May 7
Loading video player...
We had a system with around 200 agent skills, all generating code. Every time an agent needed Facebook Ads data, it wrote 200 lines from scratch. Same for Google Ads, Redshift, everything else. No agent knew what the others had already done. In this video I walk through how SenseLab fixes that, in 4 acts: 00:00 The problem (200 skills, all amnesiacs) 01:00 Setup: 4 agents generating the same code independently 02:00 Act 1 — Your agents start remembering (search, readFrom, history) 07:00 Act 2 — An agent audits your other agents (redundancy detection, 38k wasted tokens found) 12:00 Act 3 — Your agents coordinate (Entity Rooms, structured negotiation) 17:00 Act 4 — From patterns to libraries (knowledge graph, cortex briefings) 20:00 Getting started (2 env vars, 5-line check-before-generate loop) SenseLab works with CrewAI, LangGraph, AutoGen, Cursor, Claude Code, any MCP client. Python, TypeScript, HTTP. What you'll see: → A new reporting agent finding existing code with one search call and skipping 200 lines of generation. 4,200 tokens saved on a single call. → An ops review agent inventorying 4 agents, flagging that campaign-data is being generated by 3 agents independently (14,000 tokens), facebook-ads-api by 2 (11,600), ad-platform-fetch by 3 (12,300). → The Facebook agent and Google agent negotiating a shared getAdPlatformData function in a structured back-and-forth, not a chat. → The full knowledge graph and cortex briefing surfacing what's ready to extract. Start free: https://www.sense-lab.ai/ Docs: https://docs.sense-lab.ai GitHub: https://github.com/raia-live/amfs #multiagent #aiagents #crewai #langgraph #claudecode #cursor