The lab
RiftSeed is a software engineering lab. It develops a method for building software with AI agents under control and the tool that applies it, and puts both to the test on its own platforms.
It works on its own projects, and that is where the method gets written: some of its rules came out of a failure in production.
Ratchet Agentic Engineering collects the rules that code agents work under, taken in part from safety-critical software. A request is first checked against the requirements that already exist, and every change starts from an impact analysis. When something breaks in production, the incident closes only with a regression test that reproduces it, and from then on the quality ceiling does not loosen again.
This is where agents receive the work and leave the evidence. Claude Code and Codex connect through MCP: each session opens with a briefing and closes with its commits, files and tests on record. Requirements, CI results, incidents and verdicts, that is human release approvals, are kept together, project by project.
Agents propose requirements. Which decisions need a signature is set by the project’s level, chosen at the start. From level 1 a person signs the requirements and gives the verdict on every release, with the test results and any open incidents in front of them. An agent can stop a release, never sign it.
The method · How Cockpit works · The platforms
Questions about the method, about Cockpit or about early access: one line is enough. Contact
riftseed.com is run by a private individual, in a personal capacity. Privacy · Terms