Double what your engineers ship with AI agents.

GitHub Copilot made its name with autocomplete, suggesting the next few lines as you type. In agentic coding you describe a goal, and an agent plans the approach, reads your codebase, writes code across several files, runs the tests and iterates on errors. Claude Code works this way, and so do the agent modes now built into Copilot, Cursor and Codex. For a team, that means scale, because one engineer can direct several pieces of work at once.

Doubling is the target the training is built around, and how close a team gets depends on its backlog. The gain comes from work an agent can carry end to end: well-specified features, tests, migrations, first-pass reviews and bug reproduction. A team whose backlog is full of that work gets closest; a team that mostly does research or architecture gains less. Tell me about your backlog and we can work out where your team sits.

Prompt engineering is the smallest part of it. The topics that matter now are agentic workflows with tools like Claude Code, running work in parallel with subagents, MCP integrations that connect agents to internal tools, AI-assisted code review, and finding your way round unfamiliar codebases. It should also cover how review and testing keep up once agents write more of the code, and run on the team's own codebase, so participants apply it the next morning.

Look for a trainer who ships production software with these tools every day. Claude Code is my main development environment, for client work and for my own products such as Tymeslot, an open-source scheduling platform. The exercises come from that work, including where agents fail and how to recover. Cegos Integrata, NobleProg and Code First Girls book me for exactly that.

Yes. I deliver every workshop in German (native) or English (fluent), and in bilingual teams participants can ask questions in either language. The exercises run on your team's own codebase and tools, so each one is a task from your team's own week. I run it remotely, on site at your offices, or as a mix of both.

I am happy to sign an NDA. Claude Code and similar agents send prompts and the code they read to the model provider, under the data terms of your plan. If code has to stay in your own cloud or under stricter terms, we can run Claude Code through Amazon Bedrock or Google Vertex AI, or with zero data retention. For code that cannot be used in the training at all, I bring a comparable practice repository.

Each participant needs working access on the day to the tool the course uses. It is worth arranging this a week ahead.

Between 5 and 30, depending on the course. Above about fifteen, I pair people up for the practical parts.

It depends on several things: the course, the size of the team, how long it runs, on site or remote, and how much preparation your codebase and tools need. Tell me about your team and we'll work out an offer that makes sense for both sides. You get the price in writing before anything is booked.

Your team keeps everything we set up during the training, including instruction files such as CLAUDE.md, agent settings and prompt libraries. Once the team has worked with the agents for a while, you can book follow-up sessions.