Ramp says agents wrote all the code for Laminar, its in-house agent harness

Ramp says Laminar, its in-house agent harness, was built in six weeks with code generated entirely by agents steered by experienced engineers.

Ramp has published an October 21 account of building Laminar, the in-house agent harness behind its agentic products. Ramp says the first version took six weeks, built by "a small team of obsessive, experienced engineers who worked with and steered the agents," and that its code "was generated entirely by agents." For builders, this is a production example of agent-written infrastructure, and Ramp is clear that it still took senior engineers to get there.

According to Ramp, Laminar handles secure conversation history storage, session compaction, standard tool-calling interfaces, message queuing and steering, subagents, multiplayer chat and observability tooling. Ramp adds a caveat: building in-house has become cheaper, but it still needs good engineers to be on-call, keep the platform healthy and work out what product teams actually need.

daily.dev's summary of the post adds more detail. It says Ramp evaluated the OpenAI Agents SDK, Claude Agents SDK, Claude Managed Agents API, OpenCode, LangChain and Sierra before building its own. Per that summary, Ramp chose to build so it could stay model-agnostic and own its observability. The summary also says the stack runs on Postgres, Temporal and S3, the code was generated with Opus 4.7 and GPT 5.5 (a detail Ramp's own post doesn't name), and every agentic product at Ramp now runs on Laminar.

What's still unclear is how the engineers reviewed, tested and steered the generated code. Those are the practices that would make this account repeatable for other teams.

If you're weighing whether to build your own harness, freeCodeCamp's TypeScript harness walkthrough covers the core pieces: step limits, tool validation and failure testing.

Sources