Blog · Personal · 25 September 2026

What Investors Actually Want to See From an AI-Native Startup in 2025

Most AI pitch decks look identical. Investors want founders who've thought through data moats, architecture decisions, unit economics, and genuine points of vie

Right, so let's start with the uncomfortable bit. Most AI pitch decks in 2025 look identical. Same slide about the "step change moment," same screenshot of a chat interface, same vague line about "proprietary data" that nobody's actually interrogated. Investors have sat through hundreds of these now. They're not bored exactly, they're just... calibrated. They can smell a demo that's been optimised for the pitch rather than the product.

So what actually lands? I've been thinking about this a lot, and it's less about what's on the slide and more about what the slide reveals you've already thought through.

Start with the data moat thing, because everyone claims one and almost nobody has one. A real data moat isn't "we have data." It's a specific answer to why your data compounds and a competitor's doesn't. Does usage generate proprietary signal you can't buy or scrape? Does the feedback loop actually tighten your model or your product over time, or is it just... more data sitting in a bucket? Investors with an applied AI thesis (and there are a lot of them right now) have started asking this in the first fifteen minutes, not as a gotcha but because it tells them whether you understand your own defensibility or you're just repeating a term you heard in someone else's deck.

Architecture decisions matter more than people expect, too. Not the tech-stack-as-trivia version, nobody cares which vector database you picked. What they're actually listening for is whether you made deliberate tradeoffs. Did you build for agent orchestration because you genuinely need multi-step reasoning, or because "agentic" was the word of the month? Did you think about security by design from the start, or is that a slide you added after a portfolio company got burned? These aren't abstract concerns anymore. Agent infrastructure bets are a real category now, and the investors making them want founders who can explain, plainly, why their system is built the way it is. Not what it does. Why it's built that way.

Here's the thing about unit economics pre-revenue: investors don't expect you to have revenue, but they absolutely expect you to have a model. A real one. Not "we'll figure out pricing later," but a clear-eyed view of what it costs to serve a user, what happens to that cost as usage scales, and where the margin actually comes from once the model calls get cheaper (because they will, and if your business only works at today's inference prices, that's not a business, that's a bet on Nvidia's roadmap staying static).

This is where the Indie Hackers instinct actually helps, weirdly. There's a whole culture of founders who build in public, ship fast, charge from day one even if it's just a fiver, and treat the business model as something you test rather than something you narrate. That instinct, that discipline of actually knowing your numbers because you've had to, translates directly into what serious AI investors want to see. Not hype. Not a narrative about total addressable market. Just: here's what it costs, here's what we charge, here's what happens as we scale, and here's the assumption that could break it.

And then there's the point of view question, which is the one founders skip most often because it feels like the least "provable" bit. But investors are, in effect, asking you to bet with them, and they want to know you have an actual thesis about where this space goes, not just where your product sits today. What happens when the underlying models get better? What happens when a foundation lab ships a feature that looks suspiciously like your core wedge? If your answer is "we'll have moved fast enough by then," that's not a point of view, that's a hope. A real point of view sounds more like: here's what we think stays hard even as models improve, here's the layer we think persists.

None of this is generic pitch advice, and I want to be honest, I'm slightly wary of even writing something that could be mistaken for a listicle of Things To Include In Your Deck. Because it's not really about the deck. It's about whether the thinking happened before the deck existed.

Which is, honestly, where a studio with real portfolio exposure earns its keep. Not by polishing the pitch, but by having sat inside enough of these problems, architecture calls, pricing models gone wrong, security decisions made too late, to know which questions an investor is actually going to ask underneath the question they say out loud. A studio that's built alongside multiple AI-native teams has pattern-matched across failure modes founders inside a single company simply haven't seen yet. That's not a shortcut. It's more like having already had the argument with yourself before someone else has it with you in a meeting.

Building seriously, in this environment, doesn't mean building slowly or cautiously. It means being able to answer the boring questions with the same confidence as the exciting ones. Investors in 2025 aren't dazzled anymore. They're just trying to work out who's actually thought it through.