One Agent Is a Feature. Three Agents Are a Platform.

A strong team can build an AI agent in a week. Three teams doing that in the same product, serving the same users, is a recipe for chaos: fragmented UX, unreachable artifacts, and three agents that can't orchestrate into anything coherent. If you want to avoid this chaos, you must build a platform.

We built one. Three contracts, each unlocking the next, live in production at monday.com:
- A tool protocol: each team owns its MCP server, the platform owns the interface and the registry. The foundation everything else stands on.
- An outcome layer: the platform persists the typed artifacts those tool calls produce, queryable by any surface, from dashboards to our mobile app. Agents are judged by outcomes, not the reasoning behind them, so we can measure and attribute the concrete business value each agent delivers.
- A UI boundary: those outcomes get rendered through predefined microfrontend slots. This "constrained freedom" lets three teams present their artifacts, in chat and anywhere else, as one coherent experience.

The payoff: teams spent their time on agent quality, not harness engineering. We have three (and counting!) independently built agents that feel like one product, and a platform now ready to orchestrate them into one brain.

You'll leave with the three contracts and the tradeoffs behind them, portable to any multi-team agent platform, whatever it's built on.

Engineers and architects who've shipped an agent and are wondering what comes next - this session is for you.