Building an R&D org and process around AI agents at a connected-hardware company

I joined to build the R&D from the ground up for a connected-hardware product, there was no ready org to inherit. I hired the team myself, fully remote across five countries, and picked people for ownership and judgment rather than for a single narrow role.

Instead of copying the traditional structure with separate QA, DevOps and project management, I designed the whole operation around what AI agents can actually carry. Specs became precise, machine-actionable documents that agents work against. Verification moved into the pipeline, every change walks through automated tests, agent and human review, and deploy checks before it reaches production. Review shifted from reading syntax to reading intent, and releases stopped depending on any single gatekeeper.

Today a team of 5 to 7 engineers owns the entire product lifecycle, from spec to production, without the other traditional SDLC roles. Cloud infrastructure spend came down by about 30 percent along the way, and third-party integrations that used to take weeks now land 2 to 3 times faster.

The numbers above are the ones I can share in public. The rest, the team design, the toolchain and what failed on the way, I walk through on a call.