What Happens to Your AI Agents When You Quit Your Job?

And I think this notion of AI adoption having levels is true, but I think the tricky thing is it seems to change every 2 quarters. So I tend to think of what is the top 25% of companies doing versus what's the average company doing. A few quarters ago, having a GitHub repository with a bunch of shared skills on a Teams account maybe would've put you in the top. And I think every quarter it feels like the things that were top tier advanced become average.

It's the week of May 4th. Welcome to the Learning Corner, a Precursor Ventures podcast. Welcome to the Learning Corner by Precursor. 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 Goodreads that caught our team's attention over the past week. Our first piece this week is from Forbes, written by Nirat Cohen, titled, Can You Take Your AI Agents With You When You Leave a Job? It explores a question that's becoming one of the more interesting and unresolved tensions in the modern workplace. Of course, previously the deal was quite simple. Companies provided the tools, the data, and the infrastructure, and when you left, All of it stayed behind. What traveled with you was what lived in your head. That deal is potentially changing with AI. Knowledge workers are now building what Neuric calls a personal capability stack, combination of AI agents and systems that learn how you think, how you structure arguments, how you weigh trade-offs, and how you communicate decisions. And over time, these systems stop being productivity tools and start becoming something closer to a digital extension of your professional identity. The piece raises the key question: if your judgment and decision-making patterns are encoded into an AI AI system you built and refined over time, does that system belong to you or to your employer? We talked about this yesterday as a team, how skills are something that have existed since the workplace has been in existence, and now skills should include Claude skills, insert whatever foundational LLM skills, agent architecture, et cetera. The question remains, are you allowed to bring those things with you that you've built into your next stop career-wise? So Charles, I wanted to pause for your thoughts here.

I think this is one of those new new and old questions at the same time. I think there's always been like, hey, when you leave your job, if there's a cool slide format that you built, can you take that with you? That feels kind of low stakes. I think it's fascinating for two reasons. One, because of the shelf life of a lot of this stuff, it's very possible that if you leave, by the time you'd want to take it with you, the thing you built may no longer be in production. And when I read the article, my immediate thought was almost the opposite. It's like, what are companies going to do with all of these agents or things that individuals have built? When the individual leaves, like, who's going to, who's going to support that vibe-coded thing that's become? And ironically, I'm like, well, this is part of the reason that you buy software from other people is that their responsibility is to keep the thing up and running and you're not dependent on a single person in your org. I think there's also some more structural challenges here, which is if I wrote a skill that really works, the fact that, you know, some people are doing their work using corporate-provided team account or enterprise versions of AI, but some people are also using personal versions of ChatGPT or Claude. And yes, of course, no one should be using their personal thing to do work accounts, but let's be honest, people are. The question then becomes like, what happens if I built it in my personal account? Or if I built something on my free time that makes me more effective at work, does that tool even work without the context of the environment that I'm in? And conversely, if I wrote a skill that's like broadly applicable, sure, maybe I leave the SkillMD file at the office, but I still have all that knowledge in my brain. And so I think this is fascinating because it's not clear to me even that things that were built a year ago are still as useful as one would think. So I don't even be— I don't even know how to begin to think about this question. It's like so rich.

The irony of it when I was reading it is that these large companies have placed the task onto the employee of getting smart on AI and building, and they might be left with— they as in the employer— might be left with like a lot more issues than they anticipated if people actually get really really smart and up to speed, because if they do leave or if there's a layoff, those systems and tools that they asked them to build that maybe made the company more efficient don't just run autonomously in the world that we live in today. Like, we build stuff all the time internally here. I was rebuilding a tool to help us look at portfolio health, and I was working with Aaliyah, one of our teammates, on building that. I built it on my own on our team account, but you can't like share co-work projects in the way that you really need to, especially if they're complex. And then I was like, all right, I'll extract out the text of this and then be able— and you can be able to run it as well. She ran it And she had all these questions and I'm like, oh yeah, I have this like system. I know that those questions were gonna happen, but you didn't know that because you didn't build it. I have this framework of like, no, no, no, it's not making mistakes. This is just like the step process. That back and forth took 4 or 5 days just to understand what gaps there were between people on the same team. And we have a small team relatively.

