AI is the Worst Thing to Happen to Early Stage Investors

Vibe-coded products, AI-generated decks, and 20,000 applications with less signal than ever

TLDR: AI is making every B-player look like an A-player. But bigger problem isn’t the noise, it’s that the best founders have stopped showing up to the process entirely. And most of venture has no answer for that.

---

Three weeks.

That’s how long it took the founder sitting across from me to “build” their company.

The deck was flawless. The product demo worked. He had a landing page that was ripping, a waitlist that was growing, a Loom walkthrough that looked like it came from a Series B company.

I asked him what surprised him most from talking to his first ten customers, and he hadn’t even had his first ten conversations.

The product was vibe-coded. The deck was AI-generated. The waitlist was paid acquisition. The demo worked, but the codebase underneath was held together by prompts and prayers. Everything looked solid.

Three years ago, our minds would have been blown.

Today, not the least bit impressed.

This is the part nobody in VC wants to say out loud: the signal is broken.

Not just the deck. The product, the traction, the demo, the entire top-of-funnel presentation layer. AI has made it trivially easy to look like a company.

The hard part is still being one.

---

---

What used to work.

For a long time, a well-crafted pitch deck was a proxy signal. A clean narrative, coherent unit economics, a crisp competition slide. It meant the founder had done the work. Talked to customers. Thought through defensibility. The deck was a byproduct of real preparation.

Not anymore.

There are now plenty of purpose-built AI pitch deck generators trained on thousands of winning decks, optimized for the exact pattern-matching VCs do in the first 30 seconds. They know the TAM slide better than the founder. They know the language of founder-market fit. They organize a narrative better than most great storytellers.

But the deck is only the start.

The entire presentation layer has been compressed.

Founders are spinning up products in a weekend with Lovable, Cursor, and Replit that look like they took six months. Landing pages that convert. Demos that feel polished. Loom walkthroughs with real UI, real data, real-looking workflows. But, underneath?

Often vibecoded slop. Codebases that are nearly impossible to inspect, built on prompt chains that break under load, with no architecture to scale. Products that look like they work and companies that look further along than they are.

Worse founders don’t just look more credible now. They think company-building is easier than it is, and that is the real disservice of AI.

Lower barrier to looking real means lower barrier to trying. But, the barrier to succeeding is still as high as ever.

The deck has gone from diagnostic to theater, and so has the demo.

What does this actually look like from the investor’s seat?

The first meeting is no longer informative. Due diligence has to go deeper faster. The metrics lie earlier.

You can’t just kick the tires on the product anymore. You have to get under the hood, and the hood is harder to open. Vibe-coded products are a black box of prompt chains that the founder may not fully understand themselves. A founder can spin up paid acquisition, hit a landing page conversion rate that looks organic, and show you a growth curve that implies product-market fit.

None of it requires a single real customer conversation.

The traditional early traction signals, the ones investors have used for years to separate the serious from the speculative, are now reproducible on demand.

And it means the false positive rate has gone through the roof. More companies look investable. Fewer of them are. Every check you write into a company that looked real but wasn’t is capital that didn’t go to a company that was.

The VC tool-building paradox.

Yes, the signal is broken, but evaluation is breaking too.

VCs aren't sitting still.

They see the noise increasing and they're building tools to cut through it. AI-powered sourcing platforms. Automated screening. Predictive models trained on historical founder data. The pitch from every fund admin and data company right now is the same: we can help you find signal in the noise, but how do you find an outlier founder with a systematic process?

You can’t.

Every attempt to scale founder evaluation with AI has the same flaw: it optimizes for the signal it was trained on, which is now the signal anyone can fake. AI screening models. Forty-question applications. Agentic interviews. Automated video assessments. The most systematic funds are running the most sophisticated filters. And, without realizing it, they may be simply selecting the founders who are best at navigating filters. That is not always the same person as the best founder.

On the other side of this problem: the founders who don’t need to navigate anything.

On March 13, Travis Kalanick emerged from eight years of stealth. His company, Atoms, had thousands of employees who weren’t allowed to put the company name on their LinkedIn profiles. For eight years. No top-of-funnel. No pitch process. He raised over a billion dollars from sovereign wealth funds and built a $15B company entirely in the dark. When he finally went public, it was through a manifesto, not a deck.

And Kalanick isn't alone. Last week, it was reported that Jeff Bezos is raising a $100 billion fund to buy manufacturing companies and rebuild them with AI that simulates the physical world. Yann LeCun just raised $1.03 billion for AMI Labs. Fei-Fei Li raised $1 billion for World Labs. General Intuition raised $133.7 million at seed. The entire embodied AI space — World Models, autonomous driving, robotics — is being funded in billion-dollar chunks by researchers and operators who never touched a pitch deck process. They're in labs and warehouses, training AI on footage of the real world, video games, and driving data, and the checks are getting written in rooms you're not in.

They’re not avoiding investors because they’re struggling. They’re avoiding investors because they can.

Those are the billion-dollar versions. But the pattern runs all the way down.

The founders doing the most interesting work are increasingly seed-strapping. They’re using AI to compress the time between idea and traction, staying lean, staying invisible, and staying in control of their cap table.

