TL;DR: You equip development, product and project management teams with coding agents: Claude Code, OpenCode, GitHub Copilot, Codex. You set up the tooling, write the first skills with the developers, run workshops on their own codebase and measure what changes. Paris (Île-de-France), 50% on-site and 50% remote, €45–60k, permanent contract or freelance.
Idun Group is an independent AI consultancy. We measure how AI is actually used inside our clients, from the logs their platforms already produce, then help them industrialise what works. More than 40 AI assessments in under a year across more than 10 large enterprise clients; our references include LVMH, Airbus and L'Oréal. We resell no platform, so we recommend the tool that fits the team, never the one we need to move.
Almost every large engineering team now has access to a coding assistant. Few use it for more than autocomplete. Licences are open, three or four developers push ahead on their own, and the rest of the team is waiting for someone to show them how. The project manager and the Tech Lead know the way of working has to change. They have no time to build the method.
You build it with them. You join a client team, typically 20 to 30 developers with a project manager and a Tech Lead, for several months. You set up an agent-assisted delivery chain: tooling, conventions, shared skills, tests, code review. Then you make sure the team uses it every day and can keep it evolving after you leave.
Workshops are part of the job, and they run on the team's own code, tickets and pipelines. A good session ends with a merged pull request.
The role also covers product and project management: writing specs, slicing the backlog, meeting notes, sprint tracking. The same tools, with Claude, Microsoft 365 Copilot and the Jira or Confluence connectors, save as much time there as in the code, once someone sets the right habits.
We stay tool-agnostic. One client runs GitHub Copilot, another Claude Code, a third OpenCode wired to the models of its choice. You know each well enough to recommend the right one for the context, and to say when a licence does not return what it costs.
- Assess a team: tools in place, actual usage, friction points, security and compliance constraints. - Install and configure the tooling in the client environment: Claude Code, OpenCode, GitHub Copilot or Codex, MCP servers, instruction files (AGENTS.md, CLAUDE.md), hooks, CI integration. - Write the first shared skills and commands with the developers, then organise a catalogue so one team can find and reuse what another built. - Run hands-on workshops: writing and reviewing tests, refactoring, migrations, documentation, debugging, pull request review. - Equip the project manager and product owner: specs, ticket slicing, meeting notes, sprint tracking. - Train one or two internal champions per team so the practice holds without you. - Measure the effect (real adoption, cycle time, delivery quality) and report it to management with numbers.
- Coding agents: Claude Code, OpenCode, GitHub Copilot agent mode, Codex, Cursor - Extending agents: skills, commands, subagents, hooks, MCP servers - Models: Claude, GPT, Gemini, and the models offered by GitHub Copilot or the client's cloud - Project management: Jira, Confluence, Azure DevOps, Microsoft 365 Copilot - Dev environment: Git, GitHub or GitLab, CI/CD, code review
