I mapped the full AI agent infrastructure market, 6 layers (orchestration, memory, MCP/tool use, evals, deployment, security), 25+ companies, and $600M+ in recent funding rounds.
Key findings: orchestration is commoditizing fast (LangGraph/CrewAI/AutoGen/OpenAI SDK all converging), the real moats are in memory and evals (Mem0, Braintrust, Arize all raised big recently), and Anthropic/OpenAI are moving up the stack aggressively.
Also covers: MCP ecosystem timing, which layers are acquisition targets, valuation multiples vs. DevOps comps, and three whitespace opportunities that haven't been built yet. Free to read, no paywall on this one.
Happy to discuss/debate any of the takes in the thread.