pretty good but not amazing. Highest ROI thing I've ever done for the firm on the LP side, but not something that I thought would be that. So I think nobody would go to low-value networking events if they knew they were low-value, but sometimes you can't know.

Would you make any specific recommendation if founders know that they have like a fundraise in their near future? What would you say to those folks if they're like, should I start going to more events? Should I not go to more events to create this stealthy FOMO? What is the formula there?

I think it's interesting, as great as X is, there's a lot of really important information, mostly backchannel, side channel stuff about startups that isn't shared on X. Yeah. You can only get. By sitting down with somebody over coffee or on text or whatever. And I think people's willingness to share that information with you is also a function of the quality of your relationship. And not all of those relationships have to be built in person. Some can be built online, but it's still work. I know lots of people who think they're smart and have great content and publish it, and it feels like they're yelling into the void. So in the same way that being good at networking is about deciding which events to go to and figuring out how you want to show up and who you want to talk to, it's not just randomly posting on X is the strategy that most successful people I see do either.

They're thoughtful and they're making a time investment in what they post and how they post The platforms have become so noisy just because you can put together even like good content doesn't mean it's gonna be the substitute if what you're building doesn't work. So being hopeful that content will bring you these followers when you have maybe nothing going on in the background, I think is a hard argument. But I think your advice on what is your desired result and what does a successful outcome look like? I think that actual deep thinking makes a lot of sense cuz I think people just run headfirst into creating content and putting it out there without knowing like, what am I doing this for? Our next read this week is an essay by Alberto Romero from from his Substack, The Algorithmic Bridge, titled "You Spent Your Whole Life Getting Good at the Wrong Thing," Alberto's core argument is that AI is collapsing the cost of the how for knowledge workers so dramatically that the what, as in what you actually want to build, create, or accomplish, is now the bottleneck. He uses the image of a genie lamp to make this concrete. If a genie handles the how completely, you quickly realize that wishing well is actually the hard part. And that most of us have never really practiced it. The piece breaks down the specific skills that become load-bearing in this new world. It's taste, it's judgment, agency, curiosity, and decision-making, and argues that these traits were always important, but they were just buried under the weight of execution. He closes with a real and honest observation that the reason people resist this shift isn't ignorance. It's that our entire sense of self-worth is built around doing, and when doing becomes cheap, it can feel like you've become worthless. Alberto draws this distinction between the what and the how of work, and the argument is that most of us have spent our entire careers sharpening the how without ever deeply developing the what. Wanted your reflections on this one, Charles.

It's so funny. I was in a board meeting this week and this CEO and VP of engineering were giving us an AI engineering update, which I feel like has become standard practice in every board meeting. And, and the engineer was telling us how things that used to take 4 weeks now take 4 hours. And I was like, that's great, but how do you know what to build? And he's like, what do you mean? I go, well, you can ship anything. So how do you decide what to ship? And more importantly, how do you decide what not to ship? And he's like, yeah, this has become a problem. You know, I think as much as we criticize the old development frameworks, they had one benefit, which was if a customer emailed you and said, Mia, I want a purple button, you wouldn't just make a purple button. You'd say, well, why do you want the purple button?

Yeah.

What are you trying to do with purple button? What problem are you trying to solve? Why can't you do it with the green button we already have? Because building was expensive, you did some amount of planning upfront and some amount of investigation to make sure you were actually solving the problem. Mm-hmm. And I hear from so many startups, oh, everything customers want, we just put in the product now. And I'm like, well, that was never, it was never possible before. So you had to have this like prioritization and forced ranking thing. And even if it had been possible, I think most people would say, well, it's not a good idea because your customers in telling you what they want aren't necessarily telling you what to build. They're telling you, they're describing a problem that they want solved. And he talks a lot about like what, you know, what, what product managers do. And I think there's a lot of translation that happened in companies and you were able to do this because you had to move slowly. And now we're in a world where I think people have over-optimized for like, well, you can do anything, so you should just do everything. Well, that's why I think all these conversations about taste and judgment have come back because in a world where you can build anything, I agree with the core premise, deciding what to build and what to work on is really, really, really important. And ironically, I think in a world where building is cheap, planning is actually far more valuable. You know, I build a lot of things in cloud code. Everyone's like, oh, it's so cheap to build. I'm like, yeah, but once you've built something, fixing it doesn't feel cheap. Turns out like we're all becoming software developers. And I think this has been my theme of the last quarter. Every, if everyone's becoming a software developer, we have to all learn many of the best practices that other people learned about planning and code review and being thoughtful and giving yourself time to think about the full range of what you want to build, as opposed to just taking a vague prompt and slamming it into Cloud Code and getting started. And I just think it's highly ironic to me that we're all learning what my friends at Build Building Software have known for decades, but we're learning it under a very compressed timeframe.

