Claude Code for developers.

Download the deck.

All 88 slides as delivered, including the appendix on model choice, the bonus prompts and the further reading. Use them, remix them, hand them to a colleague who didn’t make it.

Four loops.

Each loop ran through the same four steps: input from the expert, an hour or more of building, demos to the room, and a discussion of what worked and what didn’t.

1. Loop 01

   Mods and MCP Apps

   09:15–10:30 · Peder Holdgaard Pedersen

   One MCP server, two surfaces. Mods change how Claude Code behaves and looks in the terminal; MCP Apps answer with an interactive interface in claude.ai and Claude Desktop. Building it once for both shows where they overlap and where they differ.

2. Loop 02

   Claude + Jev decision models

   10:45–12:00 · Anders Grandjean-Thomsen

   Small, fast models make the routing decisions, so each task goes to the model that fits it. Model fit, cost and latency as an engineering choice rather than a default.

3. Loop 03

   Building a software factory

   13:00–14:30 · Anders Grandjean-Thomsen

   Map how your organisation works, spec the work, and let agents build it in parallel, with people in the loop where it matters.

4. Loop 04

   Putting it together

   14:45–16:00 · Jacob Langvad Nilsson

   Loops, the new SDLC and where it is heading. Then the challenge: ship your first Claude Managed Agent, or close a loop from earlier in the day in your own repo, and demo it to the room.

Experts and hosts.

• Peder Holdgaard Pedersen

  Head of AI Engineering, Saxo Bank · MCP maintainer · Microsoft .NET MVP

  Led Loop 1.

• Anders Grandjean-Thomsen

  Agentic engineering consultant, Consid

  Led Loops 2 and 3.

• Jacob Langvad Nilsson

  Co-founder, Applied Futures · Claude Community Ambassador

  Host. Led Loop 4.

• Camilla Roesgaard Hansen

  Co-founder, Applied Futures · Programme lead

  Host.

You mostly don’t need anything fancy.

Recent models follow instructions well. Most loops start as a paragraph you type. The four from the appendix, ready to paste.

• > Do three more loops of adversarial review and improvement before you stop working.
• > Hill-climb on the aesthetics of my home page. Do 4 iterations. Use 5 held-out reviewers per iteration: 3 from the perspectives of specific relevant famous designers and 2 from the perspective of basic, university-level design textbook principles.
• > Babysit this pull request until CI fully passes and none of the review agents have any more feedback.
• > Save the specifics of this process in a skill so that we can reuse it and improve on it in the future.

Further reading.

The sources behind Loop 4, for after the day.

• Getting started with loopsThe four kinds of loops, and the system around them.
• Building effective human-agent teamsHow Anthropic runs teams of humans and agents. The source behind Loop 4.
• Launch your agentThe starter repo for challenge option A.