Four phases. Hard gates. Your capability at the end.

Phase 01 · 2–6 weeks

Diagnose

We score your function across all five layers of the stack, map the value at stake process by process, and locate the binding constraint — the layer or process that caps everything above it. Every finding is priced and ranked.

DECISION GATE — proceed, re-scope, or stop — with the economics on the table

What you hold at the end

• —Readiness profile across the five layers, scored and evidenced
• —Value map: the cost of manual work, process by process
• —The binding constraint, named, with the reasoning
• —Ranked case for change with a sequenced roadmap

Phase 02 · 4–8 weeks

Design

Architecture, operating model, and control framework are designed together — never sequentially. Which decisions agents take alone, where humans approve, what the auditor sees, and what the platform looks like: settled before a line of build.

DECISION GATE — design sign-off by business, risk, and audit — together

What you hold at the end

• —Target architecture across the five layers, on your existing estate
• —Agent guardrail design: action boundaries, limits, kill-switches, escalation contracts
• —Control framework and attestation design, audit-ready by construction
• —Build/buy/wait decision per use case, with vendor-independent rationale

Phase 03 · release by release

Take over

Agents go on the runbook and we operate the process in production — outcome-priced, with evidence streaming from the first transaction. No big bang: each release takes over more of the work with its controls, its monitoring, and its measured value, and the run layer stood up for the first release is inherited by every one that follows.

DECISION GATE — each release: keep, fix, or roll back — on evidence

What you hold at the end

• —Agents, integrations, and models live in production
• —Operations platform: deployment, monitoring, evaluation, cost management
• —Value measured and reported per release against the Phase 1 baseline
• —Exception model operating, with your team in the queue from day one

Phase 04 · on evidence

Ratchet

Autonomy climbs A0–A4 on the Evidence Ledger, level by level, control by control. At Lights-Out the choice is yours: we keep operating the function as a service — or your team takes the run model, the governance, and the skills against explicit exit criteria. Lock-in is a design failure; the option to leave is contractual.

DECISION GATE — operate as a service, or transfer whole — your call

What you hold at the end

• —The function at Lights-Out — operated by us as a service, or by your team
• —Governance and attestation handed over, tested in operation
• —Skills transfer: builders, operators, and control owners trained
• —Exit criteria met and signed — with a standing option, not a dependency