Blog
Releases, and agent taste tests: the same agent, the same job, several MCP servers, and what it said along the way.
Release
Baton 1.0: one open spec, three SDKs
The wire spec and the Python, TypeScript and Go SDKs reach 1.0. What is supported, how fast it is, and what comes next.
Agent taste test #2
We asked an agent to log 40 messages on 3 CRMs through their MCP servers. One took 3x as long and cost 4x as much.
Same 40 messages on Attio, HubSpot and Atomic CRM. Attio logs and searches one record per call, and the agent told us so while it worked.
Agent taste test #1
We asked an agent to tune 1 slow query on 3 Postgres MCP servers. It had notes.
Supabase, Neon and Postgres MCP Pro, one slow query, one one-line ask. 143 to 513 seconds, and 7 of 9 runs built an index nobody asked for.