The AI development tools market runs on demos. A model builds a to-do app on stage, the clip travels, and nobody asks whether anyone runs a real business on the tool. We prefer to answer that directly. Summit builds Lume, and Summit runs its own products on Lume.
The test suite is production
Anthropos-hr is an HR and payroll platform for the Polish market. ExO is a classifieds marketplace, also in Poland. Both are built and operated by our own teams on Lume, and both ship faster because of it. Payroll has hard deadlines: if a run is late, people are not paid on time. A marketplace has uptime: if it is down, sellers leave. Neither can be tuned for a leaderboard.
The workflow is unglamorous. An engineer defines a task, SOLO — our autonomous coding agent — handles everything in between, and the result is reviewed before anything ships. In the Lume IDE, Tab predicts the next edit, agent teams split larger work, and one click hands a task to SOLO. The product teams and the Lume team sit in one firm, so when SOLO stalls on a task in Anthropos-hr, the report does not enter a beta queue. It goes to the people who ship Lume.
What benchmarks cannot tell us
Benchmarks measure an agent on tasks someone else defined. Daily use measures it on ours. Friction shows up in our repositories before any customer reports it, and that loop shaped the year: SOLO reached desktop, web and mobile in April, Lume v2 shipped in June with SOLO at its core, and we told the Lume IDE story in July. We have written that Lume changes the economics of building these products. That claim comes from our own ledger, not a commissioned case study.
Where the limits are
Autonomy has a boundary, and we keep it visible. A person still defines the task and reviews the result, and the review is a feature, not an apology: it keeps the cost of an agent’s mistake at the price of a rejected diff, not an incident. We are also honest about scale. ExO is a young marketplace, and two products in one market do not prove everything. They prove enough to keep going.
The rest of the platform stands on ordinary things: SOC 2, local-first design, MCP support, an agent framework for developers who want to build their own workflows. None of it is exciting. It is what running real products requires. If you are an investor, a partner or a developer and want to see how this works day to day, write to us.