How I work with AI
I don't prompt and hope. I build the system that lets agents do real work, and I can prove the result.
Spec-driven development
One spec per feature: what it does, how it behaves, and how I'll know it works.
Harness engineering
Tests, checks and references the agent can run on its own. If I can't verify it, I don't delegate it.
Multi-agent workflows
I split the work across focused agents, each with one job and only the context it needs.
Agent loops
Once I trust the checks, I set a goal and let the agents iterate until they pass. I still review every change before it ships.
Skills and project rules
Stack, conventions and what's off limits, packaged as reusable skills so every project starts from the same ground.
MCP
MCP servers connect my agents to the tools and data they have to reach.
I've also built LLM features: tool-calling agents, multi-step chat flows and structured data extraction.
- Context engineering
- RAG