We spent some time with Dominic-Madori Davis at TechCrunch to discuss being a young founder and how we consider backing teams that were born half a decade after Y2K. It’s worth the read and she gets the perspective of young founders and old VCs and it really helps frame what I think is the open question of backing teams under 20: Are we setting young founders up to fail by giving them all this cash with no oversight or direction in the off chance they become the next Mark Zuckerberg or Cursor?

Long after the Dot-com bubble and an infant during the housing crash of 2008, these brilliant teams came of age during the pandemic and frankly, might have had more time online and around computers because of it.

Why are funds backing younger and younger founders

So the questions Dominic had are layered and relevant in this fundraising environment where I see funds jokingly (but not) start targeting the top high schools for talent. College dropouts are the new FAANG researcher are the new zero to a billion startup early engineering talent. And with all cycles, it will come back around.

So why are VCs backing the youngest teams, the downfalls, the warnings. I’ve lived through a few cycles and watched different ages be the en vogue team to back. Through Vermilion I’ve backed teams from age 19 through teams that have also lived through a few decades of tech and with all things, there are positives and negatives to all. I like having an age diverse portfolio.

“Are young founders under more pressure when it comes to building companies these days? Why and what does this mean? How has being a young founder changed in the past ten years and why is that so? Do you think today’s startup ecosystem genuinely supports young founders or just celebrates them when they fit a certain narrative?” all questions posed by Dominic as we chatted.

My long-winded musings and thoughts

On young founders being under more pressure when it comes to building companies these days. I think it is a different kind of pressure than it was 10 years ago. The credibility problem associated with being a young founder (18-24) never went away but as an active target for VC funding, you still have to convince people to take you seriously.

But now there’s a second layer that makes this funding cycle for young teams different than last. The market doesn’t give you room to learn slowly anymore and I’ve heard that feedback from younger founders that built companies in the 2010s. AI expectations are that you’ll figure out your growth engine within months, not years. The forgiveness that used to exist at an early stage and the assumption you’d iterate your way to product-market fit doesn’t exist right now. Capital is available but the patience that used to come with it at the earliest stages isn’t.

And so how has it changed in the last 10 years? During this AI boom?

Ten years ago you had fewer tools and fewer investors who’d take the meeting. Now resources and incubators are everywhere and that’s made it harder in some ways for everyone. I think there are more predatory incubator programs making promises that they can’t keep.

The other big shift: young founders today have been marinated in founder mythology since they were teenagers and told college doesn’t matter while my generation was told college was the only way forward. For me, college helped me learn how to socialize coming from a tiny town and moving to a “big” city like Atlanta. But some kids don’t need help learning to socialize and frankly, I’ve not used my Biomedical Engineering degree but I’ve used the friendships and free time hacking around on things much more in my career.

By my second year of college I was working a corporate America job almost full-time and while I’m sure I sounded like an idiot in meetings sometimes (nothing has changed) and probably wore the wrong outfits (truly, nothing has changed), I was able to do the work and learned a ton about corporate America. I had friends building startups in college. It’s not a new concept. I’ve hired kids I’ve told to drop out of college. And I’ve hired 19 year olds that graduated way early and were some of the smartest people I’ve ever worked with. One of the smartest and most accomplished people I know didn’t go to college during a time when that was unheard of and was an exec at Google for almost 2 decades. Age can be relative but I think the expectations from VCs around young founders today are almost unfair.

A meaningful share of my portfolio founders are in their twenties with a few under 21. So I’m clearly not skeptical of youth. What I’m underwriting is the quality of the insight and the technical depth behind it. The question then becomes, how deep can the technical depth or insight be without real life experience.

One example in our portfolio is Lemma from YC F25. Two brilliant founders that are young and bright and working on their third or fourth company. What they lack in experience of working at big companies or other startups, they make up for in excitement to experiment and lack of fear. Mentioned earlier, we also back founders with more experience working on teams, building for the enterprise, and domain expertise. There are downfalls to both approaches. Personally, what I make up for in experience I lost in fear of consequence.

What has really, really changed is the appetite for backing young founders with more pre-seed funds, more accelerators, more capital available earlier than at any point in the industry’s history. What hasn’t changed is that the hard part is still picking correctly but that’s always been the job.

And all incubators aren’t created equal.

I worry that some of the young founders can end up at predatory incubators that write a small check for a giant portion of their startup without bringing anything to the table and making it really hard to raise another round because of dilution.

The advice here is to reference the incubator then reference them again. If it isn’t the best in class YC or a few others, do the work to see if as a young founder they are giving you a small check (which to them is inconsequential) on the off chance your company becomes a multi-billion dollar exit for their fund. They then own so much of your company that it becomes hard to dilute the cap table to raise subsequent rounds and they are setting you up for failure. If that doesn’t make sense, run it through Claude or ChatGPT and get a deep understanding here before taking incubator capital that isn’t a very well known brand. Understand that someone giving you $100k and a desk for 15%-20% of your startup isn’t normal.

YC is obviously the best and at the other end of the spectrum of these predatory incubators. a16z’s Speedrun and a handful of others also break this mold. But of course those are hard incubators to get into but worth the try.

YC is always setting the tone of the next few quarters of venture funding. The average age of their founders has been rapidly dropping over the last few batches. Jared Heyman over at Rebel Fund backs primarily YC companies and has more data around this than I would. He writes specifically about the YC age drop here and it’s well worth the read: https://jaredheyman.medium.com/on-the-new-y-combinator-3c28e548896c

From Jared’s post mentioned above about YC average age by batch year

So what are the funds trying to accomplish by backing 20 year olds?

A lot of venture is pattern matching and trying to back enough founders that you have a better chance at the big return. We know every company can’t succeed. Companies shut down all the time and while it is hard as a fund to see it happen, it can really be life changing and demoralizing for founders. Especially ones that are so young it is the only thing on their resume. The good news is there’s nothing easier to back or hire than a second and third time founder. The bad news is no one tells founders this at the onset.

The expectations are a bit wild right now and frankly, not realistic. Startups and founders need time and space to cook. Funds are looking for revenue numbers in the millions within 12 months when we used to get excited about community growth instead of immediately focusing on revenue. Often seed funds would like to see $1M in ARR which used to be the metric we saw for raising an A. I wrote about this in Round Names are Meaningless Right Now and Redefining Round Names, Stretched Fundraising Timelines, and What’s on the Horizon for Technical Investing back in April.

But to sum it up, everyone is chasing the next Cursor. Cursor was built by a set of young, brilliant MIT founders that went from zero to $2B+ in revenue in 4 years (forecasting $6B by EOY). But that’s an atypical growth trajectory and not something you can benchmark against. I’ve said it over and over and over and over: You’re not raising the bar when you set outlier performance as the expectation.

Resetting founder and VC expectations and the myth of the young founder

Or more simply put, what do funds owe the younger generation of founders, if anything? I think it’s honesty and realistic expectations to go along with the capital. Space to grow. Not only the company but as founders and humans. And being clear that failure at the earliest stages is extremely normal.

Watch The Social Network or scroll LinkedIn for ten minutes. Nobody is posting about shutting down their company or not being able to raise their next round. The stories of failure, which are more prevalent and honestly more instructive than the wins, aren’t being told on the big screen. For a young founder who hasn’t watched peers go through the harder parts yet, that silence creates a distorted map. Not knowing what to be afraid of can be a strength. But so can the myth that if you build something great, it will succeed. Great products fail every single day. Shitty products make it.

The pressure young founders are putting on themselves right now is a direct product of that distortion. They’re measuring their private reality against everyone else’s public highlight reel and concluding they’re behind. They’re not behind. They’re just seeing the myth clearly for the first time and don’t have enough experience yet to know it’s not real and has been created through careful social media creation.

Investors have a responsibility here that not enough of them take seriously. The job is to find the winners and help them win, that’s what you owe to yourself and to your LPs. But good investors also understand what it means to be on the other side of that check for the majority of the portfolio that won’t break out. That looks like helping founders find a soft landing through an acquihire or acquisition, helping them think clearly about winding down, or just letting them keep going as long as they want to even if as a VC you can’t be as engaged as you were from day 0.

Ultimately it’s a founder’s decision when they’re done. As an investor, you can give advice. That’s it. The best thing you can do is be honest with them earlier than feels comfortable.

A lot of my intro calls with founders, VCs, and LPs start the same way. “I read your post about your first fund.” And I always apologize for the length. If you have a spare four hours, go start from the beginning and learn too many details about me and my story. But it saves me a good 5-10 minutes of giving my background and more time to focus on the person or team I’m meeting with so that was an unintentional win.

Something I think marks great founders is resilience. The ability to keep moving when you hit what feels like a dead end. And as I wrote about in the launch of Vermilion Fund I, when you get to the end of the trail, keep going.

It’s amazing what you can learn in a few years. I started raising Fund I in 2023 and announced it just over a year ago. I’ve learned more in the last three years than in the previous ten combined and not just about being a fund manager but about myself. There’s a beauty to going into something with little experience, the way I did with Fund I. Raising and deploying Fund II has been different in ways I didn’t expect. And has authentically been some of the most rewarding work of my life.

So, the news.

Vermilion Cliffs Ventures has closed its second fund. $25M, up from $13M in Fund I. Same focus just bigger checks. We back technical founders building for technical users in AI infrastructure, security, and developer tooling.

The proof that this works isn’t mine. It’s theirs. Keycard came out of stealth to secure AI agents. CopilotKit raised its Series A for app-native agents. Sequin found its home at Notion. Those are the public ones but there are so many that have made incredible progress behind the scenes. Everything moves so fast even when you think no one is looking.

The part I’m most grateful for: 95% of Fund I investors, by dollars, are coming back for Fund II. You backed me with little data. Then you backed me again and it feels good to have a community of LPs that believe in me.

Raising a second fund is hard. Most people who raise a first one don’t raise a second right now, because either the work stops being appealing or the fundraise is too punishing. My Fund I LPs came back for Fund II and I added a few more great partners around the table to round out 15 total LPs including 3 incredible anchors with LGT Capital, Screendoor, and GEM. I’m thankful for my earliest supporters and proud that the group is small, supportive, and believes in me and my vision for the fund.

Some lessons.

Working with others is more fun than going it alone. Fund I, I built solo, and I loved the independence but felt the loneliness of being a solo GP that many had warned me about. Fund II, I wanted a teammate so I brought on Meghan Murphy, a former colleague and one of the sharpest technical GTM minds I know, to join me on this ride. The work is even more fun with a partner and having her on the team means the founders in the portfolio get more access to GTM experience and I’m not the bottleneck.

Meghan Murphy and I finally getting a great photo at a recent Vermilion event in SF. She is the reason the funds runs smoothly. The founders love her, I love her, and she is my people. We’ve worked together for almost 16 years and I hope for 16 more.

Fund I, you sell a narrative. Fund II, you sell a make-believe scoreboard. With fund I you’re selling the vision especially if you aren’t coming in as a career VC with a proven pipeline and returns. With fund II, LPs can look at fund I and try to estimate what investments they think will return dollars. With only limited data, a lot of this is make believe metrics.

Not all capital is created equal. A check is a check, but the people behind it aren’t interchangeable. Partner with the ones who bring value and community, not just dollars. I tell my founders to build a cap table of different strengths. I’ve been spending more time thinking about who I want to work with long term with the founders I back and those relationships with other VCs have been so rewarding personally and professionally.

Make time for life. It's possible to be obsessed with the job every waking hour. I spend a lot of my time and energy thinking about the fund. I love it. But you have to build in the downtime, and the room for friends, family, and the people and things you love. The work will take everything you give it, so decide in advance what it doesn't get. For me that looks like calendar blocks for walks, long weekends, and leaving my laptop behind to sit in the park with a book. One of my anchor LPs gave me a great piece of casual advice when I sat with her recently. I told her I'd stopped doing some of the things that are core to my existence, even though I had the time. She told me to make time for life. She's right.

You have to fail to succeed. Venture isn’t a straight line, and it isn’t supposed to be. The failure rate is higher than almost any normal career, and at the earliest stage it’s the highest of all. For a part-time perfectionist like me, it’s been a hard adjustment period. The misses are the cost of the swings that return the fund. That’s the entire job.

Conviction is cheap when you have no track record. Fund I, I had a vision, a story, and the need to tell it. That was enough to start. Fund II, the hard muscle is discipline. Passing on a genuinely good company because it isn’t my company. Saying no to a founder I like. That’s harder than any yes.

Pattern recognition is a gift and a burden. You can know more and feel less certain. The thing that makes you fast is the same thing that makes you miss the founder who doesn’t fit the shape you’re expecting. Keep an open mind and always be ready to change your opinion or have your first impression challenged. That’s good advice in life and in work.

I used to think experience would make me more sure of my decision making. Instead it has made me cautious which I don’t necessarily like.

Choose how you spend your time. The people I get to work with, whether by investing in their ideas, taking their investment into my fund, or sitting beside them as co-investors, are people I'll work with for decades or more. None of these are short-term relationships. So be considerate about who you work with, and in turn how you spend your time. And choosing how to spend your time means spending time working on things you like with people you like.