A couple years ago, definitely within healthcare, one question I would always ask folks when they were in the initial pilot conversations with customers that seem like they were gonna be high value, I would always ask them, you know, how can you prevent from becoming a dev shop? Yeah. And I actually really liked that conversation because it showcased how thoughtful the team was on where they wanna prioritize time. I still ask this question too, if you're working with your initial customers, how do you decide what things they're requesting are actually worth dedicating resources and, and time towards? I think where we are today, to your exact point, that conversation you could remove altogether if you're a founder who says, I can become a low-cost dev shop, my customers are happy. So like, what's the problem? With that, I mean, our next piece that we're going to talk about, there are all these negative consequences of just bloat. If you've just decided to optimize for saying yes to everything and make a purple button just for the sake of doing so and making someone happy, I don't necessarily think long-term that results in being very thoughtful of who you're trying to build for and what your pain point is. You just become an agency business essentially, and like outsourcing all of your capabilities. I do believe in this concept of the taste, the agency, those things are really coming forward and being held at a premium now. I believe that.

Also, I just think there's many products that I use that are surprising and delightful because the person building them had a point of view about what I'd want before I expressed it. I also worry that the other challenge is if you're just like, well, we're just going to build what our customers want. What about the unexpressed, unarticulated needs that they have that are your ideas that they haven't yet expressed? If all you're doing is building the stuff that they tell you to build, how do you actually build something really interesting that, that's a breakthrough? So there's a tension there for sure.

My last thought there is if that's all you offer is just kind of saying yes to all their requests. I mean, if somebody else just comes in and starts doing that as well, it's just a race to the bottom on price. Like you don't have any specific perspective. I know that people really do value, oh, I'm hearing what you're saying. Maybe what you're asking for actually could come to life in this way. Is that something that's interesting to you? And it's the thought partnership that actually is the longest lasting piece of that relationship. But it's really tempting to just say yes now because it is possible. It's just what's most important in the short term and the long term and balancing that all. Our last piece this week is from Jake Handy's Substack, Handy AI, titled Your CEO is Suffering from AI Psychosis. JAKE opens by positioning himself as a genuine power user. So this isn't a skeptic writing from the outside. His core argument is that a specific kind of brain rot is spreading through executive suites and VC circles right now. One that looks like productivity and sounds like innovation, but produces almost nothing of measurable value. He points to Gary Tan shipping 37,000 lines of code per day across 5 projects while running YC full-time, and his CTO at YC even claimed this was like God Prod mode code that a developer actually later audited and found riddled with bloat, empty files, and uncompressed assets. He then layers in a Stanford study finding that AI models affirm users' actions 49% more than humans do, even when those actions involve deception or harm, creating this loop where the more your AI tells you you're doing great, the more you trust it and the less you check the actual results. The piece closes by advising people to define what done looks like before you start the agent, measure output, not activity, and audit your fleet, because if you can't account for what every agent is doing and what it produced, you have a shadow AI problem. Jake makes this distinction between the feeling of orchestration and the reality of output. And the data he cites, we've talked about before, like a long time ago, 90% of firms reported zero measurable productivity impact from AI over the past 3 years. One thing he digs into is how the allure of spinning up large agent workflows is that it gives you this illusion that you're running your own massive organization, and that can be lethal when it gives the CEO a false sense of what's truly moving forward within a company. How can you reroute as an organization if these things like token leaderboards are actually taking up space within your workspace, where you actually have to just see how it all turns out in order to reroute and change direction thereafter?

I read this and I was like, I know a lot of these people. I hope I'm not one of these people, but I know a lot of these people. And I think part of what is happening, especially at CA— I don't know about Gary's got a different He's managing a very different type of organization. But I think for a lot of CEOs, the problem has been if you're like, hey, I wanna know what's happening with our 10 biggest customers in Europe, you would have to like call your VP of marketing or go on Slack and be like, hey, I want this information. Then you and your VP of marketing would have a meeting and VP of marketing would be like, well, what do you wanna know about these customers? And you're like, I think I wanna know net revenue retention. I wanna know who's a churn risk. You got like 10 questions and you'd give it to your VP of marketing and then they would go talk to the director of marketing and tell the director of marketing, drop what you're doing. The CEO asked us to do this project. And then 2 weeks later you get the project back as a CEO and you get a presentation and you're like, well, now I have 10 more questions that I should have asked you up front. And that cycle continues. And as a CEO, you're like, well, this stinks. It's a 2-week turnaround because I don't have a login to Salesforce and to our billing system. I don't have logins to all the systems. Even if I had logins, I don't know how to query the systems and do this myself. And now you've cut all of that out with Cloud Code or Codex or Gemini. Oh, I've got MCP servers. I can connect to all of these things. I can express natural language prompts. I can get to the thing I wanted. You're also just like, every time I tell the VP of marketing, they understood 90% of what I told them. And the director of marketing understood 90% of what the VP of marketing told them. And the poor marketing manager who's pulling the data understood 90% of what they were told. So now a third of what I intended has probably been lost in translation. And I think as a CEO, it's frustrating. I waited for 2 weeks and the thing I got back was not what I wanted. If only I could just go to the data myself and touch it, I could answer my own questions all the time. I see a lot of CEOs who have a deep desire for that, not because they don't trust the people that they work with, it's that, that human iteration cycle. Because the VP of marketing doesn't want to put something in front of the CEO that isn't perfect. Claude doesn't care. Claude will give you an answer. The answer might be wrong, but you will get an answer in—

Claude, he checkmarks everything.

