ai engineer · building agentic tools, in public
i build small, sharp agentic-ai and claude code tooling that does real work in production: the unglamorous kind that holds up. most of it starts as something i needed myself and turns out useful to anyone building with agents. i build in public on x: the patterns, the tools, and the dead-ends.
i build for the way my own brain works: externalise the fiddly parts, make the system do the remembering. it turns out that's good design for anyone working with agents.
day job: on the engineering team at quicknode, across emea, building ai tooling on top of web3 infrastructure.
my thesis: ai agents are the next billion users of the internet, and the tooling layer is being built right now.
- small stdlib tools for agent engineering: eval, observability, context, mcp
- claude code skills and workflows for real day-to-day agent work
- mcp servers that connect ai models to live systems, apis and blockchain data
- building and writing in public about agentic-ai patterns
- fable-skills · 35 engineering-discipline skills authored by fable 5, plus a blind-graded benchmark. sonnet 5 loaded with the skills beat opus 4.8 running bare; both models improve in every area. no deps, mit.
- agent-eng-kit · small stdlib-only tools for agent engineering: a prompt-injection gauntlet, an mcp record/replay proxy, a session profiler, a context packer, an mcp doctor. no deps, no keys, mit.
- legwork · a project queue for claude code that survives you walking away.
/wrapmints the next prompt while context is hot;/pickupbriefs you back in thirty seconds. install as a plugin, or clone for the static dashboard and the optional unattended runner. - claude-code-tools · zero-config doc linters for claude code projects.