Community is bigger than your network. This is the most important realization I’ve had building Vermilion, and it carries over into getting older and choosing the people around you. When you’re building anything important to you, community is everything. I tell founders daily to start by finding the people who are excited about what you’re building and who give you support, and to grow that base.

One of my first LPs, in since day one, has told me the same thing for years. That I don’t understand how broad my network is and how many people I have in my corner.

She’s right that I underestimate it. But beyond network, it’s my community.

And community is everything. It’s most of the job, if I’m honest about it. Every time we throw an event, or I need a favor, or someone else needs one, it’s the community I’ve spent two decades building that comes through. It’s the people from Twilio who volunteer to help set up an event oversubscribed three times over. It’s an old CEO of mine running diligence on a deal with me. It’s another CEO of mine deep in a fundraise, the two of us locked in together like we were decades ago. And it’s the people I used to ship with every day as a bright-eyed 24-year-old, coming back together to build and reminiscing about how much fun we used to have.

Last week I got to spend time with all three CEOs from my first three startups, where I cut my teeth and made mistakes. Three different contexts for the time spent with them and all of them still in my life and now back in my work. I got to build with each of them again. It reminded me of their greatness and why I still look for founders who share their qualities. Two decades in, and the people I started with are still the people I’m scheming with.

People talk a lot about not burning bridges when you leave a job. In a community as small as SF tech, that’s one of my biggest life lessons. Leave elegantly and be kind. You’ll be working with these friends and former colleagues for the rest of your life if you’re lucky.

I’ve grown up alongside these people. Careers and companies and marriages and kids and moves and life’s surprises. Good years and hard ones. Wins and the losses in between. We’re all still growing. That’s the part I didn’t quite get. You never really grow up. You don’t arrive. You just keep going, together.

And as a fund manager your LPs become the people you call when you’re stressed and need help with a problem. My anchor LPs are my friends. They are a big part of my community. I understand it’s a business relationship, and the expectation is that I make them hefty amounts of money, and I will. But it’s more than that. They are my people. And I’m very lucky to have them in my corner.

Laura Thompson from LGT and I pictured in SF where she consistently shows up to our silly tiki parties and hacker events. She has been a friend, a source of endless wisdom, and one of the biggest supporters of me and Vermilion. She sees around corners and knows more about what it takes to build a fund than I do. She gives advice that always comes true and I’m like “Oh right,Laura said this would happen”. We clicked instantly and I’m so thankful for her friendship and mentorship. We will be in each other’s lives long after Vermilion.

Lisa Cawley from Screendoor and I when we met in SF and she told me she was expecting her second beautiful baby almost a year ago. Lisa has been a day 1 believer and I am lucky to have her. She is the person I vented to during Fund I when fundraising was hard and I was going through hard things in life. She was there for me. We’ll be friends for life.

Caroline Dallas from GEM and I meeting in NYC yesterday catching up over a casual lunch. Equal parts business and life gossip. She’s been an incredible partner and as our newest anchor LP, I’m just lucky to have someone that in a different world we would have been fast friends and stayed that way for life. Caroline is a force of nature. She is impressive in ways that are hard to explain and I watch her set her mind to a task, get it done, and blow everyone else out of the water. We have a lot in common.

It wasn't only the old relationships that carried forward. It wasn't just the new LPs who believed in me and my vision. It was the new relationships I've been so lucky to build. Founders who became friends. Co-investors who became a group chat that's only about 30% work. Other people in the industry who are just a delight and have become a core part of my day. People I met at the tail end of a raise who I now can't imagine working without. Almost twenty years in to my career in tech and so much of what drives the work today runs on relationships I didn't have a few years ago. With the new friendships and the people I've known for more than half my time on earth, I'm just so lucky to spend my life with you all. You are my people. You are my community. And you are everything to me.

It’s really hard to put the joy of my life into words. So instead I’m just going to do a millennial style mid-2000s Facebook album unedited and lacking context. But I love all of these people and it’s impossible to capture everyone so this random assortment hopefully shows that I have a beautiful group of humans in my community and I’m thankful for all of them.

So if I could tell anyone building anything one thing, it’s this. Surround yourself with people you enjoy and love. Make time for them. The work is long. The trail is long. Do it with your people. Find your community.

The point

I started Fund I in the middle of grief and life’s curveballs. Losing my grandfather, then a fast-follow cancer diagnosis for my dad that meant eighteen months of about the most severe treatment anyone could survive. But he did. He’s been on the tractor all weekend and is healthy, happy, and sassy.

Fund II I built on steadier ground, and what surprised me is that the work felt the same either way. Work and my community was always my grounding force. I think that’s the gift of loving what you do.

Fund I, the lesson was to ask for help. People like to help. Fund II, more often I’m the one being asked. The shift from the person looking for a door to the person who can open one sneaks up on you.

With Vermilion Fund II, I’ve built the team, learned from the experiences of Vermilion Fund I, and feel so lucky to work with the founders and community I’m surrounded with.

Thankful to still be on the trail.

_______

Read the Techcrunch article here

Some sent a single line. Some sent something closer to an essay. I’m running all of it, grouped by the three fault lines they kept landing on.

Common theme: Obviously founder narrative overlapping with AI.

The more subtle thread that weaves it all together: betting on a person before anyone else will, is where the edge still lives.

AI is coming for the venture business itself

The funds racing to automate judgment, and the people betting that judgment is the whole job.

“Using AI agents to screen seed founders is like using a high-tech ouija board. The criteria everyone’s scraping for today would have screened out the three $30B+ companies in Founder Collective’s Fund I. Automated filters can analyze patterns, but rarely does consensus mint the next big breakout. Many of FC’s highest-conviction investments are in ‘weird’ or unsexy spaces, because the people building them have extraordinary motivation and are more likely than others to ‘see around corners’ in their industries. Seed investing is a people business, built on relationships and randomness. Don’t underestimate getting into the wild and really getting to witness the fire burning in an ‘atypical’ founder. Hand gestures, intonations and eye contact convey a lot.”

“Most AI “technical moats” are just temporary API arbitrage. The real moat is product taste, workflow ownership, and distribution.”

““Won’t the model just do this?” is the laziest question in venture. Claude may one-shot code that passes every test and eval, but someone still has to run it at 2am when a vendor breaks their API, a retry quietly double-charges a customer, or an agent reaches for data it should never touch. Writing software and operating software are not the same thing. One is a demo. The other is a company. The moat was never the code.”

“VC firms should be less focused on investing in AI/software startups and more focused on consuming AI/software as users themselves.”

“Pattern matching is just bias masquerading as data”

Couldn’t help adding my own spicy take here.

Narrative is eating fundamentals

Pricing has detached from what companies are actually doing. Three reads on what that breaks.

“VC market in one word right now = kingmaking. VCs have shifted from funding proof to funding possibility. The more investors price startups off hypothetical endgames, the more fundraising becomes a contest of narrative gravity rather than business fundamentals, and the greater the risk of creating the next generation of zombie companies.”

“Heavy discounts on secondary sales are an inefficiency tax, a failure of process that can and should be avoided. When you practice narrative priming and early buyer activation long before a sale, you destroy the information asymmetry that comes with treating liquidity as a reactive, tactical event. Pre-warmed people who understand the story before the sale turn secondaries into a solvable data, storytelling, and relationship intelligence challenge, not a financial engineering exercise.”

“The more the VC grapevine and FOMO drive deal heat, the more founders value independent, first-principles conviction.”

The job is unglamorous, and that is the point

Takes on why the least glamorous version of this work is the real one.

“Venture (especially early stage) is currently a low status job, and that’s a good thing!”

“Inspired by Lorne Michaels on producing, seed investing is an invisible art. If you’re any good at it, you leave no fingerprints.”

Who wins this vintage

Two forward-looking bets on who comes out ahead, and neither is the consensus pick.

“VCs will make more money on college graduates than college dropouts in the next 10 years.”

“Founders win this vintage. Grow fast enough, dilute almost nothing.”

The actual point

Line these up and they argue with each other. One side says AI is rewriting what this job even is or is becoming. Another says narrative and FOMO have detached pricing from anything real. Maria Palma says founders are winning the vintage. Ashley Mayer says the job has never been lower status. Maybe everyone is right.

But read them again and they keep walking through different doors into the same room. The loud part of this market is AI, narrative, and heat. The quiet part is taste, conviction, and the read on a person nobody else believes in yet. That second part is where almost every take here sits, even the ones that start somewhere else.

The edge in early venture was never the model, the round name, or the price. It is the read on a person. Everything loud in this market is a distraction from that. Everything quiet is the work.

More soon. What’s your hot take?

Summer is here and for Vermilion that means we’re headed to San Francisco for the AI Engineer World’s Fair. Join us at two key events: AI Infra After Dark co-hosted by Amplify and Deploy, Scale, Score World Cup Social co-hosted with DigitalOcean.

✨ Register for both events here.

Basecamp updates

What’s happening in the portfolio

💰 CopilotKit raises $27M Series A led by Glilot Capital, NFX, and SignalFire, with participation from Vermilion Cliffs Ventures, to build the Enterprise Agentic Frontend Stack for generative UI and agent-human collaboration (TechCrunch)

🚀 Confident releases an open-source harness for vibe coders building AI agents — purpose-built to verify whether agent code changes actually improve outcomes, after research showing 94% of users have no reliable way to evaluate coding-agent output

🚀 Atomic ships repository-native code intelligence for CLI and Agent Tools — semantic search, scoped code search, entity extraction, and source-grounded summaries

📈 Keycard named to Redpoint’s InfraRed 100, honoring 100 transformative companies in cloud infrastructure, plus the CB Insights AI 100 for building the identity and access foundation that lets enterprises deploy autonomous agents at scale

🚀 Starsling launches Runners, the world’s first self-driving CI service — a closed loop around frontier models delivering automated DevOps after 500,000 CI jobs in production

From the trail

The latest industry news, interviews, and forward-thinking considerations from Vermilion, founders, and investors in the field.

• 📰 Why open source at GTM is having a moment via Meghan Murphy of Vermilion

• 📰 Founders are buying GitHub stars and getting funded via Elena Marchetti of Awesome Agents

• 📰 Raising a Series A with Magid and Co’s Aaron Harris via Ashley Smith of Vermilion

Who’s hiring

Find the latest roles across engineering, marketing, and developer relations on the new Vermilion Careers page here

Upcoming events

Where Vermilion will be in the next month, look forward to seeing you in person 👋

🌠 AI Infra After Dark with Amplify + Vermilion Cliffs Ventures | SF

🚀 Deploy, Scale, Score with DigitalOcean + Vermilion AI Infra Social | SF

A look at all our upcoming and past events

Aaron spent 8 years at YC where he built their Series A fundraising program from the ground up, helping founders raise over three billion dollars across two hundred rounds. Then what began as a way to help founders navigate the A series gap expanded into advising across A, B, and C rounds. The market shifts, the metrics shift, the people across the table shift. That arc gave Aaron a perspective to what venture looks like when you stop thinking in round letters and start thinking in bets.

Our convo covers the framework of what makes a venture round venture, why the founder in the room matters more than ever, and what story actually lands in 2026.

Full interview here with highlights below.

Everything is functionally a seed bet.

There’s a popular narrative that venture has fundamentally changed, that A’s look like old B’s, that seeds look like old A’s, that the letters have lost their meaning. The letter on the round isn’t the point.

“Functionally there are venture rounds and there are non venture rounds. Is this a venture bet? Is it primarily about predicting some unknown future off scant data or scant data relative to the size of the outcome you expect? Or is it a growth round where you’re putting money in and you know how it’s gonna be used and it’s gonna feed some machine that produces relatively known outcomes? And in venture, I think everything is functionally a seed bet. Do I think the founder in front of me can achieve something massive?”

What matters is whether you’re betting on an unknown future or feeding a known machine. And once you accept that frame, the partner meeting looks different. Revenue traction, novel tech, a fast-moving market: none of it separates you from the four other founders pitching that same day with the same story. What’s left is who’s in the room and whether they can convince it they’re the one who pulls it off.

The story has to do more work than ever

The bar for the pitch has moved. It’s not enough to map yourself onto a changing market and explain where you slot in. The founders who land are the ones who can credibly claim they’re the cause of the change.

“There is a much stronger emphasis on the person in the room than I think there ever was before. The founder was always critical. But at this point, that founder has to be able to do the kind of world building thing in a way that I think is different than even just a year or two years ago. Because you have to be able to present a range of outcomes that is so large..and you have to convince the investors that you’re the one to do that. It’s really hard to tell that one because it takes a certain confidence bordering on arrogance to say you’re gonna change the world.”

That confidence is the part most founders soften. It feels safer to position yourself inside a trend than to claim you are one. But the safer story is also the one that gets passed on.

Magnified outcomes in shorter timelines

A lot of the noise around “venture is broken” or “venture has changed forever” misses what actually shifted. The underlying question investors are asking is the same. The window for answering it just got smaller.

