Most people demo with AI. I ship with it.

I'm Kristian. I build software with AI agents. The kind that does real work, not the kind that looks good in a screen recording.

The shape is always the same. Agents with clear roles. Routing that sends work to the right one. Skills the repeatable jobs get encoded into once. And memory, so the system knows tomorrow what it learned today. Claude Code runs the orchestration. MCP wires it into the tools that actually matter.

I build MCP servers so someone else's AI can drive my software directly. I write Claude Code plugins and skills so the repeatable work collapses into a single command. When a new capability ships, I turn it into something working in days, not quarters.

Then there's the unglamorous half that makes AI software safe to hand to a business. Real authentication. SSO, SAML, OIDC, SCIM provisioning. The identity plumbing nobody tweets about and every serious deployment needs.

And a rule I never break: a human approves anything public, anything involving money, anything you can't undo. The agents do the work. A person signs it off. Leverage without control is just a faster way to break things.

This is not theory. It's in production - live systems, real users, real money moving through them. Not a prototype. Not a toy.

Some of what I've built: Corey, an agent OS that productised this whole system, and Presso Network, software for professional event organisers.

This site is where I write it down. Not a highlight reel and not a pitch. The actual build: what worked, what broke, and what the work is teaching me about building with AI.

That's the job. Build the thing, ship the thing, then explain exactly how, so the next person can do it too.