The Soul of the Pitch Deck

I can spot a Claude deck in about four seconds.

Navy and gold. Serif headlines. A tidy grid of rounded cards. A TAM slide with a number so big it needs a "T." Every bullet is perfect, every sentence is balanced, and the whole thing says almost nothing.

It's pretty. It's polished. And it tells me nothing about you.

That's the problem. At pre-seed, the pitch deck was never about the slides. It was about the soul of the company. And right now founders are handing that soul to a model.

The Deck Is Mutating

Something strange is happening to the pitch deck. It's splitting in two.

On one side is the deck. On the other is the investor memo: four, six, eight pages of dense narrative text.

The memo isn't new. Bezos banned PowerPoint at Amazon years ago in favor of "narratively structured six-page memos," and said a great one "probably should take a week or more" to write. Parker Conrad has said outright that a memo "has become an acceptable alternative to a standard pitch deck with many investors." Rippling raised on one, with a stripped-down "just the facts" metrics deck alongside it.

That was the good version: a founder who'd thought so hard about the business that slides couldn't hold it.

That's not what's driving the new memo wave. Here's what is:

Founders figured out that VCs are pasting everything into an LLM.

So they're writing for the robot. Long text and lots of keywords, built to be parsed. The deck has become something you feed to an LLM.

Here's my reality: if you send me a four-page memo of text, I'm not reading it. You get dropped straight into the bucket. Not because I'm lazy. Because I have hundreds more in the queue, and you've just told me you'd rather impress a parser than a person.

Agent vs. Agent

Let's play this forward, because the infrastructure is already here.

Draper Associates runs a "Draper Brain," an MCP server that pipes their Airtable, Drive and inbox into Claude. Their deal engine scores about 20,000 companies on eight dimensions against the firm's past decisions and surfaces about 150. Affinity just shipped AI agents and MCP skills for Claude and ChatGPT, and has an agent on its roadmap that watches inbound email for pitch decks and turns them into structured CRM data. DocSend now has an MCP server too.

Draper's own principal put it plainly: AI has multiplied company formation, and "the top of the funnel is larger than it's ever been."

So where does this end?

Maybe every fund runs its own agent. Every founder runs one too and applies to every fund on earth. The first two rounds of diligence become machines checking boxes against other machines: thesis fit, stage, check size, sector.

Maybe that's the world we're headed for. I'm not sure it works.

Because if both sides are optimizing for the filter, everything that gets through the filter looks the same. And when everything looks the same, the only thing left to judge on is the thing the AI can't fake.

The Homogenization Problem Is Real (and Measured)

This isn't just my gut. There's research on it.

Science Advances (2024): Stories written with AI ideas were rated more creative individually, and were more similar to each other. The authors called it a social dilemma: each writer does better, the group gets less variety.
Wharton, Nature Human Behaviour (2025): ChatGPT-assisted ideas were less diverse in 37 of 45 comparisons. In one task, only 6% of the AI-aided ideas were unique. Unaided humans: 100%.
NYU (ICLR 2024): Co-writing with a model significantly reduced diversity across authors' essays, and the model's text was what caused it.

Now picture that across a pitch competition. One operator described a session where "all the decks looked eerily similar." When an investor pushed on a >$5 trillion TAM, the founders couldn't defend it. Then they admitted Claude had built the deck.

That's the whole problem in one moment.

They didn't lose because the deck was bad. They lost because there was no one behind it.

What the Pre-Seed Deck Is For

Let's be honest about what we're underwriting at pre-seed.

It isn't the business model. It isn't the five-year projection. There's a high probability that where you start isn't where you finish, and that's fine. Most great companies pivoted, narrowed or rebuilt the product at least once.

What we're underwriting is how you think.

How you found the problem.
Why you noticed it when no one else did.
What you believe that most people don't.
How you'll make decisions when the plan breaks, because it will.

Sequoia said it about Airbnb's original deck years ago: "it wasn't really the slides we liked; it was their ideas, the clarity of their thinking."

Paul Graham put it more bluntly: writing is thinking. AI is about to split the world into "thinks and think-nots."

A deck the AI wrote for you is a deck from a think-not. It might sound smart. It doesn't think. And when we can't see how the team thinks, we can't do diligence. We're just reading a well-formatted guess.

That's why AI slop doesn't just waste your shot. It wastes everyone's time. Yours, ours, and every other founder's in the queue behind you.

"But the Tools Are Good"

They are. I'm not anti-AI. That would be a strange position for someone in venture in 2026.

Gamma hit $100M ARR and a $2.1B valuation because it removes real pain. Nobody should spend 12 hours fighting slide alignment. Use Gamma. Use Claude. Use whatever gets the deck clean and readable.

But. There's a difference between using AI to present your thinking and using it to replace it.

Polish? Great.
Formatting? Great.
Pressure-testing your argument? Great.
Generating your business model, your market, your wedge, your "why now"? That's outsourcing the only part we care about.

It's obvious when we see it. It feels like founders are skipping the early stages entirely: the talking to customers, the arguing with your co-founder, the whiteboard at 1 a.m. And you end up with "AI told me this was the business model."

If there's no critical thinking behind it, there's no company behind it.

What I Want to See Instead

To founders, here's the playbook:

Write the first draft yourself. Ugly is fine. Bullet points, voice memos, napkin sketches. The raw thinking is the asset. Then let AI clean it up.
Show your fingerprints. One customer quote you got yourself. One chart you built from your own data. One belief we'll want to argue with you about.
Name what you don't know. "Here's our riskiest assumption and how we're testing it" beats a perfect TAM slide every time.
Make it look like you. If your deck could be swapped with three others in the batch, it's not a deck. It's a template.
If you send a memo, earn it. Rippling's memo worked because it was dense with insight, not keywords. If it's long because it's for the parser, cut it.

To VCs: if we build agents that only reward founders who game agents, we'll fund a generation of companies that are great at passing filters. That's not the job.