“If before the question was, hey, how much can you accomplish in three years? The question is now, how much can you accomplish in 18 months or 12 months or six months? And so they are the same set of questions that people are trying to evaluate: is putting money into this business going to result in me making a hell of a lot of money in a relatively short timeframe.”

Speed alone used to be a moat. Now speed is the price of entry, and the question underneath every venture bet is whether you can compress what used to be a three-year arc into something closer to twelve months.

Final thoughts

The through-line here is that the venture machine hasn’t been rewired. The same question is on the table. The founder still has to be the one to answer it. The story still has to land. The math still has to math.

What’s changed is the compression. Outcomes are bigger, timelines are shorter, and the gap between the founders telling the right story and the ones telling the wrong one is wider than it’s ever been. The right story isn’t that you fit into a changing world. It’s that you’re the reason the world is changing.

For founders raising right now, the round letter doesn’t matter. What matters is whether the person across the table believes you’re going to capture the value when the dust settles.

A founder in the Vermilion portfolio recently asked if he should open source his product to help build momentum and drive adoption around a targeted pain point for developers. The answer is a layered one.

Open source used to be an ethos. A philosophy. A community thing. Now it’s a GTM strategy and in many cases, one of the most effective ones available to developer-first startups.

The same playbook built Elastic, Confluent, HashiCorp, and MongoDB into companies worth billions. Now we’re watching the next generation of founders try to run it at the AI layer. Some of our own portcos like CopilotKit and Cognee are doing it very well.

But OSS-as-GTM is not a cheat code. It has real trade-offs, real timing dependencies, and real failure modes. Getting initial traction is hard and takes grit.

Where some startups use star growth as measures of momentum, it’s getting harder to prove. A recent investigation found six million suspected fake stars across nearly 19,000 repositories. AI and LLM repos were the largest non-malicious category. Forks, contributor retention, and issue quality are harder to fake and tell you much more about whether anyone is actually using the thing.

Below are some ways to think about open source in your GTM strategy and what to consider from a product perspective to how hard it really is to turn open source users into enterprise customers. A note, this thinking assumes you are an early stage company. When you’re a larger and later stage entity, the open source strategy becomes a bit of a different ball game.

Why OSS Works as GTM (When It Works)

The mechanics of why it works are simpler than most people make them. GitHub stars, forks, and community pull requests create awareness that paid acquisition can’t replicate.

When every dev tool company is competing for the same developer attention and CAC keeps climbing, open source is one of the only distribution strategies that gets cheaper the more people use it. The motion that makes it work is simple in theory and hard in practice. An individual developer finds your project, installs it, gets value from it. They start advocating internally. The team adopts it. Someone with a budget eventually asks how to get enterprise support. The developer was your champion before you ever had a sales rep. That’s the dream. And when it works, it is genuinely one of the best sales motions that exists for infrastructure and tooling companies.

There’s also a compounding dynamic that most founders underestimate early on. Every commit, every release, every piece of documentation that goes public is a distribution asset. The community contributes, extends, and markets the product for you over time. It doesn’t feel like much in the first six months. A year in, it starts to matter. Two years in, it’s a real moat.

The last piece is credibility, and this one is specific to the categories that Vermilion invests in. For infra, AI memory, agent frameworks, security tooling, the buyers are technical and skeptical. Developers can read the code, audit the architecture, and understand what they’re actually deploying.

When the conditions are right, OSS-as-GTM can work in ways that almost nothing else does.

The Core Strategic Questions

So should you lead with open source, or build the commercial product first and open source later? There are patterns in when each approach works and when it doesn’t.

When OSS as your first product makes sense

The clearest case for going OSS-first is when you’re building something that only has value if the ecosystem adopts it. Infrastructure, tooling, protocols - these categories need distribution and developer trust before they need revenue. CopilotKit is a good example of this. Their bet on the AG-UI protocol only makes sense as an open play. A proprietary protocol that nobody builds on is not a protocol. It’s a dead end. If you need to establish a standard before a competitor does, open source is often the only credible path.

A recent GTMfor.dev podcast from Essence VC had a great conversation with Ryan Blue, the creator of Apache Iceberg and co-founder of Tabular, and his framing on why Netflix open sourced Iceberg in the first place. It was self-interest. They knew that if they didn’t do it in open source, they’d end up with a format specific to Netflix that would never be as good because you don’t get the benefit of other people doing the work. As Ryan put it: “The project unlocks the market. We are a product within that market.”

Being open is a trust signal that no amount of marketing can replicate, especially in a space where developers have been burned by close-sourced vendors. If you have genuine conviction that the open layer will drive usage of a proprietary cloud or enterprise tier, the economics can work with patience and discipline.

You will have users before you have customers, and converting one to the other without a sales motion in place is harder than it looks. The other risk is the Elastic problem. If you release under a permissive license and your product gains traction, a cloud provider can package and sell it without paying you anything. AWS did it to Elastic, to Redis, to MongoDB. And community work is genuinely expensive in ways founders underestimate early. Documentation, DevRel, issue triage, keeping contributors happy.

When OSS as a later product makes sense

The case for going commercial first and open sourcing later is simpler: you know more. You have paying customers who told you what they actually need. You know which parts of the product are defensible and which parts are commodity. You’ve earned the right to invest in the community because you’ve already validated that the thing you’re building has value.

It also makes sense when you want to expand into a new segment like self-hosted, mid-market, developer-led adoption without standing up a new sales motion. Open source can do that work for you, but only if the product is already proven.

The risks here are different. Late open source releases can feel cynical to a community that watches these things closely. If it looks like you’re open sourcing because growth slowed or a competitor went OSS-first, developers notice. Authenticity is not optional in this community. And if a competitor already went OSS-first in your category, you may be fighting uphill to earn trust you could have built from day one. I’ll say that if you are early in your lifecycle, say Seed stage, it’s less of a risk.

OSS-first optimizes for distribution. OSS-later optimizes for defensibility. Most founders need both eventually. The question is which one you need more urgently right now, and whether you have the capital and runway to build the other one before it becomes a problem.

What to open source and what to keep closed

The canonical model has a name: open-core. Open source the thing developers need to get started. Close the things enterprises need to sleep at night.

The framework is less about features and more about who needs what. Developers want something they can install, run, and trust like the core runtime, the SDK, the integrations, the CLI, whatever lowers the barrier to building a habit around your product. Procurement teams don’t care about any of that. They’re asking whether it’s safe to deploy at scale inside a company with real compliance obligations. SSO, RBAC, audit logging, deployment controls, non-negotiable for a CISO signing off on a purchase order. That’s where the revenue comes from. Keep it proprietary.

Without them, a security team will kill the deal before it reaches procurement. Audit logging matters for compliance-heavy industries. SLAs and dedicated support matter because enterprises are not willing to post a GitHub issue and wait. On-prem and VPC deployment options still matter more than most cloud-native founders expect, especially in regulated industries like finance, healthcare, and government.

Don’t open source your moat. If the algorithm, the model, or the data pipeline is what makes you defensible, keep it closed. The open layer is for distribution. Something else has to be for defensibility.

One useful test: ask yourself whether a developer could get real value from your open source layer without ever paying you anything. If the answer is no, it’s not really open source. It’s a feature-locked demo. Developers figure that out fast. Before you decide where to draw the line, Heavybit’s guide is the clearest breakdown I’ve found for founders who aren’t licensing lawyers.

Transitioning to enterprise is harder than you think

At some point, if the OSS motion is working, you will start hearing from companies you didn’t go looking for. The signals are usually clearer than founders expect. Inbound from companies with real procurement processes asking about security, compliance, or deployment control. GitHub org-level installs at accounts you recognize. Champions starting to ask not whether the product works, but whether it can survive an InfoSec review.

The problem is that most teams get here having been heads down on OSS (shipping, iterating, growing the community) and suddenly realize they have none of the enterprise readiness infrastructure in place. Building that muscle while also maintaining the OSS community is genuinely hard, and most teams underestimate how long it takes to do both well.

The harder signal to read is the champion-to-economic-buyer gap. Your OSS champion is almost never the person who signs the check. They are an engineer, a team lead, maybe a VP of Engineering if you’re lucky. The economic buyer is a CTO, a CISO, a CFO, sometimes all three in a room together. The champion got themselves convinced. Your job is to give them everything they need to convince everyone else. That is a completely different sales motion than anything that worked to get the open source community to adopt you.

The other hard part is figuring out how to intentionally go after the champions already hiding in your user base. You have thousands of stargazers. Some of them are engineers at Fortune 500 companies who would pay for enterprise support tomorrow if someone asked them the right question. The problem is most of them signed up with a Gmail address and you have no idea who they are or where they work.

This is a solvable problem but it requires deliberate effort. A few things that actually work: pull your stargazer list via the GitHub API and run reverse IP or email enrichment tools like Clearbit or Apollo against any work email addresses you do have from cloud signups or doc registrations. Set up retargeting against your GitHub and docs traffic. The person reading your deployment guide three times in a week is worth a LinkedIn impression.

Build a lightweight PLG motion into your cloud or hosted tier that captures company affiliation at signup. The OSS community got you here on pull. The enterprise motion requires some push using resources and moments to get those champions to show up and tell you who they are - and just need someone to help them make the internal case.

If you want to go deep on this transition, pricing, and more, the WorkOS Crossing the Enterprise Chasm episode with Armon Dadgar is a fantastic resource. HashiCorp’s path is instructive; they started with entitlement licensing, the model enterprises were already comfortable with, and only gradually migrated toward consumption-based cloud pricing as customers built comfort with cloud delivery. Meet buyers where they are. Build trust. Move them over time

Final Thoughts

If you’re asking yourself if you should open source part of their product, ask yourself the following: what happens the day after it works? What do you sell, to whom, and how do you find them when they’re all hiding behind Gmail addresses?

Open source is not a strategy by itself. It’s the first move in a longer game. The founders who do it well know exactly what the second and third moves are before they make the first one.

Basecamp updates

What’s happening in the portfolio

💰Cognee raises $7.5M Seed

AI memory for agents - a self-improving memory graph that takes stateless agents to durable, structured knowledge. Led by Pebblebed (Pamela Vagata, co-founder of OpenAI), with participation from 42CAP and Vermilion Cliffs Ventures. 12,000+ GitHub stars and running in 70+ companies in production.

🚀 Paper Compute launches out of stealth with StereoOS

Sandboxed agent execution with full observability. stereOS isolates agents in ephemeral VMs. tapes records every LLM call and decision to a black box. Built for teams that need to run AI in production without losing the audit trail.

🚀 Mach9 launches Digital Surveyor 2

AI-powered 3D mapping and feature extraction for CAD workflows - extracting curbs, utility poles, paint lines, and road assets from LiDAR data up to 100x faster than manual methods. Trusted by transportation agencies, telecom, and survey firms. Over 23,000 miles extracted to date.

From the trail

The latest from our team

Who’s hiring

Exciting roles across the Vermilion portfolio

The latest roles across engineering, marketing, and developer relations are on the Vermilion Careers page.

View open roles →

Upcoming events

Where Vermilion will be in the next month. Looking forward to seeing you in person.

🌠 AI Infra After Dark | With Essence VC & MongoDB Ventures · Lower East Side, NYC · Wed, April 22nd RSVP here

Event recaps

A look back at two events from March worth recapping.

Base Layer Dinner Series · New York

A small, deliberate dinner with Ethan Zou of MongoDB and a table full of people who have spent their careers building technical products. The conversation around the direction of AI developer tooling went well past dessert.

View recap on LinkedIn

RSAC After-Hours: Securing Autonomous Systems · San Francisco

Co-hosted with Citizens, Essence, and Modern Technical Fund at RSA Conference. The panel dug into what it actually means to secure agents operating with delegated authority. Prompt injection, dynamic permissions, and why existing IAM systems weren’t built for this.

View recap on LinkedIn

Thanks for reading Vermilion Cliffs Ventures | Ashley Smith. Subscribe for free to receive new posts and support our work.

Subscribe now

I mentioned this one at the end of my last piece on round names being meaningless so I’m going to try and pick up from there.

Taxonomy

Before I get into why, let me define the terms. Because part of the problem is that founders conflate these investor types and end up thinking they have more support than they do.

Lead Investor

A fund that sets the terms, anchors the round, and takes enough ownership to have a real economic reason to care about your outcome. They do the diligence. They negotiate the cap and the docs. At seed, a lead typically owns somewhere between 10-20% post-money. They are the person you call when something breaks. They may take a board seat at seed, more commonly at Series A. This is the relationship you are actually trying to build when you raise a round. Everything else is supporting cast.

Supporting Fund

A fund that joins a round after the lead has set terms usually as the second largest check. They are not setting the price. They are filling out the round alongside a lead because they believe in the company and want the exposure. A good supporting fund still brings value through intros, pattern recognition, and sector expertise. They are genuinely helpful when the lead is doing their job but they are not a substitute for the lead. We mostly fit in this bucket as we are not a lead investor.

