How I Sort AI Pitches at Hello World — The Two-Camp Framework

If you are raising for an AI startup right now, you are walking into a sorting hat. Every investor has a private framework that decides — usually in the first six minutes of a call — whether you are in the bucket they fund or the bucket they politely pass on.

Most investors don't tell you what's in the framework. That's how the game works.

I'm going to tell you what's in mine.

I'm a GP at Hello World Investments. We back early-stage founders, with a real bias toward people who came from creative industries — music, film, design, games — and are now building technology. I see roughly 100 AI pitches a year. They collapse into two camps.

The two camps

Camp 1 says: "We have better models."

Better fine-tuning. Better RAG. A clever new agent harness. A novel use of a specific frontier capability. The whole pitch is a technical wedge against a model-layer commoditization curve that, in practice, eats the wedge every six months.

Camp 2 says: "We have better judgment about what models should do."

They show me the features they killed, not the ones they shipped. They talk about which 1% of failure modes will gut user trust. They obsess over tone before they obsess over logic. They have a specific user — a real human, in a real moment, with a real feeling — and the whole product is the answer to *what does that human need right now*.

Camp 1 will lose.

I have watched four Camp 1 startups in the last eight months get reduced to features by a single OpenAI release. The moat is sand at the model layer because the layer moves too fast to defend. Every fence you draw gets filled in by the next API drop.

Camp 2 doesn't have that problem.

The moat isn't a technical artifact. It's the residue of thousands of small, opinionated decisions about what a real human in a real moment actually needs. Models commoditize. Decisions about humans don't.

The pattern I keep noticing: Camp 2 is mostly ex-creatives. Designers. Musicians. Writers. Game devs. People who spent ten years training the "is this good?" muscle before they ever wrote a line of code. That muscle is what AI products run on now.

What I actually check before I write a check

Once I've sorted you into a camp, I run five questions. They are deliberately not technical. The technical bar is table stakes — if you cleared the door, I assume the code works. The questions that decide my check are these:

1. Can you tell me, in one sentence, what your user does five minutes before they open your product?

This is a taste check disguised as a discovery question. Founders who can answer it sharply have a real user in their head, with a real life, with a real moment your product attaches to. Founders who can't answer it are usually pitching a solution looking for a problem.

If you say "they're stressed at work" — pass. If you say "she just got off the call with her oncologist and is sitting in the parking lot trying to figure out what to ask her husband when she gets home" — I'm listening.

2. Show me a feature you killed.

Camp 1 founders never have an answer. They built the feature, it works, why would you kill it? Camp 2 founders have a list. They killed it because it took a serious moment and turned it into a logistics task. They killed it because the tone of the agent's response broke the product's whole vibe. They killed it because users were using a slower-but-clearer flow they already trusted and the AI version was muscling in on it.

Show me one feature you killed and *why*, and I have most of what I need.

3. What does the worst possible AI response in your product look like?

Not "if" wrong. *When* wrong. Every AI feature is wrong some percent of the time. The question isn't whether you can prevent it. The question is whether the failure mode is recoverable.

Founders who can describe the specific worst case — who the user is, what they asked, what they got, what they did next — have thought about the bar of trust they're operating against. Founders who say "we have guardrails" are about to lose a category of users they don't know yet exists.

4. What's the smallest possible version of this that would make ten people obsessed?

This is the wedge test. Camp 1 wants to build a platform. Camp 2 wants to build a thing ten specific people can't live without. Platforms in AI lose because the model layer eats horizontal plays. Wedges win because the moat is the relationship with the user, not the breadth of the offering.

If you can't tell me the ten people, you are not ready to raise from us. If you can, the rest of the conversation is fun.

5. Why are you the right person to do this?

The honest version. Not the deck version.

Ex-creatives have a real answer here that engineers often don't. *"I spent ten years making music and the thing that made me good was knowing when a song was finished — that's the same skill that ships an agent product."* That answer is a yes from me. *"I worked at Google for three years"* is not.

This is where the artist-to-founder leap becomes its own credential. Not despite the music / film / design background — because of it.

What not to pitch us

Save yourself a no:

• A horizontal AI assistant for everyone. Wedge or pass.
• A model fine-tuned on proprietary data. This isn't a moat anymore. It's a six-month head start.
• A "ChatGPT for [X]" framing. If your whole positioning is a reskin of a frontier API, you don't have a company. You have a feature.
• A pitch that doesn't mention a single specific user. Personas are not users. I want a name, an age, a moment.
• A deck with zero mention of failure modes. I assume this means you haven't thought about them.

If you read all five and thought "that's not me, but I know someone" — send them this essay.

What we actually fund

Founders from creative industries — touring artists, designers, filmmakers, game makers, writers — building technology, especially in AI, especially with a real user obsession. Pre-seed and seed. We are happy to be the first check.

You don't need a technical co-founder. The vibe-coding stack collapsed that bar — if your taste is real, the leverage is already in your hands. If you can articulate the user, the moment, the wedge, and the kill list, the technical execution is solvable.

The pitch

If you read this and think *that's me*, the easiest path is direct. There is a form at samrenders.nl/pitch that goes straight to me. Name, email, what you're building, current traction, links. I read every submission personally. No warm intro required — though if you have one, that's faster.

Not ready to pitch yet but want to ship better AI features in the meantime? Grab the companion 6-page playbook — 3 Questions Before You Ship Any AI Feature — same operator framework, distilled.

I won't promise to write a check on every founder this resonates with. I will promise to read what you send, and if there's a fit, to move fast.

The wrapper apps will keep raising A rounds. The architecture founders will keep shipping. Twenty-four months from now, the difference will be obvious to everyone. It is already obvious to me.

— Sam