Factory Droid

Droid's trait is that it is everywhere: the same agent runs as a CLI, as a Slack or Teams bot, and headless in CI, so a team can standardize on one tool and one set of rules. The daily win is the handoff, a product manager asks in chat and the answer arrives as a branch, with the same custom droids the engineers use at the terminal.

Pick it for that consistency, especially where non-engineers hand off tasks in chat and you are tired of being the relay. Pick OpenCode if you want readable source and your own model keys with no plan in between.

The meter is the risk, and it is an unusual one. Pro, Plus and Max are governed by rolling rate limits that are not published on the pricing page, so a limit halts your afternoon rather than charging for it, and you cannot plan around a ceiling you cannot see.

The consequence is that the plan tier is a guess until you have run it for a month, and the failure mode is silence at the worst moment. Publishing the limits would change this verdict. What it does right: bring-your-own-key reaches local Ollama, so a rate-limited team can route around the plan entirely.

Factory reports 58.8% on Terminal-Bench for Droid with Claude Opus 4.1, September 2025. The figure is self-reported, but it names the leaderboard, the model and the date, which is more than most vendors on this board manage. Terminal-Bench measures harness plus model, so the score is not Droid's alone.

The harness is documented as an MCP client, hooks that fire around tool calls, and a headless droid exec surface for scripts. The source is closed, so the edit and verification loop cannot be inspected, only inferred from the score. The observation: the number is reproducible in principle and the mechanism is not.

Look at the ladder: $20, $100, $200 for individuals, then a Teams tier with a base fee plus a per-seat charge. Buyers at every altitude, which is what a company does when it has decided to sell to the org chart rather than to the individual engineer. Rolling rate limits rather than credits is a margin-protection mechanism dressed as simplicity.

Moat: the chat and CI surfaces, which raise switching cost once a team's droids live there. Likely acquirer: Atlassian or ServiceNow, someone already selling to the same CIO. Position: buy the Teams tier, negotiate the limits into the contract.

The demo answers in Slack and Teams. Procurement: Teams is $60 a month plus $40 per seat, so sixty seats is roughly $2,460 a month before the rolling rate limits, which are the number nobody can budget. BYOK routes inference through our own model contract, which keeps the data terms we already negotiated. SSO, audit logs and retention are not stated on the pricing page.

CI fit is real, headless mode drops into a pipeline step, and onboarding for a mid-level engineer is a curl install and a login. Approved with conditions: rate limits and retention written into the contract, and BYOK mandated so the model terms are ours.

I cannot read it. Proprietary binary, prompts I cannot see, a loop I cannot patch. But the BYOK doc lists any OpenAI-compatible endpoint and local Ollama, MCP servers go in a plain JSON block with a type and a url, and droid exec means I can drive it from a Makefile and pipe the output wherever I want. Hooks fire on tool events, so I can wrap it without forking it.

That is the proprietary agent I resent the least: configuration is mine, model is mine, execution is theirs. A fork cannot survive the vendor because there is nothing to fork. Grudging respect, with the usual asterisk.