Strategic Angel

An individual, usually a former operator, founder, or executive, who writes a check because they have something specific to offer beyond capital. A former CISO writing into a security company. A VP of Sales who built the GTM motion at a company you want to emulate. The check size is almost always small, $10K-$25K, but the relationship is valuable. The key word is strategic. An angel with direct, relevant, current-market knowledge in your specific problem space is worth having on your cap table at a tiny ownership percentage. An angel who just wants to be involved in startups is a line item.

Scout-Style Check from a Larger Fund

This is the one founders misread most often. A large multistage fund managing several billion dollars will sometimes write small checks into early companies to get access, build a relationship, and reserve the right to lead the Series A and beyond. The check might be $250K into a $30M post. That is .8% ownership. The large fund does not care about the return on that specific check. They are buying an option on leading the next round. If you grow the way they hope, they come back with a real check at the Series A. If you don’t, the small check was a cheap way to watch the company develop. This is not a lead. It is not support in the way a seed fund provides support. It is a financial option dressed up as an investor relationship. It can be useful, the brand name, the signal to other investors, the eventual path to a larger check, but you should know exactly what it is.

Party Round

A party round is what you get when you have no lead and fill the round with a combination of supporting funds, angels of varying quality, and scout checks from larger funds who want optionality. Terms are set loosely or by whoever wrote the first check. Ownership is distributed thin across many investors. No single investor has enough at stake to prioritize your company when things get hard. It feels like momentum when you’re closing it. It becomes a liability when you need a bridge between Seed and Series A.

Everyone comes to the party. No one is responsible for cleaning up afterward.

That last part is the problem.

The distinction that matters most across all of those types: ownership drives incentive. A fund with 12% ownership in your company has a direct financial reason to help you. A fund with 0.5% ownership is rooting for you but has 100 other companies competing for their attention. Understanding where each investor sits on that spectrum tells you everything about what you can actually ask of them.

How this looks in the data

At pre-seed, a collection of smaller checks is pretty normal and mostly fine. You’re getting started. You’re taking angel money, maybe a small fund or two, figuring out if you have something. Nobody expects a full institutional lead at a $10M post.

The party round problem is not a pre-seed problem. It’s a seed problem. We don’t lead at seed but I’m watching friends and founders around me make the party round mistake.

The data within Carta and a16z State of Seed report for Winter 2025 solidifies a few points for me. The median seed post-money valuation is now $20M. The 75th percentile is $30M. And the 95th percentile has jumped to $80.5M. The gap between the median and the upper end is widening sharply every year. More capital is going into seed, but it’s going into fewer rounds. 2025 is on pace for roughly 2,165 seed rounds, down from the 2022 peak. More money. Fewer companies. Higher prices.

Here’s what that means for founders. After a priced seed round, the median founding team owns 56.2% of their company. After the Series A, that drops to 36.1%. You are giving up 20 points of ownership between seed and Series A even in a clean, well-structured round. In a party round with multiple SAFEs at different caps stacking on top of each other, the dilution math gets messier and founders routinely arrive at Series A having given away more than they realized. That is a hard place to negotiate from.

The median time between seed and Series A is now 2.1 years. For AI startups it’s slightly faster at 1.9 years. And only about 20% of seed companies make it to Series A within the first two years, most take longer, or don’t get there at all. That’s a long time to be building without a partner to help you when you need help. That gap or what happens in those 2.1 years is worth a piece of its own.

What that actually looks like in practice

The lead investor definition above sounds clean.

A lead will tell you when your hiring plan is wrong. They’ll make calls for you when you’re trying to close a customer and need a warm intro. They’ll bridge you if your runway gets short and you’re three months from a real milestone. They’ll push back on decisions that they think are going to hurt you even when you don’t want to hear it. Sometimes especially then.

None of that happens in a party round. In a party round, no single investor has enough ownership to make your success a priority. A $150K check into a $25M post-money valuation is 0.6% ownership. That investor has no economic incentive to spend meaningful time on your company. They are managing a portfolio. You are a line in a spreadsheet. When things get hard, there is no one to call.

I asked a founder friend of mine that has been through this a few times to give examples of the lead vs. party round dynamic. His response below:

There are of course funds that take much less ownership that over deliver. But that’s not the standard.

The cap table problem

Party rounds create messy cap tables and messy cap tables create problems at the next raise.

When a Series A investor runs your cap table, they are looking for a few things. They want to see that existing investors have pro-rata rights they’ll actually exercise. They want to understand who has information rights and what that means for disclosure. They want to know there’s a lead who can speak for the existing investor group. They want a clean structure they can build on.

What they do not want to see is seventeen investors under $500K with a mix of SAFEs at different caps, a couple of convertible notes, and no clear lead. That tells them the fundraise was reactive, not strategic. It tells them you optimized for getting the round done instead of building a long-term investor relationship or that you couldn’t secure one. It makes their due diligence harder. Some will pass just because of the complexity.

And the ownership math is unforgiving. The Carta data shows the median step-up from seed to Series A valuation is only 2.6x right now, well below the 4x+ of the 2020-2021 era. That means there is less buffer between what you raised at and what you need to prove to get the Series A done. If you’re coming in with a messy cap table and a founder ownership stake already compressed by stacked SAFEs, you’re negotiating from weakness before the conversation even starts.

The failure scenario

Here’s where it really breaks down.

Let’s say you’re 18 months in. You raised a $4M party round at a $25M post. You’ve been building hard. You have some traction but not enough to make a clean Series A case. You need a bridge to get to the next milestone.

In a normal investor relationship, you call your lead. They assess the situation. If the business is worth saving, they pull together a bridge, sometimes alongside other investors, and give you the runway to get there. They have enough ownership and enough conviction to make that call.

In a party round, you send fifteen emails. Some people will respond quickly. Some won’t respond at all. A few will say yes to a small amount. Most will want to know who’s leading the bridge. Nobody wants to lead the bridge because nobody has enough ownership to justify leading. The bridge either doesn’t happen or it happens slowly and messily and you lose months of runway just trying to close it.

I’ve watched this over and over. It doesn’t end well.

The counterarguments we must acknowledge

I want to be fair here. There are legitimate reasons founders choose party rounds and not all of them are naive.

The first argument is optionality. If you’re building something where you’re not sure which of your investors has the best network for your next raise, spreading the cap table means you have more doors to knock on. Fifteen investors means fifteen potential intros to Series A funds. That’s not crazy logic. I’ve seen it work.

The second argument is control. If no one has a board seat and no one has outsized ownership, you maintain more operational independence. For some founders, especially repeat founders who’ve had difficult lead investors before, that tradeoff feels worth it. The freedom to make decisions without negotiating every major choice is genuinely valuable.

The third argument is speed. Party rounds close fast because you’re not waiting on one investor to complete full diligence and get to conviction. You’re aggregating smaller checks from investors who are making lighter-touch decisions. In a market where timing matters, that speed can be real.

The fourth argument is that the party round reflects market reality for a lot of founders. If you’re not in the small circle of companies getting institutional seed leads at $30M post, the party round might not be a strategic choice so much as the only path available. Access to strong lead investors is not equally distributed.

Where I get stuck in my head

Even with these counter arguments, I still think party rounds are a trap for most early-stage founders.

The speed to close is encouraging but the support isn’t. The optionality is nice but the accountability isn’t. Control is important but help isn’t there. You are trading long-term partnership for short-term convenience and you won’t feel the cost until things get hard, which is exactly when you need the partnership most.

When I talk to founders about this, I try to say it simply: the best thing that can happen in a party round is you grow so fast it doesn’t matter. The worst thing that can happen is you hit a rough patch and discover there’s nobody to call. The space between those two outcomes is where most companies actually live.

What I tell founders

If you can get a real lead at reasonable terms, take it. Even if the valuation is lower than what a collection of angels might give you. The lower valuation with a lead partner is almost always a better outcome than the higher valuation with fifteen check-writers who have no reason to prioritize your company.

If you genuinely cannot get a lead, be honest with yourself about why. Is it the stage? The market? The traction? Or is it that the right investors haven’t seen it yet? There’s a difference between “I couldn’t get a lead because I’m too early” and “I couldn’t get a lead because the investors who’ve met me aren’t convinced.” One of those means do a smaller pre-seed and get more traction. The other means do more work on the business before you raise.

If you do end up in a party round, be strategic about it. Pick one or two investors from that group who you think have the most to offer and cultivate those relationships proactively. Call them. Update them. Pull them in on decisions. Try to create the lead dynamic even if the cap table doesn’t have one. It won’t be the same but it’s better than nothing.

And above all: do not let the party round close at a valuation you can’t grow into. Because when you come back for the Series A and the traction isn’t quite there, you’ll need someone to fight for you. In a party round, you’re fighting alone.

Next up: Why the time between Seed and Series A is actually the most important period in a company’s life and how to set yourself up to succeed.

The reality is that round names are sort of meaningless right now. And I’m not sure they should mean anything at all anymore.

TechCrunch published a piece today that I contributed to, and I want to go deeper on a lot of my reasoning on the quotes. It’s weird to see myself in the outlets I’ve always helped my founders get mentions and coverage in. I don’t hate it.

I’ve spent my career at the intersection of operators and investors. Twilio in the early days long before the IPO run. GitLab when we were scrappy and figuring out what enterprise open source even meant. GitHub through the Microsoft acquisition. I’ve watched markets cycle and I’ve watched founders make decisions about their companies based on what a round was called instead of what it actually needed to accomplish.

So if round names don’t matter, what is actually going on in the market?

Pre-seed is sitting around $10M post. Seed is around $25-30M post, which is where the Carta data lands for this quarter. YC Demo Day was last week and most companies were raising at $40M post for a Seed. I was there, talking to founders, watching the number climb.

Those aren’t seed rounds. Not by any definition most of us were trained on. But who cares?

A $5M check into a $40M post-money valuation for a company that is eight weeks old, with pilots that haven’t had time to churn or stick, is a different financial instrument than what we used to call a seed round. The label is the same. The risk profile is completely different. For the big funds writing scout style checks into 10-20 party rounds each cycle, totally makes sense. For YC’s LPs that get exposure to the 2-3 outliers per cycle, totally makes sense. For funds like mine? It’s getting harder to manage that risk. The founders this batch were exceptional. The why behind the companies were exceptional. It is still a hard check to write if you’re a small early stage fund.

Part of what’s driving the numbers is that a few companies have genuinely rewritten the rules on how fast traction can materialize. Cursor (holy shit what a run) hit $100M in revenue in 12 months. Lovable, Bolt, ElevenLabs, there’s a cohort of companies that showed what AI-native velocity can look like. The problem is the market is now pricing every company as if it’s Cursor but in reality, the normal growth patterns are actually what is happening in the underlying non-Cursor world. The outliers are setting the floor, not the ceiling, and that math doesn’t work for most founders.

I’ve seen things further out than that. A $300M seed closed recently but super amazing founder with a great product space. Likely a good bet. But not a seed. I pointed to it when the Dominic asked about extremes. While not a seed, it is a bet that a single company will become base infrastructure for an entire era of technology. The round name is a rounding error on the thesis.

Why this is happening

So many different winds are blowing in the same direction that it’s hard to separate them, but I’ll try.

The first wind is the mentality shift around platform bets. Investors believe a small set of winners will become the base layer of the AI era, the next AWS, the next Google. If you believe that, price almost doesn’t matter, because the returns on a genuine infrastructure monopoly dwarf anything you gave up on entry valuation. Wildly, the logic makes sense. It’s just not applicable to most companies, and the market is applying it broadly.

The second wind is large funds moving earlier. If you’re managing a $5B fund and you want meaningful ownership in the companies that define AI infrastructure, you either write a lot of small checks or a few massive ones. Most are going with option two. A $50M check into a $300M seed round is a rounding error in a multi-billion dollar fund. For that fund, it makes sense. For everyone else trying to benchmark off that number, it creates a distorted picture of what the market actually looks like.

The third wind is YC. I’ll say what investors say quietly in rooms: there is a YC tax. YC sets the benchmark. I’ve watched it happen quarter over quarter, they price their batch, the SF market follows within a quarter or two, the national market catches up after that. Right now, most companies out of Demo Day were pricing at $40M post. I heard some went higher. I was standing in a room with other investors and multiple people said out loud: this is priced too high. The network, the alumni relationships, the downstream signaling from YC are all great value adds, truly, they are an exceptional group. It doesn’t change whether a team with eight weeks of runway can grow into a $40M valuation before they need to raise again at an even higher valuation. Are we setting a lot of those teams up to fizzle out because they can’t hit the higher expectations? I think so.

There is also a talent premium baked into the most extreme numbers. Investors are paying extraordinary prices for founders with the right pedigree. Repeat founders, ex-OpenAI, the names that carry their own meaning in a room. One investor quoted in the TechCrunch piece put it plainly: it’s a war for great researchers right now. That’s true. But the risk is that the premium applied to a handful of genuinely exceptional people bleeds into the pricing logic for everyone building in adjacent spaces. The market starts conflating founder brand with company value, and those aren’t the same thing.

