Claude Code and Codex training for teams

Most teams that have Claude Code or Codex are using a fraction of it. A few people prompt it like a search box, a few have figured out something clever and kept it to themselves, and the company has no shared way of working.

That is the expensive version. Partial adoption teaches the company nothing, and every person rediscovering the same tricks is time you pay for twice.

What changes when a team is trained

The job moves from writing the code to directing and verifying the work. People who were producing output become people who decide what good looks like and check that the agent got there. That is a different skill, and it does not show up on its own.

A trained team has:

• Written instructions the agent loads every session, so nobody explains the codebase, the conventions or the boundaries again.
• A shared library of skills for the work you repeat, built once and used by everyone.
• Delegation that works. Long tasks broken into briefs that subagents can execute without wandering, while the main session stays on the hard calls.
• Tools connected through MCP, so the agent can reach your systems instead of you pasting things into a chat.
• Hooks and guardrails for the things that must always or never happen.
• A review habit. Checking the actual change, not the agent's summary of it. An agent asked to grade its own work will praise it.

What we cover

The training is built around your own repositories and the work your team actually does.

1. Setup that sticks. Project instructions, permissions, and the conventions your agents should follow.
2. Skills. Turning the work you repeat into reusable skills. I wrote up how I build them in How to build Claude Code skills that actually work.
3. Subagents and delegation. When to hand work off, how to write a brief a model can execute, and how to keep the expensive model on judgment instead of supervision.
4. MCP. Connecting the agent to your systems, and the access questions that come with it.
5. Hooks and safety. Automating the checks that must always run, and knowing what a sandbox does and does not protect.
6. Reviewing agent work. Two passes: does it meet the specification, and is it any good.
7. Using more than one model. When Claude Code is the right tool, when Codex is, and how to use both. See Codex vs Claude Code.

Why me

I run Claude Code and Codex every day, on my own products and my clients'. I build the tooling around them in public. Codex on Crack coordinates several models from one conversation, and in one long build splitting the work that way cut my usage of the expensive model by roughly 95% without losing quality. I also made a free Claude Code course for people learning on their own.

How it works

Training is scoped like every engagement I take: a fixed scope, written success criteria, and a sponsor on your side. It usually sits inside a broader AI implementation, because teaching people the tool matters less than changing how the work moves. But it can stand on its own for an engineering team that is ready.

Tell me about your team: how many people, what you build, and what you use today.