Atelier Ops
GitHub Email us

Approach

A four-phase contract. Same shape every time.

Every engagement runs the same four-phase shape — discovery, ADR, migration, handoff — over six to twelve weeks. We take one problem at a time, write the decisions down, and leave a road your team can keep walking after we go.

Email us See services
  • 6–12 wk engagement length
  • ADR archive per engagement
  • Documented exit path on day one
Engagement length
– wk

discovery through 30-day shadow

ADRs published
–

Nygard format, per engagement

Disciplines per engagement
–

architecture · DevOps · SRE · SecOps · MLOps · IDP

Handoff packet
ADR + runbook + evidence

signed off by named owner before exit

  1. Discovery (1–2 wk)

    Short, sharp interviews with engineering leadership and the operators carrying the pager. We finish with a one-page brief — the problem we are solving, the problem we are not, and the success metric — because a fortnight of focused conversation beats a six-week audit nobody re-reads. Scope-as-deliverable: by Friday of week two, both sides sign a go / no-go memo or we walk.

    • 5–8 stakeholder interviews (eng leadership + oncall)
    • System map & dependency inventory
    • One-page brief: scope, non-scope, success metric
    • Go / no-go memo signed by both sides

    Not covered Formal compliance audit, vendor RFP, hiring plan. SOC 2 / ISO evidence comes out of the migration phase as a by-product — see SecOps service.

  2. ADR & dissent (1 wk)

    We draft the architecture decision record in Michael Nygard format — context, options, decision, consequences — over the lighter MADR template because Nygard forces the consequences column, and the consequences column is where engagements die. Your team gets a structured dissent window with a named blocker right: anyone on your side can stop the ADR before it ships. Decisions visible, never made in Slack.

    • ADR draft (Proposed status, Nygard format)
    • Trade-off matrix vs 2–3 named alternatives
    • Structured dissent window (anyone on your side can block)
    • Accepted ADR + migration outline

    Not covered Proof-of-concept code, vendor procurement, contract negotiation. The ADR is a written commitment, not a working system — that lands in migration.

  3. Migration (3–6 wk)

    We ship a paved road, not a platform — narrow contract, opinionated defaults, exit path documented on day one — because a snowflake migration nobody owns by week eight ships nothing. Adoption is gradual: one team migrates first, we run the postmortem with them, then widen the road to the rest. The work runs in an isolated company workspace over a shared tenant, so blast radius stays bounded and your secrets never leave your boundary.

    • Reference implementation + ≥1 migrated workload
    • Runbooks, SLOs, oncall template
    • Adoption dashboard with weekly delta
    • Documented exit path — you can leave us behind

    Not covered Parallel platform rewrites, staff augmentation, third-party SaaS migrations outside the engagement contract. One paved road per engagement — we refuse the second road mid-flight.

  4. Handoff & 30-day shadow (1 wk + 30 days)

    We hand the road to a named owner inside your org and shadow oncall for thirty days — not an open-ended retainer — because evidence over pageantry means the engagement ends with a runnable system, a packet your auditor can read, and the option to walk away. After the shadow window we are done, or we agree on the next narrow problem in writing. A handoff with no named owner is not a handoff; we will reschedule before we ship one.

    • Handoff packet: ADRs, runbooks, evidence trail
    • 30-day shadow oncall with named owner
    • Final retrospective with engineering leadership
    • Optional: next-engagement scoping memo

    Not covered Permanent oncall coverage, managed-service contracts, body-shop retainers. If you need indefinite operations, hire — we will help you write the JD instead of becoming the JD.

Operating principles

What we won't bend on.

Decisions are recorded.

If it isn't in an ADR, it didn't happen. We refuse to make a load-bearing decision in Slack — the future team needs to read the why, not reconstruct it.

Paved roads, not platforms.

A road has a narrow contract and a documented exit. A platform grows commands until nobody can hold it in their head. We refuse the sixth command.

Evidence over pageantry.

Compliance artefacts are produced as a by-product of how we work — attestation, audit trail, signed runbooks. No separate "audit prep" track.

One problem at a time.

We run at most three concurrent engagements and refuse to widen scope mid-flight. The work that gets shipped is the work we said we would ship.

We refuse work we can't ship.

If discovery shows the problem is org-shaped rather than tech-shaped, we say so on day three. Sometimes the right deliverable is a memo, not a migration.