What this means for small funds like mine

I run a small, specialized fund. We invest at pre-seed and seed in AI infrastructure. My check sizes are $750K at pre-seed, around $1M at seed. I won’t go above a $60M post-money valuation, and even that is the high end. The math stops working for LP returns above that number if you’re a fund our size. But the math works for larger funds.

What this market has done is push us earlier and out of some of the higher priced “seed” deals that are getting A and B prices. Not because we changed our strategy, but because if the first round a company raises is already priced at $40M with no product and no traction, I can’t write a check into that. So we are backing founders before that round exists. Before the product. Sometimes before they’ve left their current job. We give them enough capital to hire a couple of engineers, get some pilot customers, and start building. Then we help them raise the larger seed from funds with bigger pools of capital.

This only works because we are highly specialized. Our network runs deep in AI infrastructure. We can be useful in ways that generalist funds can’t be at that stage. If you’re trying to be a generalist fund at our size right now, I think that’s a very hard path. The market is telling you to either be specialized and early or generalist and large. As i’ve said many times before, the middle is hard. Anyway, if you’re thinking about building in AI infra or know someone that is, send them my way. :)

What this means for founders

So this is more complicated.

The most important thing I can tell any founder right now is: the headline rounds are outliers, not the market.

When you see a $300M seed, that is not a benchmark. When you see $40M YC pricing, that is a specific set of companies with a specific set of investors making a specific kind of bet. The data for everyone else shows pre-seed sitting around $10M post and seed around $25-30M post. I pulled the 2025 data from Carta, Sapphire, and others, and it all points in the same direction.

My biggest warning: pricing your round too aggressively too early is one of the fastest ways to make your next round harder. You are not just setting a valuation. You are setting a bar you have to clear in 18 months to raise again. And Series A is at record-high valuations right now, the hardest to achieve in many, many cycles. Time between rounds is the longest it has been in a long time. If you price yourself into a corner and the market resets or growth takes longer than expected, which I see happen every day, you are stuck.

Jonathan Lehr of said it well in the TechCrunch piece: higher seed valuations mean less margin for error, less room for experimentation, less tolerance for pivots, and more scrutiny if progress doesn’t match the capital raised. I’d add that this dynamic hits first-time founders hardest. When capital moves this fast, access matters more than merit at the start. Repeat founders and well-networked teams get the benefit of the doubt. First-time founders have to work harder to break through, and then they get less runway to prove themselves once they do.

Raise what you need to get to the next meaningful milestone. Optimize for working with the best investors, not for hitting the highest number. Early valuation is not a win. It is a constraint you carry into every conversation that follows. And I need to write a piece on Party Rounds and why they can also set founders up to fail. Another day.

The bigger question

Dominic asked me at one point whether we should just retire round names altogether. I said yes. It might be time.

Round names exist to create shared benchmarks, a way for founders, investors, and LPs to align on what stage a company is at and what the risk profile looks like. When a seed round can mean anything from $1M on a $5M post to $300M with no traction, the benchmark is gone. You’re not benchmarking anymore. You’re just making it up as you go. Which is really sort of fine? Isn’t that what early stage venture and startup building is anyways? Anyone that says otherwise is bullshitting.

And storytelling is fine. Narrative has always been part of venture. The problem is when we start using it to price rounds instead of to describe them. When “this could be the next AWS” becomes the reason a pre-product company is valued at $150M, we’ve crossed from investing in potential to investing in hopes and dreams. Sometimes fiction turns out to be true. Most of the time it doesn’t, and the founders are the ones who pay for it but they also are the easiest teams to back on attempt 2 and 3.

The round name is dead. What actually matters is: what do you need, what will you do with it, and what does your next milestone look like when you come back to raise again?

Get those three things right. The label on the round doesn’t matter at all.

Dominic, thanks for being so thoughtful with your questions and your writing. Excited to see where round names and valuations go next.

First, a little history for anyone that didn’t come up during the earliest/earlier days of the public internet. Around 1991, Tim-Berners Lee made the internet publicly accessible. I really started finding my way around the internet probably around 1994 from school computers and my actual addiction to the internet kicked off around 1997 which I detailed in painstaking length when I launched the fund.

The consolidation of online communities

For my 2000’s babies, there was a time in the mid-2000s where we had a set of community driven platforms emerge and Reddit and Digg (okay and 4chan) were the ones that really caught traction. Within the developer community, Hacker News was the center of the universe. For everything else, Reddit and Digg upvotes were driving what people were exploring online.

Screenshot from the front page of Digg circa 2008. It is not lost on me that the ad displayed is to help you get more fans on FB or followers on Twitter. Evil foreshadowing.

I think what was super special about that time on the internet for me (any many others) was that it was a place where all of your interests would combine under one umbrella and it was organic. You were into indie music, math, yoga, and cooking? There was a subreddit for that. You could find your people. You could find conversations that were authentic.

As a child of the internet, I’ve been addicted to online communities since I got my first computer at the age of 12. The connecting thread between the chatrooms of the 90s (Yahoo Chat please come back) and the message boards of the 2000s was always humans interacting around a topic, interest, or location that everyone had in common, for better or worse.

A lot of these interactions were based in anonymity and while a lot of us made offline friends out of online friends from those times (I still have several that I’ve now known for over 25 years), it was a great way to connect.

Social media and the de-anonymization of the internet

Social media came along and de-anonymized the internet. Suddenly, we knew what every person in our social circles were doing. All the time. Then came the monetization of social media and the influencer of it all.

Social media and message boards always felt pretty separate from one another and then the algorithms and content machines came along and it felt like the fake profiles and bots started to grow. Bots had always been present but it started to feel malicious in the mid 2010’s. People were buying followers and likes and even worse people were influencing elections. The bot economy ramped up.

The pandemic changed how people interacted socially, both offline and online. TikTok and Instagram pushed things even further. Algorithms started delivering content that felt like community. But the monetization layer made it feel less organic. This felt like some blended world of social media and message board communities.

Nothing is authentic online anymore

So here we are in present day. AI agents came along and content is getting ruined by bots using AI to generate content then promote it across all the social networks and online communities. It’s sort of hell. The internet which once felt so free and authentic is now full of fake people (non-human, not just influencers) and inauthentic content supercharged by AI agents.

This brings me back to Digg. Around 2010 Digg did a terrible redesign and killed their product. Everyone left for Reddit. Digg died a slow death. The learning here is don’t change your entire product and ignore the VERY LOUD complaints from your community.

Fast forward through social media’s rise (and fall), influencer culture, algorithms, bots, and now AI agents, and we’re in a time when I think everyone is craving authenticity and communities that aren’t full of AI generated content, influencers, and bots, DIGG ANNOUNCED IT WAS RE-LAUNCHING. Maybe this was the fresh start we all needed online?

Or not.

After 3 months in beta Digg shutdown again. The message on their homepage is sobering.

via Digg:

Building on the internet in 2026 is different. We learned that the hard way. Today we're sharing difficult news: we've made the decision to significantly downsize the Digg team. This wasn't a decision made lightly, and it's important to say clearly: this is one of the strongest groups of people we've ever had the privilege of working with. This is not a reflection of their talent, their effort, or their belief in what we were building. It's a reflection of the brutal reality of finding product-market fit in an environment that has fundamentally changed.

We faced an unprecedented bot problem

When the Digg beta launched, we immediately noticed posts from SEO spammers noting that Digg still carried meaningful Google link authority. Within hours, we got a taste of what we’d only heard rumors about. The internet is now populated, in meaningful part, by sophisticated AI agents and automated accounts. We knew bots were part of the landscape, but we didn’t appreciate the scale, sophistication, or speed at which they’d find us. We banned tens of thousands of accounts. We deployed internal tooling and industry-standard external vendors. None of it was enough. When you can’t trust that the votes, the comments, and the engagement you’re seeing are real, you’ve lost the foundation a community platform is built on.

This isn’t just a Digg problem. It’s an internet problem. But it hit us harder because trust is the product.

Building social is hard, incumbents are harder

We underestimated the gravitational pull of existing platforms. Network effects aren’t just a moat, they’re a wall. The loyalty users have to the communities they’ve already built elsewhere is profound. Getting people to move is a hard enough problem. Getting them to move and bring their people with them is something else entirely.

So where does that leave us? Is this what we get for monetizing attention?

AI, agents, bots, and inauthenticity are starting to feel like a core societal problem, not just a product problem. When you can’t trust that the posts you’re reading, the comments you’re replying to, or the engagement you’re seeing are actually from real people, something fundamental about the internet breaks.

Community platforms only work if the participants are human.

Harvard has been publishing reports about the epidemic of loneliness in America, and it’s hard not to see the connection. The internet was supposed to connect us. For a while, it did. But if more and more of what we interact with online isn’t real, it starts to have the opposite effect. You feel like you’re participating in something social, but the room is half (maybe more?) bots.

The irony is that the technology industry helped build the systems that are now eroding the very communities that made the internet magical in the first place. It’s all sort of collapsing in on itself.

Which makes me think (and hope) something else is about to happen.

If the internet is less trustworthy, the value of actual human interaction goes up. This is why we are seeing so many events and in person events in tech (and beyond tech). Events are fun again. People are getting creative with the meetup style community events we saw begin to emerge in the mid 2000s and really spike in the early 2010s.

People are meeting people IRL again. It’s pretty nice, even for an introvert. Apparently a walk down the West Side Highway in NYC or Marina Green in SF (The Sluttiest Mile on the West Coast) is the easiest way to get a date these days. We even ruined online dating which you all already know if you’ve talked to any of your single friends trying to sift through Hinge or Grindr. Romantic wasteland.

I think the swing back towards in person gatherings, meetups, and communities where you know the people around you are actually people is a welcomed change after living in the false reality that’s been created on the internet lately.

The early internet was magical because it helped humans find each other.

Right now it feels like we’re losing that. Facebook just acquired Moltbook, a social network for AI agents.

Someone please solve the bot problem before the internet becomes a ghost town full of machines talking to each other. Or at least create what comes next.

Until then, I’m trying to read more books, meet strangers in person, and turn my iPhone into a flip phone. Fuck, I might get a landline.

Ashley, what does this have to do with VC, investing, and developers? Everything. And maybe just a little bit of me entering my Old Man Yells At Cloud era. I’m tired of seeing the printing press analogy but it might be real. What is next.

I copied and pasted this entire rant into ChatGPT and told it to create a timeline visual. It made this hideous sales Powerpoint-looking visual circa 2009. Deep belly laugh. Thanks Chat.

This was ChatGPT’s second try after I shamed it for using old Powerpoint templates. Sigh. Thanks for the resume styling.

All of these views can be true at once. Capital is flowing, particularly into AI, but access to that capital has narrowed. Fewer firms are consistently investing, LPs are concentrating commitments, and the number of rounds has not kept pace with the dollars being deployed. From the outside, the market looks healthy. From the inside, it feels selective, slower, and harder to navigate.

Here’s my hot take: This is not a bad venture market. It is a narrow one. And that distinction matters for founders choosing partners and for funds deciding how they want to build through this cycle.

These experts did incredible research. I wanted to pull the most important signals from the data, and what I think they mean in practice for founders and emerging funds.

Why Venture Feels Confusing Right Now

AI investment is strong (investment dollars rebounded the last two years due to this) and optimism is returning to conferences, headlines, and fundraising decks. Yet most founders report fundraising is harder than it has been in years.

Processes are slower. Meetings are harder to secure. Conviction is rarer. The gap between interest and actual checks has widened. This disconnect exists because capital availability and capital access are no longer moving together. The market is deploying more dollars, but through fewer firms and fewer rounds.

Venture Is Contracting for the First Time in a Generation

Despite what looks like a large and active venture ecosystem, the number of firms consistently investing has declined meaningfully. One of the clearest takeaways from Beezer’s analysis is that the venture industry is no longer expanding by default. While thousands of firms still exist on paper, fewer than half are meaningfully active.

Zombie firms - making zero investments in the last four years - inflate the perception of market breadth without contributing to actual capital formation. And now LPs are consolidating commitments into fewer managers, raising the bar for who remains active.

This is the first time since dotcom that new funds are not keeping up pace with attrition.

Capital Is Flowing but Access Is Narrowing

Carta’s data makes the lived experience concrete. Total dollars invested have increased, but the number of rounds has not. Fewer companies are raising even as more capital is deployed overall.

This shows up in cluster outcomes. Strong companies still raise and the top of the market remains competitive but the middle thins quickly. Companies that would have raised before now struggle to find traction.

For founders, this shows up as longer fundraising timelines, more “wait and see” feedback, and fewer second chances.

The Seed Market Is Mis-priced

Seed valuations have risen faster than any other stage, even as graduation rates have declined. The traditional logic of paying for early risk has inverted and it’s the most fragile part of the current market.

Equal Ventures highlights that, on a risk adjusted basis, later stage investing currently offers better pricing than seed. Carta’s data reinforces this, showing higher dilution earlier and fewer companies successfully moving through the funnel.