Yeah.

If you work in a massive organization and you're handing off multiple agents for multiple different projects, I don't really know how you're able to handle that transition.

I think the thing that I find hilarious I rediscover this lesson every week is everybody, whenever I tell people these problems, they're like, you know, these are solved problems in software development, right? Software developers have encountered these problems before, and this is why we have comments in code, and this is why you have documentation, and this is why you have all of these things. And I'm like, yes, all of these non-developers are learning all of the basics of software development, whether that's good documentation, whether that's version control, whether that's collaboration. Like we're all learning all the things that professional software developers have solved, although we're learning them in a different context because many more people are building and the things they're building are in some cases more siloed, but also more autonomous. So I do think at some point companies are gonna figure out better systems for building this stuff, better systems for documentation, because I think most companies deep down abhor being reliant or dependent on a single individual. I just think people haven't figured out how do we both push the work of building the stuff down to the individual, but also make sure that we have a way to recover and operate if that individual leaves.

Where do you think legal arguments or any sort of regulation is going to fall on this if there starts being conversations in the background, which I assume there are happening right now.

Well, I mean, I think in some instances, if you have to build your agent using corporate infrastructure and it lives there, it's kind of, I think to me it's obvious, but I keep asking myself like, well, what does it mean to take an agent with you if the agent's power was the data to which it had access? So I do think that the notion that you can take agent outputs that are like corporate data, I have a feeling companies are going to win that argument. Hey, you can't use an agent to walk out the door, but a framework for doing an analytical project on top of a database, I don't know how you stop somebody from doing that. They're going to have the knowledge of how to do it. Whether they copy the exact computer-generated Python code to do it is another issue altogether. So I suspect this is going to be very murky because it's not like, oh, I like stuck a USB drive in my computer and sucked down a bunch of customer records. We've, we've kind of decided you're not allowed to do that. But hey, an architecture that cleverly helps me you know, do a cohort analysis on customer data in a given system with a killer prompt. I don't know how you stop somebody from— even if you don't let them take the text out the door, the knowledge is going to walk out the door with them when they leave.

That's protected. Yeah. And I can, I can imagine there being some sort of infrastructure to protect this when you work at like a sub-100 employee organization. I do wonder, how does this work in scale? How do you run after all these people if they have 5 to 10 agents themselves that they've built up and you work at like a 20,000+ organization? I don't know how you watch that in today's world.

Again, I think the answer for most companies will be eventually will be centralization and that individuals will do less of this development. Right now it's like a thousand flowers bloom, push it out to the edge. But at some point I think people will say, hey, this customer pipeline agent is a critical thing. We can't have the one person in sales who's like, I build, build this. We need to have a centralized team that builds and manages this agent. And I think we'll get there eventually. We're just far from— it feels like far from that.

One last comment back to your earlier statement about this is why people have worked with B2B software companies for so long. When I was in consulting, you would go and arrive at a client and you would take a look at their entire tech stack. When you saw that there were homegrown tools in the tech stack, when you were doing a review, you would start to freak out because you were, you were being like, oh no, who's responsible for this? How did they build this? Like, how janky is it? Sometimes it's really complex, but that's even scarier. How do we get rid of it? This is almost like that, but thousands of times over. If each person has built their own homegrown tool. I, I literally remember we would have our own work stream assigned to homegrown tools, running around our client, finding out the people who built them and seeing if they still worked here.

And then in most cases you're like, well, oh, we can't touch it.

You can't.

Cause it, it works and nobody knows how to work. Nobody knows how it works. No one knows.

