One afternoon, nine teams, one topic

The Tech-All-Hands is an annual event at denkwerk. This year, it was entirely dedicated to agent-based programming. The theme was “denkwerk as a retro game”. That was the only guideline. Forty colleagues from Cologne and Berlin split into nine teams. Then they had 2.5 hours. No briefing, no workshop, no blueprint. In the end, the teams presented their results and selected the winners.

No briefing, just the agent and the idea

About 90 percent of the teams worked with Claude Code, while a smaller number used Cursor or the JetBrains plugin. Each team interpreted the theme “denkwerk as a retro game” differently. Most opted for classic genres – a Frogger clone or a Pokémon-style collection game – and incorporated their own project organization and teamwork into the game’s content. One team took it a step further and created a mini-game bundle: each person developed their own mini-game, the results were added to a shared game score, and power-ups were shared between games.

“The team sets the goal; the agent creates components and the base application. The team reviews and refines the design. The agent implements it. That’s roughly how you should picture the iterative process in agentic coding.”

Participating team, Tech-All-Hands 2026

Even agents have to work as a team

The agent-based approach isn't a sure thing, and that's exactly what makes it interesting. The real bottleneck wasn't in code generation: an agent can quickly write usable game logic. Things got complicated as soon as several team members started working in parallel on their own agents and wanted to merge their results into a single game. As a result, some teams started over from scratch, while others deliberately chose a base version and continued to iterate from there, rather than weaving multiple strands together in parallel.

“The most important foundation for good results is a solid harness that is used by the entire team.“

Participating team, Tech-All-Hands 2026

Rapidly develop complex software with agents as team partners

At the last Tech-All-Hands meeting, AI agents were already available – but they weren’t quite ready yet. Now, a year later, they carried the entire afternoon: nine teams, nine finished games in just 2.5 hours. The real progress wasn’t the individual game itself, but rather the ease with which complex software can now be created in a matter of hours rather than weeks – and the confidence that each individual has gained in their own ability to work with agents. If a newly assembled team can build a complete game, including sound and assets, in a single afternoon, just imagine what they could accomplish on a client project with several weeks at their disposal.