For founders, high seed valuations can feel like a win. In reality, they often pull future risk forward. For funds, seed has become harder to underwrite unless there is genuine conviction and long term alignment.

Two Markets Are Running in Parallel

The thing is that AI aligned companies are operating in a different market. They raise faster, at higher prices, and with broader investor interest. Capital is flowing disproportionately toward AI across every stage.

At the same time, strong non AI businesses face much tighter conditions, regardless of fundamentals. This does not mean those companies are weaker. It means category positioning has become a gating factor for access to capital.

Understanding which market you are in matters more than ever.

Just a note on the AI optimism, especially called out in A16z’s report. Many bullish perspectives correctly identify AI as a generational platform shift. Technology markets will be large. New categories will form. Value creation will continue.

But large markets do not guarantee broad venture outcomes. Capital still concentrates. Liquidity still bottlenecks. Power laws still apply. The question in this cycle is not whether AI will matter. It is who gets paid, when, and under what constraints.

What This Means for Founders

Fundraising is now a partner selection problem more than a pitch problem. Who you raise from matters as much as whether you raise.

Valuation discipline matters. Durability matters. Default alive paths matter. Founders should assume fewer rounds, not more, and build companies that can survive financing cycles rather than depend on them.

Carta’s data shows founders already adapting, hiring later, staying leaner, and preserving ownership longer. I believe that instinct is right.

What This Means for Emerging Funds

Undifferentiated strategies struggle in concentrated markets. Survival favors conviction, consistency, and real partnership.

Very few funds make it from Fund I to Fund IV and fewer still become enduring franchises. The funds that survive are not the loudest. They are the most deliberate.

Small focused funds can still win in this environment, but only by being clear about what they do, why they exist, and how they help founders build through cycles.

Closer

This market rewards clarity over confidence. Founders who assume abundant follow-on capital will struggle, while those who build durable businesses and choose partners able to stay engaged through multiple cycles will preserve more optionality than they expect. Valuation still matters, but survivability matters more, and survivability is built through real product usage.

For many companies, that means prioritizing pilot customers, design partners, and early launches over perfect positioning. Getting something into production, learning quickly, and proving that a specific customer will actually rely on what you are building is now the fastest way to create leverage. Momentum comes from use, not narrative.

For funds, this is a separating cycle. Capital concentration makes undifferentiated strategies harder to sustain and places a premium on conviction, consistency, and real partnership. Smaller focused funds can still win if they are deliberate about where they invest, how they show up, and how they help founders move from idea to adoption faster.

The firms and founders who treat this as a craft cycle rather than a volume cycle, and who build alongside real customers early, will shape what comes next.

Vasilije came to this through a genuinely unusual path: business school in Italy, a career in data engineering across German unicorns, and then a left turn into cognitive science, where a course on higher cognition sparked the idea for Cognee. He built a prototype, it got traction, and here we are. The background in cognitive science is not incidental. It shapes how he thinks about what memory actually is and why the way most AI systems handle it is fundamentally wrong.

As AI agents move from demos into production, the systems underneath them are not ready. Not the models, those are extraordinary right now. The problem is everything that surrounds them. Dig in with Vasilije’s perspective on this below.

Full interview here with highlights below.

The data model breaks the moment agents enter the picture

When teams finally push AI agents from sandbox to production, the first thing that breaks is rarely the model. It is the data model underneath it. Every traditional data warehouse was built by humans, maintained by humans, and updated on a human timeline. Agents do not work on that timeline.

“With the LLMs, these data models go out of the window. Because they are producing so much diverse inputs. You’re processing all kinds of data that was never processed before, like PDFs and unstructured data. And all of these things, you need a way to really dynamically define how these data models look like and make them evolve because no one can humanly analyze and keep these updated in the way they analyzed it in the past, especially if you have a million agents doing some type of operations. This whole system needs to be automated and outsourced to one external system.”

The latency problem is equally real. Processing a single pipeline run used to cost $325. It is now fractions of a cent. But inference latency is still a bottleneck, and designing systems that can manage it gracefully at scale is something most teams underestimate until they are already in production.

What actually changes when memory becomes persistent

The RAG approach was a meaningful step. But it was designed for a world where you load data, forget about it, and search it when you need it. That model is already showing its limits. Vasilije traces the progression clearly:

“ I think embeddings are a good building block, but they are not solution in itself. What we are trying to do is we're trying to create this graph on top that gives us all the relationships and possible interaction points of these embeddings. And we then get another layer of meaning. And these can be layered up… merged and cross-analyzed by different other algorithms. And that gives us a lot of flexibility in the design. I think the RAG approach went into GraphRAG, which went into the dynamic memory optimization. Now we’re working with Berkeley and Brown to actually have this predictive memory that we’re predicting and paths of possible futures and then merging them. Sounds a bit like minority report type of a thing, but it’s interesting.”

The underlying shift is that embeddings alone cannot capture relationships, time, or evolving domain knowledge. Cognee builds a graph on top that encodes how pieces of information relate to each other, layered and cross-analyzed in ways a vector store cannot do on its own. The goal is that developers do not have to think about whether their indexing is correct, whether their collections are well-managed, or whether representations are stored in the right element. The system handles it.

The honest advice no one else is giving about deploying agents

There is a lot of noise right now about the right framework, the right stack, the right implementation pattern. Vasilije cuts through it with two pieces of advice that are more useful than most of what is being written about agent deployment.

The first is about evals. Everyone knows they matter. Almost no one has figured out what good ones look like. The second, get your hands dirty to figure it out.

“One of the major things is that no one knows what good evals are, I think. So that means that I think someone in Anthropic called it vibe evals. So you kind of sit down and you define the vibes until it works. And I think that’s the fortunate reality of the market….Ultimately, Don’t think too much about implementation details. Just get it to work and you’ll find a way because there is, I think, a lot of noise in the market. There is a lot of tools. Half of them are unnecessary. Half of them are somewhat necessary. I would say just get it to work and then you’ll figure it out. And if it scales, you’ll solve those problems.”

Final Thoughts

The through-line here is that the bottleneck in agentic AI is not the model. The models are extraordinary. The bottleneck is the infrastructure underneath: how knowledge is stored, structured, recalled, and shared across agents over time.

Vasilije’s background in cognitive science is not just a talking point. It is the reason Cognee’s architecture looks different from a vector store with a graph layer bolted on. Memory, in his framing, is not a retrieval problem. It is a reasoning problem. And solving it at the infrastructure level is what makes the next generation of agentic systems actually reliable.

From 2,000 pipeline runs in November to one million this month, the traction is already telling a story. The question is how quickly the rest of the industry catches up to the problem Cognee is already solving.

Hector and I go way back. We worked together at Parse, the mobile backend-as-a-service that developers loved and Facebook eventually acquired. We spent a year traveling the world doing meetups, events, and conference talks together, riding the mobile wave when it was the hot new thing. It feels more than a little fitting that we’re back working together now, at a moment when AI is doing to software development what mobile once did: changing the conditions of the game entirely.

Hector spent a decade at Meta, growing with the engineering org from 3,000 to tens of thousands of engineers, and later worked on the React Native team. That background gave him a rare lens on what infrastructure actually has to exist for parallel, high-scale engineering to work. Wallfacer is his bet that that same infrastructure logic applies now. Not to thousands of humans, but to each engineer running multiple agents at once.

Listen in not just for the technical depth, but for Hector’s clear-eyed take on where today’s developer tools are already falling short, what teams that are winning in this new era are doing differently, and what it means for the craft of engineering when AI stops being an assistant and starts being a collaborator.

Full interview here with highlights below.

The tools got 10x faster. The workflow didn’t.

Most teams discovered AI coding tools the same way: one engineer brings it up, the team tries it, and suddenly features that were scoped for next quarter ship in a week. The productivity gain is real. But the existing workflow like the pull requests, the code reviews, the branching structure was never designed for this volume.

“A year ago when Claude Code had just come out…and we actually were shipping features in a week that we had budgeted for next quarter. So that’s when I first saw how these tools could really empower engineering teams to move faster. But quickly, I started seeing the issues. When every team member on your team is now 10x in their output, code reviews quickly became the bottleneck. When I joined Meta, it was like 3,000 engineers. By the time I left, it was tens of thousands of engineers. And I made the connection that we have all this tooling in place to handle having thousands of engineers committing code on the same monorepo, without stepping on each other. And that’s not too similar to the problems we’re seeing now with having each person on your team having multiple agents.”

The teams winning right now have figured out where humans belong

The question isn’t whether to use agents. It’s knowing where to keep humans in the loop and where to get out of the way. The teams that are moving fastest have already started making those decisions deliberately rather than defaulting to the old model.

“The ones that are better prepared for this new way of working are the ones that are just really taking a good look at where are humans still in the loop and where can agents just do that job better. I mentioned code reviews earlier, but it’s not just code reviews. It’s just do we need to wait for a human to deploy, to merge a PR? If you can set up your system where you’re not waiting on humans and you have the confidence that these agents are just doing their job, these are the companies that are just gonna be able to move way faster in this new era.”

This is a process question as much as a tooling question. Infrastructure can remove the bottleneck, but someone has to decide where the bottleneck should be.

For the first time, the question isn’t how long it will take.

There’s something Hector said near the end of our conversation that stuck with me. It wasn’t about infrastructure or architecture. It was about what it feels like to build right now and why this moment is different.

“I’m so excited. I don’t think I’ve had this much fun just working. Because now the question is not how long is it going to take. It’s just, OK, do we need to build it? And it just gets built.”

That’s the shift. Not just faster output, but a fundamentally different relationship between intention and execution. The constraint is no longer time or cycles. It’s clarity about what you actually want to build.

Final Thoughts

The through-line here is that the bottleneck in AI-assisted engineering has already moved. The code generation problem is already solved. It’s everything that surrounds the code: the environments, the previews, the reviews, the deploys, and the decisions about where human judgment still belongs.

Hector has seen this movie before, at Meta, at Parse, through every wave of infrastructure change that made distributed work tractable. Wallfacer is the bet that the next version of that infrastructure needs to exist now before the gap between what agents can produce and what teams can actually ship becomes the defining constraint of the next decade.

For founders and engineering leaders adopting AI, the tools are ready. The question is whether the workflow around them is.

It’s also been a milestone month for exits: Graphite joined Cursor, Sequin was acquired by Notion, and Keycard acquired both Runebook and fellow portfolio company Anchor.dev - read the lineup here.

If you’re at Upfront Summit, Equity Summit, or RSA this month, let’s connect.

Basecamp updates

💰Atomic building the next era beyond Git | $2.5m Seed by Slow Ventures, Irregular Expressions, and Vermilion

💰 Persistent AI Workspaces with Wallfacer | $1.23M Seed by Freestyle Capital and Vermilion

🤝 Graphite (Angel Portco) acquired by Cursor for better code collaboration

🤝 Sequin (Vermilion Portco) acquired by Notion to organize data across the stack

🤝 Keycard (Vermilion Portco) acquires Runebook for easier production-ready MCPs and fellow Vermilion Portco Anchor.dev

Who’s hiring

Find the latest roles across engineering, marketing, and developer relations on the new Vermilion Careers page here

From the trail

The latest industry news, interviews, and forward-thinking considerations from Vermilion, founders and investors in the field.

• The Second Desert in Venture Capital via Charles Hudson @ Precursor Ventures

• Goodbye Git, Hello Atomic: Interview with Lee Faus @ Atomic

• I See Dead VCs Elizabeth “Beezer” Clarkson @ Sapphire Ventures

• Is Venture Really a Team Sport? via Robby @ Modern Technical Fund

• What Early Stage VCs Really Look For via Pepe Agell @ Pear.vc

• The Agents I Use as a Founder via Tori @ Tadata

Upcoming events

Where Vermilion will be in the next month, look forward to seeing you in person 👋

🚗 Upfront Summit | Feb 23-26 Los Angeles

🌵 Equity Summit | Mar 4-6 Scottsdale

🤖 Securing Autonomous Systems - RSA Happy Hour | Mar 25 San Francisco

Today, I’m excited to share one of those moments.

Keycard has acquired Anchor.dev, bringing two Vermilion portfolio companies together to unlock a major step forward for autonomous coding agents.

This isn’t just a team or technology acquisition. It’s about removing one of the biggest blockers to agentic development: trust.

Coding agents are already shipping production software. Developers increasingly rely on tools like Cursor, Claude Code, Codex, and Windsurf to write and modify real systems. But most organizations still depend on static credentials and constant human approvals, which makes true autonomy impossible. So teams either constrain agents to narrow workflows or keep developers permanently in the loop approving routine actions.

Keycard brings a different approach.

Instead of governing access at the protocol layer, Keycard governs at the application layer. Agents get real identities, short-lived credentials scoped to each task, and policies enforced per action. Whether an agent is invoking an API, running a CLI command, or generating its own tooling, the same identity and policy model applies. That lets agents operate autonomously within clear guardrails, while still preserving control for sensitive operations.

