Blog
Field notes from building AACP
Honest accounts of what worked, what didn't, and what turned out to matter more than expected.
AACP added to the LangChain integrations directory
aacp-langchain is now listed in the LangChain Python integrations directory as a community integration. What that means, and what it does not mean.
What auditability actually means in a multi-agent workflow
Auditable AI usually means logging model outputs. AACP treats structured coordination messages as the audit trail itself. The difference matters when an auditor asks for it.
I tested AACP against four agent frameworks. Here is what I found.
Same workflow, same data, same model, four frameworks: LangChain, CrewAI, AutoGen and Pydantic AI. 59 coordination hops each. The saving scales with how verbose the framework's default coordination is.
AACP and JSON: why not just use what already exists?
JSON is excellent at what it does. AACP is not a replacement for it. An honest answer to the most common first question about the protocol.
The hidden cost in multi-agent LLM systems
Every agent-to-agent coordination hop in a default LangChain or CrewAI system costs an LLM call. Across a realistic department workflow, that is 59 calls before any task work begins.
I built a coordination protocol for multi-agent LLM systems
An honest account of how AACP started as a token-saving experiment and ended up being more interesting for determinism and audit trails than for compression.