Vera 1.0: running a real operation

As of this month, Vera 1.0 is in production with a paying Finnish field-service customer. Not a pilot, not a proof of concept running next to the real process — the actual process. Disruptions come in, Vera proposes resolutions verified against the company's own rulebook, and dispatchers approve them with one click.

I want to be precise about what we mean by production, because the word gets abused in AI announcements.

Production means money. The customer pays a monthly subscription because the product saves senior-dispatcher time every single day. Vera has been revenue-funded from the start; this milestone is why.

Production means the rules are theirs. During onboarding we ran a rule-capture workshop with their dispatchers — the people who carry the operation in their heads. SLA tiers, certification requirements, working-time limits, customer quirks. All of it now lives in the rulebook in their own words, with the machine-checkable form alongside. When a rule changes, they change it themselves. No ticket to us.

Production means approval mode. Every operationally significant decision waits for a human click. We could run more autonomously — the engine supports it — but trust is earned per decision class, not granted by a settings page. Some routine classes will graduate to autonomous execution over the coming months, with the audit trail as the safety net.

Production means boring reliability. The deployment is a handful of Docker services on modest hardware. No GPU. The constraint solver does the demanding work, and solvers are gloriously cheap compared to the things people usually throw GPUs at.

What surprised me most wasn't technical. It was how quickly dispatchers started trusting the audit trail in conversations with each other — "look at the trace" has become how they settle disagreements about why something was scheduled the way it was. The proof turns out to be a communication tool, not just a compliance artifact.

There's a second deployment in onboarding now. More on that soon.