With Anchor.dev joining the team, Keycard now extends this model directly into autonomous coding workflows across development and production environments.

Anchor brings deep experience building trusted developer infrastructure from companies like Cloudflare, GitHub, and Heroku, with a sharp focus on making identity primitives like certificates automatic, reliable, and invisible to developers. That expertise now strengthens Keycard’s agent-native identity platform in a meaningful way.

Concretely, Keycard now supports:

• Application-layer policy enforcement at runtime: Policy is evaluated per task and per tool call, where agents actually execute actions, rather than at login, the network boundary, or through protocol-specific proxies.

• Protocol-agnostic support for the full agent toolchain: Agent behavior is governed consistently across MCP workflows, CLI commands, APIs, and agent-generated tools, enabling the long tail of applications and patterns used in modern software development.

• Explicit tool governance with full visibility: Teams define which tools an agent is allowed to use, and every tool invocation, successful or failed, is evaluated, logged, and attributed to a specific agent and task with support for additional metadata.

• Identity-bound, task-scoped credentials: Static secrets are replaced with short-lived credentials issued per task and cryptographically bound to the agent and the application being accessed.

• Portable agent configuration across environments: Agents are configured once and can be used across laptops, sandboxes, and production systems without reconfiguration, using cryptographically attested runtime context.

• Autonomous workflows with built-in guardrails: Routine actions proceed without human involvement, while sensitive operations can still require explicit approval, enabling autonomy without sacrificing security or control.
  

Wesley Beary, CEO of Anchor.dev and now part of the Keycard team, captured it perfectly:

“Anchor focused on removing operational complexity from certificates through automatic issuance, validation, and renewal. At Keycard, that same discipline is being applied to coding agents: real identity, scoped access, policy enforcement per action, and full visibility. That’s what makes autonomous coding practical in production.”

From my perspective as an investor in both companies, this combination makes all the sense in the world.

Agents are becoming first-class actors in modern software systems. But autonomy without identity, governance, and auditability is a non-starter. Keycard is building the foundational layer that lets agents move from local laptops into production safely, without re-architecting workflows or compromising security. Anchor accelerates that mission by bringing deep infrastructure expertise directly into the core platform.

Huge congratulations to Keycard CEO, Ian Livingstone, and the entire Keycard team, and to Wesley, Ben, and the Anchor crew. I’m proud of both teams, grateful to have worked closely with them through this process, and excited to see what they build together next.

Let’s keep building.

It’s been quite the year.

Quick recap

• We made 20 investments in pre-seed and seed stage early stage technical founders for a total of 35 portfolio companies in our first fund.

• 5 companies were acquired as we see the M&A market is heating up again.

• We hosted 2 Tiki Socials in NYC and SF with a great crowd and tropical vibes. Another one coming soon. We co-hosted several events with companies in the portfolio and other VC funds.

• We attended events all over the world and spent time with some of our favorite founders, VCs, and LPs in person.

• My Twilio colleague Meghan Murphy joined the team to help scale our founder support across technical GTM.

• Made so many new friends and caught up with old friends along the way which has been such a life highlight.

  

  

What is still true today

Vermilion backs technical founders building technical products at the earliest stage.

When we launched, we said Vermilion would write meaningful checks to technical founders building technical products, developers, infrastructure builders, data engineers, security innovators, and everyone in between and we’d be operators alongside them, not spectators.

This year, that commitment has taken shape in:

• Portfolio companies shipping meaningful early products that solve real pain points for developers and high-growth teams.

• Founders hiring critical early talent and gaining traction with customers, closing their first 10 deals and starting to scale GTM.

• Teams rallying around execution and focus, even in an era of shifting market dynamics.

• Portfolio companies raising follow-on rounds from some of our favorite funds to work alongside.

It turns out the next wave of infrastructure isn’t just about models, it’s about how humans and machines actually build together. That’s exactly the layer the founders we work with are thinking about and where we love investing.

Who we invest in

One of the enduring themes from our launch announcement was this idea:

The great thing about being an early stage VC is being able to watch so many early stage founders go from idea to product. It’s always been the favorite part of every job I’ve had, shipping quality product, so getting to see that over and over again is a dream.

We are backing founders that are obsessed with how software will be built in 5-10 years and want to create a product or protocol that will be instrumental in that shift.

What I’ve learned (and what I’m still learning) about backing early-stage founders

Our first year wasn’t just about writing checks, it was about the craft of early-stage building. A few things that seem to be true amongst all of our portfolio companies:

• The early stage landscape is changing.

You can feel it in the products shipping, the teams rallying, and the speed of feedback loops. There’s less patience for vague vision and more demand for real progress. Fundraising is harder. Buyers are more thoughtful. And teams that find traction early are the ones pulling ahead.

It’s 2026, not 2021.

Capital is no longer abundant, timelines are compressed, and expectations are higher across the board. Raising a round is the hardest it has ever been, especially at the earliest stages. Founders aren’t just competing with other startups, they’re competing with a market that now requires proof instead of promise.

And once you do raise that early round, the clock starts immediately.

There’s less room to experiment in isolation. Less tolerance for wandering roadmaps. Less appetite for “we’re still figuring it out.” What used to be a long exploration phase has turned into a short window to demonstrate momentum. You still need to learn quickly, but you need to ship while you learn.

That changes how companies get built.

Teams are staying smaller for longer. Founders are closer to customers earlier. Products are getting into users’ hands faster. Iteration cycles are tighter. Decisions feel heavier because every month matters.

I’ve also noticed that resilience has become a real differentiator. The founders who are winning right now aren’t waiting for perfect conditions. They’re making progress in imperfect markets. They’re scrappy. They’re focused. They’re prioritizing signal over noise.

This environment rewards clarity, speed, and conviction.

It rewards founders who can move from idea to product to customer feedback without overthinking every step. It rewards teams who understand their ICP early and build directly toward it. And it rewards builders who treat constraints not as blockers, but as forcing functions for better execution.

The bar is higher. The margin for error is smaller.

But for founders who can ship, listen, adapt, and stay close to their users, this is also one of the most exciting times to build.

And with that…

• Execution and iteration matters earlier than ever.

This year reinforced what I underscored at launch: the best products win because they solve real problems and teams that obsess over shipping product instead of ideas get the most traction possible, earlier.

There’s no time to sit in obsessive internal feedback cycles.

You have to ship what you’re building from the first week and onboard design partners and pilot customers as early as possible. Throw up a landing page with a beta request email field the minute you have something to get in the hands of developers.

• Grit and the ability to keep pushing through hard things is a competitive advantage.

Not defined by background, pedigree or resume, but by how founders respond when growth feels slow, bugs pile up, or the market shifts. You can’t freeze. You have to push through it and be honest with yourself and others about what is working and what is not. It can kill your company if you freeze and hide (which I tend to do so I totally get it).

• Founder-market fit really matters when building for technical audiences.

When founders are building for developers, security teams, data engineers, or infrastructure teams, they need to be building for themselves. Not hypothetically. Not through secondhand research. Through lived experience.

It’s just easier that way.

The best products I’ve seen come from founders who are solving problems they’ve personally wrestled with. They know where workflows break. They feel the friction in existing tools. They understand the tradeoffs because they’ve lived inside the systems they’re trying to improve.

You can’t really be a tourist in the current technical market.

• Early GTM starts on Day 0, whether you plan for it or not.

The days of quietly building for years and then “turning on” go-to-market are over. Even at pre-seed, founders need to be ready to answer some very real questions: What does monetization look like? Who are your first 10 pilot users? How do you plan to reach them? What signal tells you this is working?

Those answers don’t need to be perfect, but they do need to exist.

Too often I hear, “We’ll figure GTM out later,” or “We’ll just hire a head of marketing.” That’s almost never the right move early on.

Hiring a head of sales or marketing too early can be a death sentence. It creates the illusion of progress while quietly pulling engineering in the wrong direction. Suddenly you’re optimizing for pipelines and campaigns instead of learning. You’re shipping features for hypothetical customers instead of listening to real ones. You’re building process before you’ve earned it.

At the beginning, GTM is not a job title. It’s founder behavior.

It’s founders talking directly to users. Writing the first cold emails. Running the demos. Sitting in customer Slack channels. Watching how people actually use the product. Hearing objections in real time. Feeling friction in onboarding. Internalizing what lands and what doesn’t.

This is where product-market fit starts to form.

The strongest early teams treat GTM as part of product development. Customer conversations inform roadmap decisions. Early pilots shape architecture. Pricing experiments influence scope. Everything is connected. There’s no clean separation between building and selling yet and there shouldn’t be.

I’ve also learned that the first 10 users matter as much at the first 1000. Those early design partners help define your ICP, your messaging, your product shape, and often your confidence as a founder. They tell you what’s confusing, what’s valuable, and what’s missing. They help you find signal in the noise.

Later, when there’s repeatability, when the motion is clear, when you can explain why customers buy and why they churn, that’s when you start layering in GTM hires.

Until then, founders need to stay close to the work.

• Technical differentiation is table stakes. Narrative is leverage.

Storytelling is core to winning. In a world where everyone is tackling similar sets of problems and trying to figure out how developer tools evolve, your story is everything.

There is a lot of noise. Be the music.

• Small teams with extreme clarity outperform larger teams with vague ownership.

Hiring is not a strategy.

Early on, every additional person adds coordination cost. Meetings multiply. Context gets diluted. Decisions slow down. When roles are fuzzy, work falls through the cracks or gets duplicated. But when a small team has crystal-clear ownership, things move fast. Everyone knows what they’re responsible for. Everyone feels accountable. There’s no hiding behind org charts or process.

You can get an incredible amount done when you aren’t worried about hierarchy.

The best early teams I’ve seen operate more like tightly aligned problem-solvers than departments. Engineers talk directly to users. Founders ship alongside their teams. Product decisions happen in real time. Feedback loops are short. There’s very little ceremony.

Yes, you will need to hire as you scale. Eventually you’ll need structure, managers, and more specialized roles. But there’s a window early on where small, aligned teams are truly magical. It’s when velocity is highest, creativity is strongest, and everyone is deeply connected to the product and the customer.

That phase doesn’t last forever, and that’s okay. But I’ve learned not to rush past it.

Protect that moment.

• Capital helps, but operator support moves the needle.

I’ve noticed a pattern with a lot of founders after they close their first round. There’s a brief moment of relief and excitement, and then it hits: okay… so now what?

I think that feeling is completely normal.

Raising money feels like a milestone, but it’s really just the starting line. Suddenly you’re responsible for turning a deck into a product, a vision into a roadmap, and a small idea into something real. There’s no playbook handed to you when the wire lands.

The next step is usually a text or an email to investors and advisors asking for help. How do I get a beta product out the door? Who should I talk to first? How do I find design partners? Can someone introduce me to potential customers? What should I prioritize in the next 30-60-90 days?

This is where the difference between capital and partnership becomes obvious.

Checks don’t ship product. People do.

I’ve also learned that early-stage building is deeply emotional. Founders are carrying a lot at once: product, hiring, fundraising, customer discovery, and often self-doubt. Having someone in their corner who’s been there before, who can normalize the chaos and help create clarity, makes a real difference.

Capital gives founders runway. Operator support gives them momentum.

And momentum is what turns early ideas into real companies.

What I’ve learned (and what I’m still learning) about building an early stage venture fund

I’ve learned more in the last few years than I had in the 10 years prior. Part of why I didn’t want to be an operator anymore is because I felt like I was doing the same job over and over at different companies. I get bored easily so having tons of new companies to talk to and learn about while also learning new things every day about building a fund has been so much fun. A few things that I’ve had to learn so far:

• Fundraising is a long strategic game, not a direct process.

You’ll meet most of your Fund 2 LPs while raising Fund 1.

In many cases, it’s a years-long process. This is one of the biggest surprises for first-time managers. Fundraising for a venture fund looks nothing like fundraising for a startup.

As a founder, you can meet a VC and have a term sheet the next day. There’s urgency. There’s momentum. There’s usually a clear decision-maker on the other side of the table.

LP fundraising usually doesn’t work that way.

The LP/GP relationship is fundamentally different. It’s built on long-term trust, not short-term conviction. LPs aren’t underwriting a single product or market. They’re underwriting you across a decade or more. Your judgment. Your consistency. Your ability to source, support, and return capital through multiple market cycles.

It just takes time. And a great LP is worth the wait.

• Your story matters as much as your strategy.

LPs aren’t just underwriting your fund model. They’re underwriting you. Your background, your operating experience, your point of view on the market, and why you’re uniquely positioned to see and support great founders early.

Your story is how people remember you. It’s what sticks after a first meeting. It’s what LPs repeat internally when you’re not in the room.

Strategy explains what you’re doing. Story explains why you’re the person to do it.

The strongest emerging managers I’ve seen are clear on both. They can articulate their thesis, but they can also explain the path that brought them here. What they’ve built. What they’ve learned the hard way. What they care about. That authenticity builds trust faster than any slide.

• Early believers are everything.

