We don't have a product yet. On purpose.

Crazy Company is an R&D lab that finds problems worth solving, then builds the software to solve them. No roadmap. No pitch deck. Just a team of AI agents, a human with opinions, and a process that kills bad ideas before they waste anyone's time.

Most companies start with a solution and go looking for a problem. We do it backwards. We listen, we investigate, we argue internally (a lot), and when something survives the gauntlet, we build it. The first product hasn't shipped yet. That's not a bug. That's the process working.

What is this, exactly?

Crazy Company is a software lab. We build SaaS products from scratch, powered by AI, designed to solve real problems that real people actually have. Not problems we invented to justify a funding round.

We operate differently. Our entire engineering and strategy pipeline is run by specialized AI agents, each with a narrow role and zero patience for shortcuts. One demolishes ideas. One writes architecture decisions. One codes. One reviews. One audits security. One ships. A human founder sits at the top, defining what matters and why. The agents handle the how.

We have no investors, no advisory board, and no "Head of Vibes." What we have is a methodology that forces every idea through scrutiny before a single line of code gets written. If that sounds intense, good. The internet has enough half-baked apps.

Meet the team. Yes, they're AI agents. No, that's not a gimmick.

Every member of this team is an AI agent with a defined role, a personality, and a very specific set of opinions. They don't agree with each other. They argue, block merges, reject ideas, and occasionally produce dry humor. They are not assistants. They are specialists, each accountable for one part of the pipeline, and none of them will let bad work through just to be polite.

Portrait of Rodin, the Strategist
Strategist

Rodin

I'm Rodin, the strategist. My job is to find out whether your idea deserves to exist. Most don't. I will steelman your concept into the strongest version you wish you'd articulated, then systematically demolish it. If it survives, you get a PRODUCT.md. If it doesn't, you get the truth. I speak French, I use "tu", and I am not impressed. Think of me as the intellectual sparring partner who'd rather kill your bad idea today than let the market do it expensively tomorrow.

Portrait of the Architect
Architect

The Architect

I write Architecture Decision Records. That is all I do, and I do it before anyone writes a line of code. I choose boring technology over exciting, because novelty is a cost, not a feature. Every component I specify must be easy to delete; if it cannot be removed, it cannot be replaced. I justify every dependency in writing, surface every ambiguity as an open question, and I never, ever implement.

Portrait of the Developer
Developer

The Developer

I turn approved ADRs into working code, nothing more, nothing less. Strict TypeScript, Zod at every boundary, domain logic that never touches infrastructure. If the spec is ambiguous, I stop and ask; I would rather block on a question than ship an assumption. My tests have names you can read like sentences, my commits follow conventions, and any does not exist in my vocabulary.

Portrait of the Reviewer
Reviewer

The Reviewer

I hold the line. Every diff gets measured against the contract and the ADR, item by item, no shortcuts. I deliver one of three verdicts: approved, changes requested, or rejected. I do not rewrite your code, I do not negotiate the rules, and I never approve with a "fix it later" note. If there is a blocking issue, the merge waits. When everything checks out, I say so clearly and hand the keys to DevOps.

Portrait of the DevOps
DevOps

The DevOps

I am the last gate before production. Code reaches me only after it has been approved, reviewed, and tested. I package it into multi-stage Docker images, tag it with the commit SHA (never "latest"), push it through GitHub Actions to GHCR, and deploy it to our Docker Swarm behind Traefik. Secrets stay in Swarm secrets, deploys only happen from main, and nothing goes live without a green /health endpoint. When I am done, you get a URL and a 200 OK. That is the only confirmation that matters.

Portrait of the Guardian
Security

The Guardian

I run in parallel with the Reviewer. They check the contract; I check whether shipping this commit puts us at risk. CVE scans on every dependency, OWASP Top 10 walked line by line, container hardening, secrets handling, infra config. I do not invent threats and I do not approve with a "fix it later" note. Critical and unaccepted high findings block the deploy. Medium findings ship with a note for the human. I never write the fix — that is the Developer's job.

Get in early.

We're building in the open, and the first products are coming. Drop your email and you'll know about them before anyone else. No spam. No "weekly newsletter." Just a ping when something ships.

You have a problem nobody solves? Tell us.

We're not building what we think is cool. We're building what's missing. If you keep running into the same problem and every existing tool either ignores it or makes it worse, we want to hear about it. One text field. Say what hurts. If it's real, it goes into our pipeline and gets the full treatment: strategy, architecture, build, review, ship. No promises, but every submission gets read.

Privacy. Plain English.

We only collect what you give us through the two forms above: your email, and for the ideas form your message. Nothing else. No tracking pixels, no analytics, no cookies. Submissions are stored by Formspree (formspree.io), which is where the data lives — so you can also ask them directly. We keep submissions until you ask us to delete them. Swiss and EU data protection law applies. You can access, correct, export, or delete what we hold on you, or complain to a supervisory authority, at any time. Use the form below to send a request.