Senior Software Engineer · AI Agents
Nine years building software. The last three building AI agents that run in production.
By day I am a Tech Lead at Nationale-Nederlanden, the largest insurer in the Netherlands, working in the claims domain: new product lines, legacy systems moved into modern services, and the internal tooling other teams build on. Insurance and banking taught me that shipping software inside a regulated company is a different sport than shipping it anywhere else.
The rest of the time I build FiveTimesFaster, a multi-tenant AI-agent platform I wrote solo. Every company that signs up gets its own isolated agent that builds and runs workflows across 500+ tools. I built the agent runtime, the workflow engine, the billing, the interface, and the infrastructure that keeps one customer's agent away from another customer's data. Two-minute demo.
The interesting problems have not been the models. They have been the boring parts around them: what happens when an agent hallucinates a tool call, what happens when someone closes the tab mid-run, what happens when the credits run out at 3am.
| Project | What it is |
|---|---|
| icpfinder | Paste a product description, get buyer archetypes with verified decision-maker emails. Structured LLM extraction, published on npm. |
| copilotkit-langgraph-history | Agent runs that survive a refresh: rehydrating LangGraph conversation history in CopilotKit. Written after I hit the problem in production. |
| MicroFes | Module Federation demo from my micro-frontends talk. Separate builds, one application. |
TypeScript Python LangGraph MCP Next.js NestJS PostgreSQL AWS CDK
I podcast and write at danielfrey.me about what building actually looks like, including the parts that go wrong.
If you are building agents in production, I am always up for comparing notes.