Your first LPs matter more than you realize.

They become your references. Your connectors. Your sounding board when things feel slow. They help validate you to the next wave of LPs. They’re often the ones who lean in when momentum dips.

Early believers aren’t just capital. They’re signal. They’re emotional support. They’re the people who took a chance on you before it was obvious.

Those relationships tend to last far beyond Fund I.

• Traction beats credentials.

Titles don’t close funds. Progress does.

LPs care far more about what you’re actually doing than where you worked before. Portfolio construction. Deal flow quality. Founder feedback. How quickly your companies are moving. How you support them.

Real traction speaks louder than resumes.

This applies to both you and your founders. Seeing companies ship, hire, sign pilots, and find early product-market fit is infinitely more compelling than any pedigree.

• You need to show your work.

LPs want to see how you think: investment memos, portfolio construction logic, founder support examples, market perspective. Transparency builds confidence.

• Fundraising while deploying is brutal, but powerful.

Doing both at once is exhausting. It’s also your strongest proof point. Nothing validates faster than great founders choosing you early.

So what’s next?

I still believe the venture landscape is changing and early technical founders deserve more than just capital. They deserve partners who know what it feels like to build in the trenches, who can help guide go-to-market from day zero, and who bring empathy for both the logic and the chaos of early product work.

• Doubling down on founder support, not just checks, with hiring a team that can help scale our GTM support.

• Deepening our network of operators and experts who can add tangible value to the Vermilion portfolio of founders.

• Backing the next category-defining infrastructure builders, early.

Because I still believe what I said on Day 0: the next era of innovation won’t be built in boardrooms or pitch decks. It’ll be hacked together at 2 a.m. by teams obsessed with solving problems no one else has solved yet.

Thanks to the community around me

To the founders who trusted me early, to the LPs who believed in my vision, and to the peers who’ve helped shape me, thank you. You’ve let me ask really basic questions as I’ve learned to build a fund and I deeply appreciate the support. I’m trying to pay it forward.

This first year has been about doing the work together. And that’s the best foundation any venture firm, or founder, could ask for.

Here’s to Year Two.

Lee has more than two decades experience building and scaling developer infrastructure at GitHub, GitLab, and Red Hat. He sees the next wave of AI adoption forcing a reset at the foundation, not another layer of tools added at the edges. Atomic doesn’t bolt AI onto an existing workflow, it rethinks version control for a world where humans and machines both produce code at scale.

Listen in not just for the technical depth, but Lee’s articulation of where today’s systems are already failing, why enterprises are paying attention, and what it means for how engineers work over the next decade.

Full interview here with highlights below.

AI turned developer productivity into a weapon, but the process never changed

Most teams stack copilots and models onto workflows designed for slow, human-only contribution. That mismatch hits hardest at validation, where Git starts to strain under scale it was never designed to handle.

“Sure, developer productivity is awesome. It’s only one measurement when we talk about building software. So that whole workflow has to change.

If I’ve got a thousand engineers and each one is solving three problems a day, that’s three thousand branches. Most pull requests are open for three days, so now you’ve got nine thousand pull requests sitting there waiting to be approved.

Now give every engineer a hundred agents. You end up stacking all of those changes on top of each other, and Git doesn’t have a good way to reason about that.

That’s where Git breaks. Not because it’s bad software, but because it was never designed for this world.”

This isn’t a future problem. Large mono repos are already breaking under human scale. AI just makes the failure mode impossible to ignore.

We don’t need AI for Git, we need a different way of storing change

Git’s core abstraction assumes changes arrive slowly and are reviewed asynchronously by humans. That model breaks once humans and machines are producing code continuously and in parallel. Fixing that isn’t about adding AI to Git, it requires rethinking how change itself is represented.

“If we think about how Git is architected - I’m going to simplify this a lot - Git is really about comparing snapshots. You’re looking at one picture and another picture and trying to see the differences. But what happens when you’ve got a thousand of those pictures layered on top of each other? It becomes really hard to understand what changed and why.

Now think about LEGO blocks. If I have a red block and a green block, they know how they relate to each other. When I insert a new block in between them, it knows exactly where it belongs. This is Atomic.

Because of that, we always have a clean state of the repository, and changes commute automatically.”

That decision enables large-scale agentic development without pushing humans or machines into serialized workflows. It also embeds provenance directly into the system instead of tacking it on later.

Enterprises aren’t worried about future scale, it’s the limits that already exist

Enterprise teams aren’t speculating about what AI might break someday. They’re already hitting limits around repository size, validation speed, and accountability. AI doesn’t create those problems but it makes them impossible to ignore.

“Could you actually go back to be able to look at that piece of code and say, was that written by a human? Was it written by AI? If it was written by AI, what was the prompt that generated that code?

If it ended up going through a security process, who was the reviewer? Did AI review the code or did a human review the code? All of those things are now becoming critical to people adopting AI.”

Repository size, I/O bottlenecks, and centralized CI pipelines are already creating friction. AI accelerates the volume of change to the point where those bottlenecks become operational risks. AI doesn’t create these requirements. It makes them non-negotiable.

In an AI world, correctness and trust can’t be proprietary

Atomic’s commitment to open source is about credibility, not ideology. Open source gives developers and enterprises the ability to inspect and verify the systems they’re betting on. In the next generation of developer infrastructure, trust isn’t a feature. It’s the baseline.

“Do you open source it? If you open source it, what license do you want to use?

Because we’re extending Pijul, we’re sticking to their license. We can always append the copyright to say we’re using their code plus our code, so there’s an extension of the copyright.

But the dataset needs to hold the rigor for the math to math. If we were doing this closed source, we couldn’t get any sort of independent validation that it’s actually doing what we want it to do. So going down the open source path was important to me.

I wanted to make sure we get open source people to come in, play with it, help us find where the math maybe doesn’t math, but at the same time come in and say, what other innovations do you want to put in?”

Final Thoughts

The through-line here is that AI doesn’t just add pressure to existing systems. It exposes where those systems were already fragile.

Git reshaped software development by enabling distributed human collaboration. Atomic bets that the next shift comes from enabling distributed human and machine collaboration, with correctness, provenance, and trust built in from the start.

For founders, operators, and enterprise teams adopting AI, this shift isn’t optional. The only question left is whether the foundation evolves fast enough to keep up with how teams already write software.

We’re entering the AI-native era. Not the model arms race, but the connective tissue that lets humans and AI build together. And as Tech Week showed, the go-to-market motion is becoming human again: smaller rooms, real context, and trust as distribution.

It’s the best time to be early again.

Basecamp updates

💰 Keycard building trust in an autonomous agent world | $38m Series A by A16z, Boldstart, and Acrew

📈 Osmosis solving the static model problem with reinforcement learning | $6.3m Seed by CRV and Audacious

Who’s hiring

Find the latest roles across engineering, marketing, and developer relations on the new Vermilion Careers page here

From the trail

The latest industry news, interviews, and forward-thinking considerations from Vermilion, founders and investors in the field.

• New Stack, Same Game: Why the AI-native era feels different yet familiar

• How Screendoor Became a Key Signal for Emerging VC Talent Allie Garfinkle at Fortune

• Gem’s Emerging Manager Selection with Jay Ripley

• The Trillion Dollar AI Software Development Stack via Andressen Horowitz

• Introducing the Venture Codex from Aaron Harris

• Why You Need an MCP Tool Layer from Tadata

• A look at Startup Growth from Zero to $25m ARR via ChartMogul

Upcoming events

Where Vermilion will be in the next month, look forward to seeing you in person 👋

👾 Unofficial AI Code Summit Happy Hour with Keycard

• Infrastructure is maturing
• Developer experience is improving
• The market is expanding for the tools that build on top

No one lost their moat. If anything, the space for new companies opened wider. This is the best time to be early again. While this current tech stack is most definitely new with models, orchestration layers, and agent frameworks, the underlying patterns to this landscape are recognizable.

We’ve seen this before:

• When the web went mobile native
• When SaaS went cloud native
• When Git gave way to CI/CD and DevOps ecosystems

Now we are entering the beginning of another shift: AI-native software development infrastructure. Not the model arms race, but the connective tissue that lets humans and AI build software together in real workflows.

Infra Matures, Behavior Shifts

Every technology wave starts with infrastructure. The real shift happens when behavior changes and we’re seeing that today.

• Platforms like Copilot and Cursor moved AI into the flow of writing and reviewing code, making model assistance a default part of creation rather than a separate step.

• MCP and agent frameworks turned single prompts into ongoing context, enabling models to maintain memory, state, and multi-step workflows across tools and environments.

• Development environments are starting to adapt to developer intent in real time Tools like Warp, Zed, and Aider treat code, commands, and context as part of the same reasoning loop instead of isolated actions.

What cloud was to SaaS, AI infrastructure is now to every function, including how we build, ship, document, and communicate software.

This is the layer where the next decade of developer platforms will be defined. It includes tools for developers building with AI agents, security layers for AI-generated code, and systems for multi-model data and workflow orchestration.

The Market Is Not Consolidating, It Is Fragmenting

There is a common assumption that someone like an OpenAI will simply build everything. The same assumptions appeared about AWS, GitHub, and Atlassian in their key eras. But every time the platform layer stabilizes, specialization accelerates.

History is consistent:

When core workflows stabilize, entire ecosystems emerge around them.

There will be consolidation later. But right now the opportunity surface is expanding faster than any single platform can absorb. This is the phase where new category-defining companies are created.

The Return of Human-Native

As the tooling goes AI-native and dev tools rise into a golden age (again), go-to-market is becoming human-native again.

Tech Week made this visible. Dozens of small, founder-led gatherings. Conversations in living rooms and back rooms and rooftops. Not because people are nostalgic for community, but because trust and context matter more when everything else accelerates.

It feels like early SaaS again:

• Founders are the marketing team

• Brand is shaped by narrative instead of volume

• Signal travels through relationships, not feeds

AI increases the speed and scale of production. As a result, audiences filter by who they trust, not what they see.

Human-Native, AI-Driven GTM

The last decade focused on optimizing funnels. The next decade focuses on optimizing context.

We are emerging from the hype cycle and entering another craft cycle.

It’s the Best Time to Be Early Again

This moment rewards developers who think in workflows, agents, and composable systems. It rewards founders who understand both model interaction and human interaction. It rewards builders who see tools not just as products but as practices that reshape how software is made.

The stack is new. The game is familiar. The opportunity is open. It is AI-native and human-native, and it is happening now.

Software isn’t static anymore. It’s dynamic, generative, and increasingly autonomous. AI agents can now make decisions, execute actions, and operate across systems on behalf of people and organizations. That evolution introduces enormous potential and an equally enormous surface area for risk.

Today, we’re excited to share that Keycard has launched publicly and announced $38 million in funding from Andreessen Horowitz, boldstart ventures, and Acrew Capital. Vermilion is proud to be an early investor and partner to the team as they build the identity and access layer for this next phase of computing.

The Problem: Trust in an Autonomous World

For decades, identity and access management has revolved around static users and predictable workflows. Humans log in, perform a set of actions, and log out. Permissions are granted through roles, groups, and API keys.

In the age of AI agents, that model collapses. Agents spawn by the thousands, move across systems, and act autonomously. They need different permissions depending on what they’re working on, who they’re acting for, and what resources they’re touching. The security model for this world can’t rely on static secrets or brittle access lists. It requires something dynamic, real-time, and verifiable.

Keycard was built for that world.

The Solution: Real-Time Identity for AI Agents

Keycard provides a dynamic identity and access platform for AI agents that integrates with existing enterprise identity systems. It cryptographically proves who an agent is, who they act for, and whether they’re authorized to take a specific action. Permissions are task-scoped and ephemeral, adapting as context changes.

Instead of long-lived API keys, Keycard issues identity-bound, short-lived tokens tied to the agent’s verified workload identity, compute provider, or device. This allows real-time policy enforcement and instant revocation without code changes.

In short, Keycard brings trust, transparency, and control to a world where software acts on its own.

The Team

Ian Livingstone, Matthew Creager, and Jared Hanson have spent their careers building the developer and security infrastructure that powers modern software. Ian and Matthew helped scale Snyk’s platform engineering and developer experience teams as revenue grew from $30 million to $300 million. Jared was the Chief Architect at Auth0 (acquired by Okta) and created Passport.js, the most widely used authentication framework for Node.js.

They’ve each seen how developers struggle to connect systems securely at scale. And they know what it takes to build products developers trust.

Why We Invested

At Vermilion, we back technical products built by technical founders. We look for people building the foundational layers that enable new paradigms in how software is created, deployed, and trusted.

The AI agent ecosystem is evolving fast. But without secure identity and access, it can’t safely move into production. Keycard’s approach bridges that gap. They’re not just securing the edges of agentic systems. They’re redefining how identity itself works in this new world.

We believe that trusted agents will underpin the next wave of computing. Keycard is building the infrastructure to make that future possible.

If you’re building in the agentic space or rethinking access, identity, or policy in distributed AI systems, take a look at keycard.ai