Twenty-five engineers, one batch, one working build, inside their own environment.

The opportunity

A technology team that wanted to build with AI inside its own security perimeter.

The insurer’s technology team was ready to build with AI rather than read about it. The constraint was where that learning could happen. Regulated data, approved tooling and internal controls meant a programme run on a sample project outside the perimeter would produce work nobody could use.

The client is under NDA, so the system itself is described here only in outline. Three things shaped the programme:

• Everything inside the perimeter

  The batch ran in the insurer’s own environment, with its own tools and its own controls, so the work never had to cross a boundary to be useful.

• Nothing to rebuild afterwards

  Work produced outside an approved environment has to be rebuilt before it can go near production. Building inside removed that step.

• Engineers, not an audience

  Twenty-five people who write code every day, learning on their own stack rather than on a tutorial.

The impact

One batch, twenty-five engineers, and something that ran at the end of it.

The batch followed the same six steps every programme follows: agree who joins, map the real work, hands-on sessions on that work, a graded assessment, a one-day hackathon, then champions named and support continuing afterwards.

It started something new rather than changing a system already in service, and it finished with a working build.

• 25

  Engineers through the batch

  One batch, July 2026, drawn from the insurer’s own technology team.

• 1

  Working build at the end

  The programme finished with something that ran inside the insurer’s environment, rather than a deck describing what could be built.

• 6

  Steps, the same as every programme

  The method does not change between a business cohort and an engineering batch. Only the work does.

• What the batch built is not published

  The client is under NDA. A reference is available on request, and the build can be described on a call.

The setting

What it ran on.

Environment

The client’s own, inside its security perimeter

Tools

The AI tools the client had approved

Work

The team’s own stack and its own repositories

Data

Sanitised or sample data only; no customer or employee personal data in any prompt

Review

A person reviewed and approved any AI output that left the team

The road ahead

The champions carry it forward.

The people graded highest in the batch hold the playbook and the templates, and office hours run for three months after the last session.

1. 1

   Extend the method to the next engineering teams

2. 2

   A leadership session on where AI fails and what that means for a regulated business

3. 3

   Champions running the next batch, with our support in the room