ChatGPT and Codex training for startups and product teams in Berlin

Codex is already on your developers’ laptops

If your developers installed Codex on their own, each of them runs it differently. One lets it edit files and run commands without asking, another approves each step. One repository has an AGENTS.md, the next has none. The code review sees the result and cannot tell which rules produced it.

Your product team is in the same position with ChatGPT. Each person keeps private prompts for user stories and release notes, and a new colleague starts from zero. The training turns these habits into shared ones, in two tracks: ChatGPT for the product team, Codex for the developers.

The ChatGPT track for product teams

Product managers and support staff learn ChatGPT on the texts they write each week.

• A project per product area that holds the roadmap and the glossary, so each chat starts with that context.
• User stories with acceptance criteria from interview notes, checked against what the user said.
• Release notes in English and German from one changelog.
• A competitor overview in which each claim links to its source.
• Which data may go into ChatGPT, as your data protection officer decides.

The Codex track for developers who already use it

This track starts with the setup your developers have now and turns it into one the team shares.

• The AGENTS.md at the root of the repository, plus files in subdirectories where a part of the code needs its own rules. Codex applies each file to the code below it.
• Approval and sandbox settings: when Codex asks before a command, and whether it may reach the network.
• Checks that must pass before Codex proposes a change: tests, linter, type check and build.
• Review of agent changes. Codex can review pull requests itself; a person still approves the merge.
• Hooks, scripts that check a tool call before it runs. My method is on AI agents under the same rules as humans.

Six Codex settings your team should agree on

Before the training, find out how each developer runs Codex today. The table lists the settings that differ from laptop to laptop and the question your team answers for each.

Setting	Where it lives	Question for the team
Approval policy	Codex configuration or a command-line flag	When must Codex ask before it runs a command?
Sandbox mode	Codex configuration	May Codex write outside the repository or reach the network?
AGENTS.md	Repository root and subdirectories	Who reviews a change to the rules?
Secrets	.env files and the shell	Can the agent read keys for production?
MCP servers	Codex configuration	Which company systems may Codex reach?
Pull request review	Your Git host	Who approves the merge of an agent’s change?

Who I am

I am Michael Wutzke, an IT architect for data platforms based in Frankfurt, with more than twenty years in IT and media. I was Head of Decentralized Finance and Node Operations and then Chief Information Officer at the Frankfurt-based company Blocksize Capital. Today I build software with Codex and Claude for a client and teach at Claude Hacker House. I am also interested in open-source AI models that a company runs on its own servers. I train in English and German. Details: Career stages.