Carta data shows pre-seed deal volume down 28% year over year while the best companies are reaching meaningful milestones with less capital than ever.

They’re not out of options. They’re running a new process.

---

What real signal looks like now.

If the deck is broken as a filter, what replaces it?

The answer is embarrassingly analog.

It’s the stuff AI can’t generate. It’s the obsession to build. It’s the maniac mentality.

Take Dakotah Rice for example. He showed up to Antler’s NYC residency having already failed publicly. His previous company, Poolit, shut down in 2023. He’s been open about it. He told TechCrunch he should have shut it down a year earlier, that his ego got in the way. That kind of honesty is rare in a pitch. It’s rarer in a room full of people trying to impress you.

But that’s not why we backed him. We didn’t even back him because of his idea. In fact, at the time, he was working on something he called Tatch, an AI-native data room for fund admin and private equity firms. I had real questions about how he would compete with the likes of Hebbia, but ultimately it didn’t matter. Dakotah was inevitable. His sense of urgency, intellectual honesty, and undeniable slope told me all I needed to know.

Later, he and his co-founder Tushar Nair made a hard pivot to insurance, and got into YC. It was a perfect pivot for them. His family owned a brokerage. His entire life he had been witness to others’ struggles in a process that hadn’t been redesigned in decades. They initially planned to build AI tools for existing brokerages. Then they realized they should just build the brokerage themselves. The obsession to build came first. The company came second. And the pivot closer to home came third.

We were their first check, Tatch is now called Harper, and Dakotah and Tushar just raised $47 million in a combined Seed and Series A led by Emergence Capital. They have over 5,000 customers. What used to take a traditional broker five to seven days, Harper does in one to two. None of that was visible when we backed him. What was visible was a founder who’d been through failure, came back with a chip on his shoulder, a work ethic like no other, and a co-founder he’d already been in the trenches with.

That’s what backing maniacs at inception sounds like.

That’s what AI cannot replace.

Or, take Arthur Leopold as another example.

He was the first employee, first investor, and President of Cameo. He helped raise $100 million at a billion-dollar valuation. Before that, we worked together at LinkedIn where he was known as “the five quarter guy” for his work ethic.

A few years ago, we sat at the counter at Mel's in Chelsea and he told me he was thinking about leaving to start something. He had two ideas. No company yet. No deck. I told him right there that we would back him to figure it out. He still needed to incorporate and decide on the direction that would later become Agentio. We came in alongside Craft and AlleyCorp in the first round, later backed by Benchmark and Forerunner in the A and B respectively. Today, Agentio is valued at $340 million, with 35 people, and growing to 150 by end of year. None of that was on a slide at Mel's. It was a person, a conviction, and a track record of building things that mattered.

No amount of AI polish produces that, and investing at inception, you need to underwrite at the obsession level, rather than waiting for the company or the pivot.

Of course, every founder will tell you they’re obsessed, but finding out if they truly are is a very different exercise than simply asking. Every exceptional founder has a strongly-held view they’ll defend in a room of skeptics. Not because they enjoy the fight, but because they think they’ve solved something the market hasn’t.

Marc Andreessen has famously described how he finds the real ones through an approach he calls an interrogation:

“You ask increasingly detailed questions and people have trouble making things up and things just fuzz into obvious BS... fake founders basically have the same problem. They’re able to relay a conceptual theory of what they’re doing — but as they get into the details, it just fuzzes out. Whereas the true people you want to back can do it.” — Marc Andreessen

The AI-generated narrative never fuzzes on the surface, it holds itself indefinitely, coherently, with no typos. That’s exactly the problem. The Andreessen test still works though, but only in a room, with follow-up questions, in real time. Not on a slide.

Early stage VCs need to stop spending the first thirty minutes on what the founder is building, and start spending the time on why the founder can’t stop.

---

What this means for inception.

As AI compresses the time between idea and traction, founders are reaching inflection points faster, but the market is pricing it as durability, when in reality it may not be. Carta data shows the 95th percentile seed valuation hit $80M in 2025, nearly 3x higher than 2019. Buying into a fully-formed narrative at a fully-formed price is a math problem that gets harder every year.

The funds that win aren’t going to win with AI outreach, AI screening, and AI memos. They’re going to win by getting to founders before the screening exists. Before the deck is written. Before the narrative is polished. Before the AI has cleaned up the rough edges.

At inception, the deck hasn’t been written yet. There’s no computer-generated problem statement, no traction to benchmark, no competitive matrix. There’s just the founder, the idea, their obsession, and the energy in the room.

For inception investors, their valuation will still reflect the uncertainty, which means the upside still makes the math work. For them, the signal isn’t broken when it’s done right. It’s tactile, it’s visceral, and it’s the least noisy stage to underwrite.

The edge in this market isn’t a better process. It’s being there before the process starts.

Somewhere right now, a researcher is training an agent inside a simulated world built from gaming footage. No deck. No demo day. No LinkedIn post. Just obsession and compute.

Those are the founders I want to find.

See you Monday.

Apply to Antler | Work with me on Intro