Level 3 of 3

AI Builder.

Two days for analysts, operators and marketers who want to build small internal tools themselves. Participants plan, build and ship one tool on their own work with a coding agent such as Claude Code, and check it against a security checklist.

Book AI Builder All levels

Duration: 2 days, two weeks apart
Group size: Up to 8 per group
For: Builders outside the IT team
Before joining: AI Practitioner
Tools: Claude Code and similar coding agents
With: The client's IT team informed

Two colleagues reviewing code on a screen

Plan before building. Ship only what passes the checklist.

Builders outside IT, with the guardrails IT would ask for.

Three rules for anyone who builds.

Plan before building.A one-page spec comes first: who uses the tool, what goes in, what comes out, which rules apply.

Keys stay out of code and chats.Secrets live in one place, and a leaked one is rotated the same day.

Owned and documented.Every tool has an owner, a README and a plan for when it breaks.

Two days, module by module.

Day 1 plans and starts the build. Two weeks of building on real data follow. Day 2 makes it safe and ships it.

1
Day 1, 9:00 to 12:30
Plan, then build
A written spec and a first working version.
What it covers
- What a coding agent does, and where it fails
- The spec: users, inputs, outputs, rules
- A project rules file so the agent follows the team's conventions
- Small steps: run, test, read what changed
Practice, on their own cases
Each participant writes the spec for their tool and builds a first version on sample data.
They leave with
A spec and a working first version.
2
Day 1, 13:30 to 17:00
Connect it to data and tools
The tool reads the data it needs, with the least access possible.
What it covers
- Spreadsheets, exports and databases
- APIs and connectors to email, drives and the CRM
- Secrets in environment variables, never in code or chats
- Reusable skills for the steps the tool repeats
Practice, on their own cases
Connect the tool to one real data source.
They leave with
The tool running on real data, with keys stored correctly.
Two weeks of building on real data, with one office hour
3
Day 2, 9:00 to 12:30
Ship it safely
Every tool passes a security checklist before others use it.
What it covers
- What a quickly built app can expose
- Prompt injection from documents, emails and web pages
- Guardrails: what the agent may run, edit and access
- Peer review against the checklist
Practice, on their own cases
Peer review of each tool against the checklist, then fixes.
They leave with
Each tool passes the checklist.
4
Day 2, 13:30 to 17:00
Run, share and hand over
The team uses the tool, and someone else can maintain it.
What it covers
- Versioning in a shared repository
- Sharing with the team: an internal page, a scheduled run or a shared skill
- A one-page README, an owner, and what to do when it breaks
- When a tool moves to the IT team
Practice, on their own cases
Demo to the group and a manager; README written.
They leave with
One internal tool in use, documented, with an owner.

Around the sessions.

Before
A call with each participant to choose a tool worth building, and with IT to agree the rules.
After
One review a month later: what is used, what broke, what moves to IT.
Adaptation
The tool changes with each participant. The checklist stays the same.

Safe use at this level.

API keys and secrets, and how to rotate a leaked one
Least access for every tool and account
Untrusted input: how a document or a web page can steer an AI tool

Safe use across the Academy

AI for Leaders: one-to-one coaching

AI for Leaders

AI Builder.

Three rules for anyone who builds.

Two days, module by module.

What it covers

Practice, on their own cases

They leave with

What it covers

Practice, on their own cases

They leave with

What it covers

Practice, on their own cases

They leave with

What it covers

Practice, on their own cases

They leave with

Around the sessions.

Before

After

Adaptation

Safe use at this level.

AI for Leaders: one-to-one coaching

Start with one team.