Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
Claude Dual-Agent Development Loop
See a live demo of building a 44-tool MCP server using two collaborating Claude agents: one builds, one tests, refining specs through use.
I built a 44-tool MCP server that gives Claude full control over task management. But this talk isn’t about the architecture — it’s about how I built it with two Claude agents collaborating: one writes code, one uses the tool in real workflows.
Live demo of the development loop:
- Tester agent hits friction (“reordering 5 tasks took 11 API calls”)
- Builder agent implements fix
- Tester validates — in minutes, not days
The punchline: specs emerged from use, not upfront design. Started with 33 tools, ended with 44 — each one added because the tester needed it.
I’ll show the actual conversation flow and where this is heading: config as a shared channel for agent-to-agent communication without the human in the loop.