And leadership doesn't want you to touch it either. They're like, just don't touch that. And you're like, okay. Well, our next piece is by Ioana Olteniu, published on her Substack Motive Notes titled When Your VC Leaves, You Are Fundraising Again. It opens with a founder texting Ioana the morning their partner posted on X about leaving the firm, asking how to manage the risk. The core argument is that when your VC leaves, your company gets quietly reevaluated inside the firm, not in a formal meeting or a memo, but in where attention flows, who's speaking up for you in partner discussions, and who uses their political capital on your behalf. All of that institutional memory, the reasoning behind early bets, the contacts that made your thesis legible before it was obvious, walks out the door with your partner. Iwana walks through the two lenses founders need to hold simultaneously: the founder lens, which says, which says your job as a CEO doesn't change, and the structural lens, which says the firm's internal ownership just reset regardless of how warm the emails stay. The piece is full of really specific advice from asking directly who owns you internally now to recognizing that go raise from the market and we'll take our pro rata is not actually formal backing. It's a firm waiting for someone else to take the risk first. Obviously our friend Hunter wrote a really good piece about this a few years ago. How real do you think this phenomenon is? How much has it grown? Grown in occurrence, and what should the priorities be of a founder when they encounter this, since it's going to happen more and more with increased movement across venture from firm to firm?

I've seen, ironically, the firm-to-firm stuff is less disruptive than the firm-to-industry stuff, because the firm-to-firm, the person is still a VC. They can't help you in the same way, but they're still in the mix and in the flow. When someone's in industry, they're just done with venture. Yeah, whatever connection that person had to your company is very different. And I think we're seeing— I don't know about your LinkedIn feed— I, every I see someone I know. I see two flavors, junior partners leaving and going into TBD jobs that have not been announced, but I assume that they've already decided what they want to do. I've seen more senior folks either leaving the business altogether or leaving to start their own firms. And so I think this is going to be— every year I feel the half-life of venture careers gets shorter on a per-firm basis. I'm seeing more and more instances of this. And I think a lot of founders I find are getting caught flat-footed because the partners oftentimes don't tell them they're leaving until the very last minute. And I think Alana's advice is really good. It's very similar to what I tell people is like, you need to immediately figure out who's responsible for you. And if you don't think you have good sponsorship internally, you should ask for a better sponsor. I had one founder in our portfolio who was like, well, I really like this person and I want them to be my sponsor. I was like, well, that person has no power in that firm. And the person who did this deal was a managing partner. So you're going to go from a managing partner who could look out for you and make sure things happen for you to someone who's basically an operating partner. Who has a much looser kind of— I was like, I don't think that's a very good strategic decision for you. You know, in a lot of cases, there's a moment when you've been abandoned as a founder where you can actually ask the firm for more. You can be like, hey, you guys, like, this person left. I'm feeling kind of abandoned here. I want that person or I want this thing. And I think in a lot of cases, firms do feel bad about doing this. But like, at some point, then it's just over. And I really do think it's hard because what Oana highlights is the truth. It's it's really the advocacy and people's willingness to spend social capital on your behalf that goes away. And I've been on, I've seen this from every side. When I was a founder, the person who led my seed round left and the person who took over that person is like a very good person, but didn't have the same connection to the company. We've had people leave here at Precursor where I've inherited companies where like I didn't have a connection to the founders. And in many cases I've tried to get to know them, but in some cases it's just not the same. And I've seen our portfolio companies go through this. I think it's really, it's really hard. And all the rules I used to give them, make sure it's senior person, ideally the founder of the firm. Even those rules don't apply anymore because people are leaving and leaving for all kinds of reasons. And what used to be like a safe profile of a sponsor doesn't feel safe to me anymore.

Alana outlines different states for where founders could find themselves, and one of them is a truly abandoned one where there's no other person getting assigned. It just feels like people have, have ghosted, which sounds crazy if you're outside of venture hearing that people kind of abandon from a board perspective, but I see it happen a lot. What do you think of her advice to CEOs seek help outside the cap table, especially if you're in the founder position where maybe the route is to start finding a soft landing for your company? Is it going to find other founders who've been through this situation and have gone down the M&A path? Is it seeking out other investors that you've built relationships with who have, have hopped around? What do you think of that?