You'll get an answer in 45 seconds or a minute or 2 minutes. How do you compete? As a CEO, that's like way too tempting to be like, I could get an answer in 2 hours that's statistically probably right, or I could wait for 2 weeks and get something that's maybe no better or maybe actually worse. And I think it's created this desire to solve all of these time-to-value problems that you have as a— and the higher you're up in the, in the company, the more communication layers there are between what you ask for and what you get, and the the longer it takes for it to come back up the chain, which is why I think you see so many CEOs who are AI-pilled. I think the challenge is with agents, you can just build them. And, you know, CEOs spend a lot of time talking to other CEOs. I can tell you, me and no CEO or VC I know wants to go into a meeting and be like, oh, I'm not running any agents. People be like, well, but what are you even doing? You're not running any agents?

None?

Zero? So everyone's got it. You got to have an agent. I'd made an old man analogy to one of my friends. I was like, I remember when people first got BlackBerrys. You don't have a BlackBerry? Are you like not important? Does your company not think you're important? And it was sort of like, oh, I don't want email on my hip all the time. Well, if you're an important person, you have a BlackBerry. That's just like the way it works. And I think AI agents have become the same way. And so I've actually decommissioned a few agents because they don't serve me anymore. I've like found better ways to solve those problems. I also now write pretty detailed specs about what problem I'm actually trying to solve with the agent, how I'll know that I'm solving it, how much time I actually think I'm gonna save by doing this, what my objective is. And sometimes I write the spec and I'm like, this isn't really a good use of my time. This isn't actually, not that I'm, not this actually isn't that big of a problem. Yes, I could make it agentic, but why?

Do you think that becoming almost in this cyberpsychosis, which is what Gary coined it as, being so lost in the sauce on all the value you can extract from agents and spending time dedicating to build there, is the natural progression to getting smart on them so that at the end of it you can decide what the value is that you specifically can extract from it? Or is there another way that you can get smart on this without becoming so lost in, I don't know, ecosystem that's like saying this is the best use of of your time, and if you're not doing this, then what are you doing?

I think there's two different things. One, I've told one of my friends who's getting into agents, I'm like, be careful. This is productive doom scrolling. It will make you feel like you're doing something, but you might be avoiding some other work you should really be working. I'm guilty of that sometimes. I'm, I'm speaking from personal experience. The other thing is, I think this feels to me like this is more of a tech thing than a general CEO thing. When I talk to my friends in tech, I think AI is impacting their engineering organizations and their businesses in such a profound way. It feels almost irresponsible to not have firsthand experience with this stuff so that you understand the power of it so that you can, when people are resistant, you can push them harder. I have some friends who are in much bigger Fortune 500 companies where that agent development is not being done by executives. It's being done by centralized teams and they're observing it, but they're not the ones up at 2 in the morning doing pull requests and GitHub. Like they're not doing that. I think the like hands-on Gary Tan thing is very much a CEO, a tech CEO and VC thing. Predominant, not exclusively predominant, definitely.

Talk to a friend who works at a larger fund. They have a whole IR system set up. They're in the middle of the fundraise, and they were sending out a DD request back to, uh, an investor. And last stop, after like everyone had reviewed it, somebody on the actual investment team was like, can I take a look at the DD form? Takes a look at it, completely made up numbers throughout, and was like, where are you getting these numbers? But we talked to everyone, it seemed like everyone was vetted off. Email chains floated back and forth, and it was very clear that somebody had just put in specific reporting metrics into Claude and then put in the DD request and then was like, match it, make it work. And they caught it before it went out, but it said, check, check, check. Everything's done. Everything's filled. That's an example of, I would say, no one's owning how deep you actually need to get in terms of architecting what an agent looks like. We're staying at the surface level. And that actually, I think, happens a lot at larger organizations. And I think actually that's where a lot of the mistakes are going to be occurring. Nobody's taking ownership over what that task looks like.

Also, in relatively flat organizations, it's easy. At Precursor Ventures, if somebody gives me a work product, I can look at it and be like, that's right, or that's just intuitively I can look at it like, that's not right. I'm one level removed from the data. My fear is that there are some big companies where if you're the CEO of some— I'm making this up— an airline company and something lands on your desk and your team's like, we all vetted this, this is accurate, you're like, it should be, there's a lot of money that was spent before this got to me. You know, in the old days, you know, people hand-reviewed things to make sure there weren't mistakes, whether it was a presentation going to a client or a spreadsheet that was going to your boss. And now I think people talk about all the laziness of thinking. I don't worry about that as much. I worry the laziness of review, because I get work products sometimes, not from anybody on our team, where I'm just like, even a cursory scan of this would have like given you cause for alarm, would have caused you to think, is this, is this correct? And it takes time and it like undermines the AI makes everything cheap and fast if you still have to do human review in the end.

Well, we are all out of time today. Thank you, Charles, for jumping in. Thank you to our listeners for hopping in as well. Please like and subscribe and we will talk with you all next week. Thank you.

Bye.