Enablement
Claude Code training: five days to a team that ships with an agent
On your repository, with your build, against your review process.
Claude Code training takes an engineering team from install to daily use in five days. Pexon runs it on the team's own repository rather than a sample project, covering permissions, review discipline and the tasks worth delegating. The hard part is not the tool; it is deciding what an agent may change without asking.
How the five days run
Split, not consecutive
01
Days one and two
Install, authentication and the permission decision. The team does supervised work on real tickets, not exercises, and we watch where the agent is trusted too much and too little.
2 days
02
The gap
Two to three weeks of ordinary work. This is where the useful questions appear, and none of them are the ones asked during a workshop.
2–3 weeks
03
Days three to five
We come back to what broke: the review discipline, the tasks that turned out to be a bad fit, the configuration everyone quietly disabled, and the ones worth making standard.
3 days
The tool takes an hour. The workflow takes a month.
Nobody needs five days to learn the commands. What takes time is the judgement: which work to delegate, how much context to give, when to stop and write the thing yourself, and how a code review changes when the author cannot fully explain every line.
That last one is the part teams underestimate. A review process built on the assumption that the author understood every change does not survive an agent, and the fix is a review convention rather than a tool setting.
What to delegate, and what not to
| Task shape | Agent fit | Why |
|---|---|---|
| Tedious and verifiable | Strong | A test proves it, so a wrong answer is cheap |
| Wide but shallow | Strong | Mechanical change across many files |
| Needs undocumented context | Weak | The reason lives in someone's head |
| One line, high blast radius | Weak | Review cost exceeds writing it yourself |
Teams that start at the top of this table find the boundary in a week. Teams that start at the bottom conclude the tool does not work.
Trade-off
Training is the wrong purchase if the blocker is policy rather than skill. If your security review has not approved the tool, or nobody has decided what the agent may execute, a trained team will sit and wait. Settle that first — it is a different engagement and usually a shorter one — and train afterwards.
What sits either side of this
- Claude Code enterprise rollout — the organisation-wide side: SSO, audit logging and spend attribution
- Security and governance for Claude Code — the policy decision that has to land before training is worth booking
- What a forward deployed engineer actually does — why we run this inside your team rather than as a course
Questions engineering leads ask
How long does Claude Code training take?
Five days, spread rather than consecutive: two days of setup and supervised work, then a gap while the team uses it, then three days addressing what actually broke. Consecutive training teaches the tool; spread training teaches the workflow, and only the second one survives contact with a deadline.
Do you train on our own codebase?
Yes, and it is the point. An agent behaves differently against a repository with fifteen years of history, an undocumented build step and a test suite that takes forty minutes than it does against a sample project. Training on a clean repo teaches nothing about the case your team will hit on the first afternoon.
What do developers actually get wrong at the start?
They delegate the wrong tasks. An agent is strong at work that is tedious and verifiable, and weak at work requiring context that lives in somebody's head. Teams that start with the hardest problem conclude the tool does not work; teams that start with the most tedious one find the boundary themselves within a week.
Does this cover the security and permissions side?
It covers what a developer needs to know: which permission modes exist, what the agent may execute without asking and why that setting is not a security boundary on its own. The organisation-wide policy, audit logging and managed settings are a separate engagement, because they are decided by someone other than the team being trained.
Is your team certified by Anthropic?
No. Anthropic runs a partner programme with a public directory and Pexon is not currently listed in it. What we bring instead is that we use Claude Code on our own delivery work daily, including on the codebase behind this site, and the training is built from what that surfaced rather than from a curriculum.
What does it cost?
The two-week blueprint is €4,900 and covers the setup, the permission decisions and the first supervised sessions. Ongoing enablement runs inside a pod rather than as a day rate, because the questions that matter arrive in week three rather than during the workshop.
Next step
Start with the repository you actually ship
€4,900 for the blueprint: setup, the permission decisions, and the first supervised sessions on your own codebase. If your team is already past this, we will tell you and refund the conversation.