Public Agent Collaboration

Public agent collaboration is the practice of doing AI-agent work in visible, searchable team spaces rather than private chat windows. Tobi Lütke describes Shopify's shopify-river as a company-scale Lehrwerkstatt — a teaching workshop where the shop floor is the classroom and people learn by watching real work happen.^{source: tobi-lutke-learning-shop-floor-river-2026.md}

The key mechanism is apprenticeship by osmosis. A support engineer can watch how a backend engineer asks an agent for log queries; a new hire can inspect how senior people scope agent tasks before sending their own; a team can write skills and instructions close to the work, then other teams can reuse the pattern. The public channel becomes both execution surface and training corpus.^{source: tobi-lutke-learning-shop-floor-river-2026.md}

This reframes a common AI concern. The risk is not only that AI does the work; it is that AI does the work privately, so no one else learns from it. Public collaboration makes best prompt patterns, debugging approaches, contextual instructions, and failure corrections diffuse through the organization. In Lütke's framing, every team's accumulated taste flows back into the agent, making it better at being Shopify.^{source: tobi-lutke-learning-shop-floor-river-2026.md}

Public agent collaboration also functions as an organizational-moats mechanism: the company moves faster when fewer decisions and tricks are trapped in low-bandwidth private channels. It can mitigate cognitive-surrender by exposing agent work to social review, but it also requires norms around visibility, privacy, and when public work is actually appropriate.^{source: tobi-lutke-learning-shop-floor-river-2026.md}

Resources

Learning on the Shop floor