New — live in production across lending, industrial & clinical. See the customers →
Blog / How we work
How we work · field notes

Inside an Aminobots engagement.

Four ways in, one entry point. Diagnose is the wedge. Then you decide whether to Deploy, then whether to Operate, and License is there for teams who want to bring the Blueprint in-house. Here's what each stage actually delivers.

We're an implementation company, not a platform you log into and hope. An engagement moves through four stages in the order most customers actually travel them — Diagnose, Deploy, Operate, License. Each stage stands alone, and each one earns the next. Nobody is asked to commit to a build before they've seen the evidence.

Stage one — Diagnose.

Diagnose is the wedge, and every engagement starts here. It is a two-to-three-week, written, evidence-backed scorecard of your current or planned AI system. We run the Gate 0–4 assessment, a cascade test to find where errors would amplify, a token budget audit, and a CLEAR baseline.

What you get back is a document, not a sales deck: a topology recommendation, a remediation plan, and honest cost ranges. There is no commitment to build at the end of it. Plenty of Diagnose engagements end with us telling a customer their system doesn't need agents at all — and that's a successful Diagnose.

Stage two — Deploy.

If the Diagnose says build, Deploy is the full lifecycle: Blueprint, build, deploy, handover, typically over six to twelve weeks. CLEAR telemetry is plumbed in from week one. HITL checkpoints sit at every high-stakes boundary. And critically, we deploy in your tenancy, under your identity, with your keys — the system runs in your cloud, not ours.

At handover you receive everything: full source, the runbook, and the eval harness. There is no black box you have to keep paying us to operate. You could take it in-house the day we hand it over.

Stage three — Operate.

Most customers choose to have us keep the system healthy rather than carry that load themselves. Operate is ongoing operation — 24×7 observability, model upgrades as better models ship, prompt tuning, and a quarterly methodology-compliance audit. This is also where the system gets better: every deployment we operate feeds what we learn back into the next one.

Stage four — License.

For teams who want to bring multi-agent implementation in-house, License gives your engineers the Blueprint System and Workbench directly. It's an annual platform license with a per-deployment seat fee. You stop hiring us to build and start building the way we build — same gate discipline, same eval harness, your own team shipping.

What a first year looks like.

Concretely, take a mid-tier NBFC on a BFSI lending pattern. A three-week Diagnose, a Gate 0–4 audit. A ten-week Deploy of a Helm-shaped lending system. Then ongoing operation beginning in month four. The first year lands around the cost of a single senior hire — and at the end of it the customer owns the system outright, with the option to keep us operating it or take it in-house.

"The order isn't an upsell ladder. It's the order in which we earn the next yes."

Diagnose earns Deploy. Deploy earns Operate. And License is always available the moment a team decides they'd rather own the capability than rent it.

Begin with a Diagnose.

2–3 weeks. A written scorecard with a topology recommendation, cost ranges, and a remediation plan. No commitment to build.

Book a diagnosis