AI-Assisted Development on Axelerant Engineering Handbook

Project Configuration

Mon, 01 Jan 0001 00:00:00 +0000

Project-Level AI Configuration#

Claude Code reads configuration from the project directory. When this configuration is committed to the repository, every team member starts with the same context, permissions, and tool integrations. This is no different from how we treat DDEV configuration or linting rules — it’s part of the project, not a personal preference.

See the official documentation on memory and project configuration for how these files work.

What every project should have#

CLAUDE.md#

A project-level .claude/CLAUDE.md (or CLAUDE.md at the repo root) gives Claude context about the project: the stack, architecture patterns, conventions, and common pitfalls.

Skills and Workflows

Mon, 01 Jan 0001 00:00:00 +0000

Skills and Automated Workflows#

The difference between using Claude Code conversationally and using it effectively is skills. A skill is a structured instruction set that tells Claude exactly how to approach a specific type of work on a project. Instead of explaining your project’s conventions every session, you encode them once and invoke them by name.

For how skills work technically, see the official skills documentation.

Our approach: skills as institutional knowledge#

We treat skills as the place where project-specific knowledge lives. A good skill encodes:

Development Practices

Mon, 01 Jan 0001 00:00:00 +0000

How We Use AI in Development#

AI tools are powerful but they require discipline. These are our expectations for how engineers at Axelerant work with AI.

Human in the loop#

AI augments your capability. It does not replace your judgment. Always review AI-generated code — it’s rare not to find something to fix or improve.

Be aware of cognitive biases that come with AI assistance:

Automation bias — over-relying on AI suggestions because they look confident
Anchoring — difficulty thinking beyond the AI’s initial suggestion
Framing effect — accepting the AI’s phrasing or approach at face value
Sunk cost fallacy — reluctance to discard AI-generated code you’ve already invested time reviewing

If a solution from the AI isn’t quickly helping or you feel you’re losing control, revert the changes and try a different approach. Sometimes writing the code yourself is faster.