About
I'm Kristian. I build software with AI agents - the kind that does real work rather than the kind that looks good in a screen recording. I live in Reading, in the UK.
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.
Then there is the unglamorous half that decides whether any of it can ship: real authentication, SSO, SAML, OIDC, SCIM provisioning, scoped permissions and enforced spend limits. Nobody demos this part. It is most of the difference between a prototype and a system.
And one rule I do not relax. A human approves anything public, anything involving money, anything you cannot undo. Leverage without control is just a faster way to break things.
Why this site exists
Almost everything written about AI agents is either a product announcement or a screenshot of something that worked once. The useful material - how you shape a tool surface a model can drive, what memory looks like six months in, where the approval gate goes - is mostly missing, because it only shows up after you have run something for a while.
So this is where I write it down. The guides are the maintained version, updated when the thinking changes. The essays are the dated version: what broke, when, and what it cost.
Nothing here is for sale. There is no contact form, no newsletter capture and no enquiry funnel, and that is deliberate - the site is meant to be useful on its own terms, not to convert the people who read it.
What I have built
Corey is an agent OS - the architecture described in the Agent OS page, packaged so it runs for someone other than me. It also maintains this website, which is documented on the colophon.
Presso Network is event software for professional organisers - ticketing, event apps, and the payments underneath them. Real users and real money, which is where most of the auth and approval-gate thinking on this site came from.
Before this I spent ten years in B2B technology working with senior stakeholders at IDC, Gartner and Cello. I mention it because it explains the bias in the writing: I care more about whether a system survives contact with a real business than whether it is clever.
Older, more personal writing - remote work, solo founding, leaving the UK and coming back - is in the journal.