Kerber.ai: A Venture Studio Powered by AI

2026-01-31

After a year of AI coding experiments—tool-hopping, disasters, and finally finding what actually works—I wanted to put it all into practice.

Today I’m soft-launching kerber.ai.

What Began as Curiosity

I’ve spent the last year deep in the AI coding trenches. Cursor, Claude Code, Copilot, and a graveyard of abandoned configs. I’ve written about my journey from chaos to actually shipping code.

The biggest insight? AI is 10x faster at setup. Not the coding itself—the setup. Build processes, testing pipelines, documentation, CI/CD configs. That’s where AI shines.

The actual coding? Maybe 2-3x, not 10x. And that’s fine.

I wanted to take these learnings and build something real with them.

The Setup

Kerber.ai is a venture studio with an unusual structure:

→ Me (20+ years product experience)
→ Henry (AI collaborator orchestrating 100+ specialised agents)
→ A workflow that lets us ship like a large team

Henry isn’t magic. When I define an objective, Henry breaks it down, routes subtasks to appropriate specialist agents, and synthesises the results. Code review goes to code specialists. Design feedback to design agents. Research on research tools.

The result: consistent output across disciplines, available 24/7.

Testing a Modern Stack

I also wanted to test a modern stack from scratch:

SvelteKit with Svelte 5 runes ($state, $effect, $derived, $props)
Tailwind CSS v4
Deployed on Vercel

Fast. Clean. No bloat.

The site supports 14 languages out of the box—from Swedish to Japanese to Arabic. Because when you have AI helping with translations, why not?

The Workflow

Every feature follows the same flow:

Plan first — AI proposes, human reviews. We iterate until the approach makes sense.
Implementation — AI writes code while I review diffs in real-time. Like pair programming with a talented junior.
Verify everything — Trust but verify. AI will confidently generate garbage. You need to catch it.
Document failures — When AI does something stupid, it goes in the docs. Institutional memory beats repeated errors.

The key: AI handles tedious work (tests, configs, boilerplate) so I can focus on architecture and hard problems.

Is This a Full-Time Company?

Not yet. It’s a side hustle alongside client work and existing projects.

But I’m onto something here. The velocity is real.

We’re building a few products—a community/dating app, a productivity tool, and a marketplace. Each is at a different stage. Check kerber.ai/ventures if you’re curious.

Honest Truths

Some things the hype cycle won’t tell you:

AI won’t make you 10x productive overnight. Anyone who says otherwise is selling something.
Setup time is real. Good workflows take weeks to build.
The real winners are senior developers. Not because we write better prompts—because we understand what good code looks like.
80% of code gets rewritten anyway. So why put prestige on writing it yourself?

The goal isn’t perfect code. The goal is shipping the right solution, with the right architecture, using the right tools.

Let’s See Where This Goes

Kerber.ai is deliberately minimal right now. A landing page, a process description, and a calendar link.

I’ll be sharing more about:

How I structure AI agent workflows
Lessons learned from shipping fast
What works and what doesn’t