AI Platforms & InfrastructureAI Platform Partnerships & Professional Services
Claude Code Rollout Programme
An engineering team of up to twenty people using Claude Code every day on their own repositories, with written standards for what it may touch, a review gate every PR passes through, and one internal MCP server so Claude Code can read their tickets and specs instead of being told about them.
Claude Code Rollout Programme
$30K–$50K
- Duration: 6 weeks
- Tier: Transform
- Format: Cohort / programme
You leave withAn engineering org productively using Claude Code, with standards and internal MCP tooling
Book a Free SessionSound familiar?
Every answer arrived after the week it was about. Six portals, six exports, and a spreadsheet to make them agree. 'Why did Ludhiana drop last week, were we out of stock, unlisted, or just outspent?' was four logins and an afternoon, and by then the sale was gone; a stock-out cannot be back-dated. Every new question from a brand manager turned into a ticket for an engineer, and the engineers were the queue.
We had the AI seats. What we did not have was a way for the model to reach our data, or any rule about what our engineers could let it touch. So the three people who got it ran ahead and everyone else waited to be told it was allowed. We did not need one more report. We needed the data to describe itself so the assistant could answer, and engineers who could ship that safely, with a standard and a gate, instead of one hero at a time.
What you leave with
CLAUDE.md per pilot repository
Format: Merged in the client's repository
How we know it’s acceptableWritten with the team; a new joiner can set up and open a first PR from it alone
Written standard
Format: Markdown in the client's engineering handbook
How we know it’s acceptableSigned off by the engineering lead; the gate enforces it
Review gate
Format: Installed in the client's repositories
How we know it’s acceptableRuns on every PR for two weeks; first-pass rate recorded and above half after tuning
Internal MCP server
Format: Client repository, deployed with authentication
How we know it’s acceptableA cohort engineer reads a real ticket through Claude Code and opens the PR without us
Facilitator guide and participant materials
Format: Client's shared drive
How we know it’s acceptableA champion ran the week-five session from it without us in the room
Assessment results
Format: One sheet per engineer
How we know it’s acceptableScored against the rubric agreed in week one
Measurement report and ninety-day plan
Format: PDF and the data sheet
How we know it’s acceptableThe five numbers taken by the client's platform engineer, not by us
Also included
- A cohort of up to twenty engineers trained over six weeks on two of their own repositories, not a sample project.
How it runs
6 weeks
Week 1
Baseline and set-up
Take the five numbers from Git and PR history. Confirm seats, the security route and which repositories Claude Code may touch. Name the cohort and fix the session dates against the release calendar.
OutBaseline numbers, cohort list, repository access, the assessment rubric agreed.
Week 2
Foundations
Two sessions and one exercise on their repositories: daily workflow, context, permission modes, what to put in CLAUDE.md.
OutA CLAUDE.md in each pilot repository and the first ten PRs opened with Claude Code.
Week 3
Standards and the gate
Write the standard with the team, install the review gate, tune it on their real PRs.
OutThe standard signed off by the engineering lead, the gate running on every PR, first-pass rate recorded.
Week 4
Internal MCP tooling
Build one MCP server over the agreed internal system to the Hector tool standard, with authentication, and connect it to the cohort's Claude Code.
OutA cohort engineer reads a real ticket through Claude Code and opens the PR.
Week 5
Champions
Three or four engineers run the week's session themselves from the facilitator guide. Assessment runs.
OutCertified champions, assessment results per engineer.
Week 6
Measure and hand over
Re-take the five numbers, write the report and the ninety-day plan, hand the playbook to the named owner.
OutBefore and after figures, the report, the owner.
Is this right for you?
This is for you if
- Buyerthe person who owns engineering delivery and is being asked by the board whether the AI seats are paying for themselves. Usually a VP Engineering, a CTO or a head of platform. They have bought or are about to buy Claude Code seats and have three people using them well, ten who tried once, and the rest waiting to be told it is allowed.
- Also has to say yessecurity, because source code goes to a model API and they decide whether that is direct, through Bedrock or through Vertex. The engineering managers whose teams lose four hours a week for six weeks; if they do not agree in week one, attendance collapses in week three. Procurement or Finance for the seats themselves, which are separate from our price.
This is the wrong answer if
- Fewer than eight engineers. A cohort does not form. Run the free Possibilities Session and give them a written guide instead.
