Warp

Picture this: a Sentry alert fires, someone on your team types @web-factory in Slack, and a coding agent picks up the job. It checks out the right repo, pulls context through a shared Sentry MCP server, investigates the issue and opens a PR, while the team keeps working in Slack and Linear. That's the workflow Warp is built around now. The company that started as a modern terminal is pitching itself as "open infrastructure for cloud software factories," with fleets of coding agents handling code review, bug investigation, refactors and incident response across your SDLC.

The interesting part is how it's configured. Warp calls it factories-as-code: a factory.yaml file plus folders for agents, automations, runners, skills, webhooks, scorers and benchmarks. You declare which repositories agents can touch, which secrets they get (encrypted at rest and injected as environment variables only during runs), which MCP servers they share, and which cloud roles they can assume through federated AWS or GCP access instead of long-lived keys. Agent defaults set the model, runner and environment, and specialized agents can override them. Because it's all versioned files, you can measure and tune cost and performance the same way you'd tune any other code, and Warp says evals, benchmarks and self-improvement are built in.

If you just want a better shell, the Warp Terminal and Warp Agent CLI are still here, and Warp claims 800k+ developers use its products. But the homepage makes clear the bet is on teams running many agents in parallel, not one person chatting with one assistant. The listing shows a free tier with paid plans from $20/mo. The catch: the schema is labeled v1alpha1, and the scraped material is heavy on concept and light on hard limits, so expect to learn the details after signup.