Next platform deadline: Amazon EKS Nov 26, 2026  

All 14 deadlines

For engineering leaders

How the work actually gets done.

Phase gates, artifacts, who does what, how we test and roll back, and what you hold at the end. The same method across data, cloud and AI.

Phase gates

Six phases. Nothing moves on without a sign-off.

Each phase has a defined start, a set of artifacts and a gate your owner approves. Small engagements combine phases; none are skipped. Practice detail: data, cloud, AI.

  1. 01

    Mobilize

    Week 0–1

    Starts with

    Signed SOW, NDA and access request

    Artifacts

    • Kickoff deck: goals, scope, contacts and cadence
    • Access plan: read-only first, named accounts with MFA
    • RAID log opened
    • Communication plan and escalation path

    Gate

    Access in place; plan for discovery agreed

  2. 02

    Discover and assess

    Weeks 1–3

    Starts with

    Read-only access or exports

    Artifacts

    • Inventory of systems, versions, dependencies and support dates
    • Baselines: performance, cost, reliability or evaluation scores
    • Findings and risk register with ratings
    • Options with trade-offs

    Gate

    Written assessment accepted by your technical owner

  3. 03

    Design and plan

    With the assessment

    Starts with

    Accepted findings

    Artifacts

    • Target architecture and architecture decision records
    • Wave or release plan with windows
    • Test strategy and acceptance criteria
    • Rollback approach for every change

    Gate

    Design review sign-off; change approved through your process

  4. 04

    Build and rehearse

    Per wave or release

    Starts with

    Approved design

    Artifacts

    • Infrastructure and configuration as reviewed pull requests
    • Automated checks in CI
    • End-to-end rehearsal with measured timings
    • Runbook and rollback tested in rehearsal

    Gate

    Go/no-go criteria met; rehearsal report reviewed

  5. 05

    Cut over or release

    In your window

    Starts with

    Go decision

    Artifacts

    • Runbook executed step by step with named owners
    • Data, smoke and performance checks against the baseline
    • Change record updated in your tool

    Gate

    Agreed checks pass; your owner signs off the change

  6. 06

    Hypercare and hand-over

    Agreed per SOW

    Starts with

    Change accepted

    Artifacts

    • Close monitoring and tuning after go-live
    • Runbooks, diagrams, ADRs and dashboards in your repository
    • Recorded knowledge-transfer sessions
    • Close-out report: what was done, measures against baseline, open items

    Gate

    Hand-over accepted; access revoked; material returned or deleted

Roles

Who does what

R = responsible, A = accountable, C = consulted, I = informed. Your team stays accountable for scope, approvals and acceptance; we are accountable for the engineering. Adjusted per SOW.

Sample statement of work

Who does what: responsible, accountable, consulted, informed
ActivitySRASHYour team
Scope, priorities and acceptanceCA/R
Access provisioning and approvalsCA/R
Discovery, assessment and findingsA/RC
Architecture and design decisionsRA
Build, automation and pull requestsA/RC
Code review and merge approvalRA
Change approval (CAB) and windowsCA/R
Runbook execution and rollback callsA/RC
Application smoke testsCA/R
Status reporting and RAID logA/RI
Documentation and knowledge transferA/RC

Testing and rollback

Proved in rehearsal, not hoped for in production.

Every production change has go/no-go criteria, a tested rollback and checks against the baseline captured in discovery. Sample cut-over runbook

  1. Data correctness

    Row counts, checksums and sampled record comparisons between source and target for every migrated table or collection.

  2. Performance

    Critical queries and endpoints compared with the baseline captured in discovery, under representative load.

  3. Resilience

    Failover, restore and node-loss tests with measured recovery times, not assumed ones.

  4. Infrastructure

    Policy and plan checks on every Terraform or GitOps change before it merges, and drift detection afterwards.

  5. AI quality and safety

    Evaluation suite scored on every change, plus permission-leak and prompt-injection tests before release.

  6. Rollback

    Every rollback path rehearsed and timed before the production change is approved.

Engineering standards

Everything as code, everything reviewed.

  • Infrastructure, configuration and pipelines as code (Terraform or OpenTofu, Helm, GitOps)
  • Every change through a pull request with automated checks and your approval
  • Environments promoted in order: development, staging, production
  • Secrets in your secret manager, never in code, chat or email
  • Least-privilege, time-bound access through your identity provider (security practices)
  • Significant decisions recorded as architecture decision records

Definition of done

  • Acceptance criteria in the SOW met and signed off
  • Code and configuration merged through reviewed pull requests in your repository
  • Automated checks passing in CI
  • Runbook updated and rollback tested
  • Monitoring and alerts in place for what changed
  • Documentation and ADRs updated
  • No open severity-1 or severity-2 defects caused by the change
  • Knowledge transfer held with the people who will run it

Communication

You always know where things stand.

A fixed rhythm from week one, in your tools. Replies within one business day; the engineer on the call is the engineer doing the work. Sample status report

TouchpointWhenWhat you get
Async updateDaily, in your Slack or Teams channelWhat moved, what is blocked, what is next
Working sessionAs neededPairing and reviews with your engineers
Status report and callWeeklyStatus, progress, RAID log, decisions needed
DemoWeekly or per milestoneWorking changes shown, not slides
Steering reviewMonthly, or per phase gateSponsor-level progress, risks, scope and outcome measures
Close-outEnd of engagementClose-out report, measures against baseline, open items and recommendations

Hand-over

What you hold at the end

Designed so your team can run, change and audit what we built without us.

Runbooks

Step-by-step operating and recovery procedures, tested during the engagement.

Architecture record

Diagrams and ADRs explaining what was built and why. Sample ADR

Code and pipelines

Terraform, manifests, scripts and CI in your repositories, with history.

Dashboards and alerts

Monitoring for what changed, with alert routes your on-call team owns.

Recorded knowledge transfer

Walk-throughs with your engineers, recorded for people who join later.

Close-out report

What was done, measures against baseline, open items and recommendations.

Questions engineering leaders ask

How we fit into your team, tools and process.

Ask a question

Do you work in our tools or bring your own?

Yours. We work in your ticketing (Jira, Azure DevOps, ServiceNow), your repositories (GitHub, GitLab, Bitbucket), your chat and your change process. Nothing lives only on our side.

Who reviews your code?

Your engineers approve merges. Every change goes through a pull request with automated checks; nothing is changed by hand in production outside an approved runbook step.

What does hypercare cover?

A period after go-live, length set in each SOW, when we watch what changed, tune it and fix issues caused by our work within the agreed fee.

How do you handle a failed change?

We roll back using the rehearsed path when go/no-go or post-change checks fail, record what happened in the RAID log and agree the next attempt with you. A rollback is a planned outcome, not an emergency.

Can our team learn as we go?

Yes. We pair with your engineers, record knowledge-transfer sessions and hand over runbooks written for the people who will run the system.

Do you follow ITIL or SAFe?

We fit the process you have. We file change records for your change advisory board, work in your sprints and report in the format your PMO uses.

Walk through the method with the engineer who would run it.

On the free 30-minute review we look at what you run and how this method would apply to it. We reply within one business day.