- No CI and no tests on the repositories they want to use. Claude Code makes it faster to write code; without a gate it makes it faster to ship bugs. The SDLC AI Maturity Audit comes first.
- Security has not cleared source code going to a model API, and will not clear Bedrock or Vertex either. Nothing in this programme can start. Say so on the first call.
- They have not chosen Claude Code. If the real question is Claude Code against Copilot or Cursor, that is a comparison, not a rollout, and it belongs in the SDLC AI Maturity Audit. We only run this for teams that have decided.
- Leadership wants a tool installed but no change in how work is reviewed. Adoption stalls at the first PR that gets rejected the old way.
What’s included, and what isn’t
In scope
- one cohort of up to twenty engineers
- two pilot repositories
- CLAUDE.md and the written standard
- the review gate installed and tuned
- one internal MCP server with authentication
- champion training, facilitator guide and materials
- assessment
- week-one and week-six measurement with a report
Out of scope
- Claude Code seats and licences, which the client buys from Anthropic or through the partnership and are not in our price.
- A second cohort.
- Rebuilding their CI or writing the tests they do not have.
- Product features for their customers; that is the Claude-Powered Product Launch.
- A second MCP server or one over a system with no API.
- Support after week six; that is Managed Services or an embedded engineer.
Before we start
Before day one
- Seats provisioned for every engineer in the cohort.
- Security sign-off in writing on the route and on which repositories Claude Code may read and change.
- Two pilot repositories with CI that runs and passes today.
- The cohort named, with each engineering manager's agreement to four hours a week for six weeks, and the session dates checked against the release calendar.
- Read access to Git and PR history for the baseline.
- The internal system for the MCP server chosen, with an API and a service account.
What we’ll need from your team
- An engineering manager as sponsor, two hours a week, to unblock attendance and sign off the standard.
- Each engineer in the cohort, four hours a week: two sessions and one exercise on their own repository.
- One platform or DevOps engineer, three hours a week, for seats, CI, the gate and hosting the MCP server.
- A security contact, about three hours in total, in week one and again in week four.
- Three or four champions, two extra hours a week from week five.
Pricing
What we need to give you a quote
- Headcount in the cohort.
- The two repositories, and whether CI with tests runs on them.
- Whether seats are already provisioned.
- The security route: direct, Bedrock or Vertex.
- Which internal system the MCP server should reach, and whether it has an API.
- Access to Git and PR history for the baseline.
After we hand over
Who owns it after we leave
A named engineering manager owns the standard and the gate. The platform engineer who hosted the MCP server owns it. The champions run a monthly refresh session for new joiners. At first that is about two hours a week for the manager and half a day a week for the platform engineer; we size it against what we measure in week six and write it in the report. If the manager cannot be named by week five, it goes down as a dependency and the Managed Services conversation opens.
What we measure
- PR lead time
- Review turnaround
- Engineers active on Claude Code each week
- PRs passing the gate first time
- Time from ticket to PR on the exercise tasks
All five taken in week one and week six by their platform engineer from their own Git and PR history.
What comes next
- The Cohort AI-Native Engineering Programmefor the whole way of working, once Claude Code is in daily use.
- An Embedded AI Engineerto carry the rollout to the next fifty engineers with the champions.
- Managed Services, Essentialfor the MCP server and the gate.
- A second cohort at a lower price, since the materials and the CLAUDE.md exist.
Questions buyers ask us
You will get the three people who were going to use it anyway. The other seventeen need the standard, the gate and someone sitting with them on their own codebase for six weeks. That is what we are selling.
Then this is not the right programme and we will say so. We run this for teams that have chosen Claude Code. If you are still choosing, the SDLC audit compares them.
A fair one, and for some clients it ends the conversation. Bedrock and Vertex keep it inside your cloud account. If neither is acceptable to your security team, we stop before day one rather than in week three.
Yes. That is exactly the week-three problem and it is why the gate exists. Review turnaround is one of the five numbers we measure, and the gate is tuned on your real PRs until it earns its keep.
Five numbers taken in week one and again in week six, measured by your own platform engineer, not by us: PR lead time, review turnaround, engineers active on Claude Code each week, PRs passing the gate first time, and time from ticket to PR on the exercise tasks.