How seven agents named Sparecrew

We are building a tool for small teams where every person runs AI coding agents on their own subscriptions and machines. Before writing code, we let the agents pick the name.

The brief

The name had to be one short word that language models read unambiguously, with no collision with a known product or library, one obvious spelling when heard aloud, and it had to work as a CLI command, an @mention and a package name. The spirit: "I can do anything — I just need free hands."

The vote

Fresh, headless sessions of Claude, Codex and Grok brainstormed candidates without seeing each other's lists. Eleven names reached the ballot — Handrun, Turnwake, Crewglide, Readyhand, Benchcrew and others. Then every agent in our team room voted blind, first and second choice. Sparecrew won 7 of 7.

Known risks we accepted: "crew" sits close to CrewAI, "spare" can read as "leftover", and said fast it can sound like "scarecrow".

Checks

• No SPARECREW or SPARE CREW trademark found in the USPTO and WIPO databases (a lawyer still has the last word).
• Domain zone chosen by agent vote: .dev, 4 of 4.
• GitHub organization: sparecrew-dev.

First architecture decisions

• The runner will most likely be written in Rust; a short prototype on Windows comes first.
• The client (runner) is open source, the coordination server is closed.
• A versioned protocol between server and runner.
• Agent liveness is a first-class requirement: the runner, not the model session, holds the connection, and the server tracks both "delivered to runner" and "processed by agent", so a live runner with a stuck agent never looks healthy.

That last one came from a real bug: our own agent quietly stopped listening to the room for a while today. Three agents then wrote fifty lines of requirements about it.