Training track — developers
AI Engineering
Effective use of AI coding tools in a real codebase, with the review discipline that keeps generated code safe to ship. Taught from production work under audit and compliance constraints.
Who it suits
- Software engineers working on established, non-trivial codebases
- Engineering leads who want consistent quality rather than uneven individual adoption
- Teams shipping into regulated or safety-critical environments
Prerequisites
- Working professional experience writing and reviewing code
- Familiarity with the team's own version control and test tooling
- Access to the codebase and coding tools the team uses, ideally on the day
Session-by-session outline
Session 1 — Working with agentic coding tools
Productive use in the terminal and the editor: where each is stronger, how to run and interrupt an agent, and how to keep changes reviewable.
Session 2 — Scoping and specifying tasks
Turning a change request into a task an agent can complete usefully on a large existing codebase, including what to hand over and what to hold back.
Session 3 — Context management
Supplying the right context and managing it as work grows: relevant files, conventions, interfaces, and knowing when to restart rather than continue.
Session 4 — Review and testing discipline
Reviewing generated code properly rather than accepting it: reading diffs critically, writing tests that would actually fail, and gating what reaches main.
Session 5 — Where these tools fail
Invented APIs, plausible but wrong logic, silent breaking changes and confidently incorrect refactors — how each shows up and how it gets caught.
Session 6 — Beyond new features
Applying the tools to testing, refactoring, migrations and documentation, where the return is often larger and the risk easier to contain.
Session 7 — Security, licensing and data
The implications of sending source code to a model provider: confidentiality, licence and provenance questions, and what your obligations require.
Session 8 — Team conventions
Establishing shared conventions so quality holds across the codebase rather than varying by developer.
Typical duration
Two days on-site, or a four-week programme of half-day remote sessions with practice between them.
What participants can do afterwards
- Work productively with agentic coding tools in the terminal and the editor
- Scope and specify tasks so an agent produces useful output on a large existing codebase
- Supply the right context and manage it as work grows
- Review and test generated code properly rather than accepting it
- Recognise where these tools fail — invented APIs, plausible but wrong logic, silent breaking changes
- Apply them to testing, refactoring, migrations and documentation, not just new features
- Understand the security, licensing and data implications of sending source code to a model provider
- Establish team conventions so quality holds across a codebase
Book a conversation
Tell us about your team and we will suggest a shape for the training. Or email usman@bukhariconsulting.co.uk directly.