To me, there's like this progression where it's like partner shows up to the board meeting in person, then the partner only shows up on Zoom, then the partner brings their associate and the two of them are on the board meeting, or principal as the case may be. Then the partner comes every other board meeting and all of a sudden the person's just not there. To me, that's a very common slow walk toward— it's a sign. And I think most companies are smart enough to realize that if you're getting a less senior resource, you're either helping to train that person, A, B, the firm thinks that person's ready for board work and you're still trying to train that person, or C, the signal from the firm that like this isn't a high priority investment anymore. And I think as a founder, when you get that signal, you have to decide, well, what am I going to do with this information? If I agree with the firm's assessment, that this isn't going well, maybe I should be more actively trying to sell the company. If I don't agree with their assessment, it frees you from having to listen to them to some extent. You're now free to run the company as you see fit, which could include M&A, but which could also include a pivot, could include a lot of different things. But I do think if you don't have a great sponsor who's senior and who cares about your company, you should assume that no one in the firm is paying attention to you. And unless you're a top, top performer, you're probably not going to get any money or attention from that firm going forward. So your options are either turn your company into a top performer so that you can once again get the their time and attention, or use your time elsewhere. I'm involved with the company where most of the existing investors are checked out, and it's sort of me, the founder, and one angel. And that's that person's support system. And we're working with this person because we care about them. I'm not sure that it's gonna have a huge financial outcome for, for me or for them, but they're still working. Yeah. And I was like, they're working hard. We gave them their, our money, and we should continue to give them some amount of our time too.

It is just a constant reminder for founders to really think deeply about who you bring on your board and that it sticks with you round over round. I was catching up with a founder not not within our portfolio, but one that I've known for a while. And she's going through like a really tumultuous time and the board has kind of left her in the abandoned state where they're no longer interested. They're not interested in re-upping and providing more capital, but they're also not interested in supporting her pursue any acquisition paths. And I asked her if she was surprised by that, just knowing the folks around the table. And she said yes and no. She said no, because candidly, they hadn't been that helpful strategically for her in the past 5 years of the business. But yes, she was surprised in that she confused why they didn't seem more active to find routes to make back more money if that was a path. And I think that actually that, that happens more times than you think. People just become complacent and not helpful at all.

I also think there is a psychology in a lot of firms. Well, we've written that thing down to zero. Unless something changes, this is a zero. And unless you see something obvious you can change about the company, you're like, well, this is a zero. It's a zero on the books already, or it's a zero in my mind. So why do any work? And going from a zero to 10 cents on the dollar at most firms doesn't really change anything. So you're like, well, really to change something for this to matter, I've gotta go from a zero to maybe 75 cents on the dollar for anybody to care. And boy, that feels hard when you're on zero.

Our last piece this week is from Anne Uricoh on X titled Everyone Wants to Be AI-Pilled. Most Companies Are Still Level 1. And it outlines a framework to cut through the noise around what it actually means for a company to be AI-native. The piece opens with a comparison between small startups where being AI-native can be a founding team personality trait because the teams are so strong and they're so close to the customer, and scaling companies where it absolutely has to be embedded into the DNA to be considered AI native. Anne draws an analogy to the levels of autonomy that defined the autonomous vehicle industry debate for many years. Cruise control was not autonomy. Lane keeping was also not autonomy. And for this instance today, 80% of our employees using AI weekly is certainly not an AI native organization. And she maps out 6 levels from L0, where AI is essentially theater, and and nothing of consequence actually happens, all the way to L5, a virtually self-driving organization where the system notices things, synthesizes across sources, and updates shared memory without a human initiating the process at all. She says that this doesn't really exist yet. Most companies today are operating at L1, personal productivity, where the best AI users' workflows walk out the door with them and the org chart hasn't changed at all. So kind of back to our first piece today. Charles, curious from your perspective, what percentage of companies companies that you talk to day to day, what level do you feel like they actually sit on, and why do you think they believe they probably sit on a higher level than they really are? And then maybe zooming out at the larger company level, what percent of companies are truly operating at, uh, I would even say like L1 or above?

I think all startups always believe they're exceptional, so it's not surprising to me that everybody would rate their own AI adoption quite high. Also, nobody would want to admit that they're L0 or even if they are. To me, I read this and I'm like, we are so— and I liked Anne's framework, but between token maxing and this, we are so desperate to find a framework to figure out what matters. And I've always thought the token maxing thing is weird because in some ways it rewards inefficiency. I refactored two of my agents last weekend, brought the cost down by 90%. They consume less tokens. It's actually better. It's a more efficient agent. But if there was some token leaderboard, I would've probably gone—

Undinged. Yeah.

And I think this notion of AI adoption having levels, I think is true, but I think the tricky thing is it seems to change every 2 quarters. You know, a few quarters ago, having a GitHub repository with a bunch of shared skills on a Teams account maybe would've put you in the top. So I tend to think of what is the top 25% of companies doing versus what's the average company doing. And I think every quarter it feels like the things that were top tier advanced become average. And so I don't know, even know how, by the time you end up measuring which level you're on, the levels will have shifted. But I do think a lot of these frameworks, I think, imply a couple of things that I'm not sure. Like one, I, I appreciate Anne acknowledging that level 5 doesn't exist yet. I'm not even sure that that's the right objective for most companies. Yeah. Given the complexity of their business. I think what makes more sense to me is I've enjoyed following Jason Lemkin's journey of like using agents at Saastr. And I think what he describes feels closer to me to what top usage looks like, which is mostly autonomous agents doing a bunch of discrete tasks that are then rolled up either to some kind of orchestrator or a human, and they're being overseen and watched by a person who has judgment, who can look at something and say, oh, this is broken, or this isn't producing the output I expected, or wow, there's a new use case we need to help teach the agent. I think this idea of like humans managing agents to me is what I see many of the most sophisticated companies doing. Are there some processes that run fully autonomously? Absolutely. And it's That's great for discrete tasks. I think that is great. But I think also a lot of companies, it's uneven. Your finance team might be L1, but your engineering team might be L4. What does that make your company? I don't know.

Yeah.

Do you take the average? Do you like, do you min-max? And so I think we are so early in all of this stuff that my experience has been there are very few companies where everybody's consistently operating at a super high level unless it's 3 people. One of the things I've, I've been having a long-running conversation with one of our portfolio companies that's built It's been, it's a fund one company. It's, it's been around for a while. We've been sharing our AI building journeys and her last update has a bunch of really amazing things that she and the company has built. Whereas a quarter ago she was like, maybe I'm going to have AI help us with customer support. And now she has an AI agent that is closing 6 or 8 deals a quarter. And so I think the other thing is people, I think, are overestimating where people are on the average, but underestimating how quickly you can go from L0 to maybe L3. And I think we're gonna continue to look for frameworks and measurement methodologies that help us figure out kind of, am I doing this right? And I think it's gonna take a long time before we know. And as soon as we find something that feels like it's working, the world will change again.

She had a good quote, which was, Steve Blank once said that a startup is not a small version of a large company. Mm-hmm. Similarly, an AI-pilled company is not simply an AI-assisted version of an old company. And so the one theme that rings through in the piece is that the entire infrastructure for what we think an AI-native company looks like will likely not look like anything else that we've seen prior. To your point about managing agents?

I think the funny thing is every time someone tells me something, I'm, I hate like the whataboutism. I'm like, but what about Google? They seem to have done a pretty good job. They didn't throw out search. So I also think it's very comforting as a VC to believe that AI-pilled startups or companies that are AI-native from scratch will beat old companies in the same way it was attractive to believe that every cloud would beat on-prem every time. That isn't actually what happened. It isn't. On average, yes, cloud won, but there's a lot of funky, weird on-prem stuff still Bought, serviced, and in production. And I suspect that not every AI-pilled company will win every market. There will be the revenge of the incumbents and like, I wouldn't count them out just because they don't always fit a framework.

We are all out of time. Thank you, Charles Hudson, for jumping in. Thank you for our listeners for joining as well. Please like and subscribe and we will talk with you all next week. Thank you. Bye.