Why your ai shouldn't decide alone: the 3-options pattern

Michel Faure avoided a costly rework by asking his AI agent for three structurally distinct solutions before coding — not just one recommendation [devto]. The task? Updating a trainer's name in an ERP system, a change that看似 simple but touches audit trails, role-based access control, and workflow integrity [github].

The '3-options pattern' forces explicit trade-offs across three axes: business effect, code surface, and operational cost. In this case, the AI proposed options ranging from direct database edits to full workflow-integrated updates. Faure picked the option with strong audit logging — a choice validated when a parent later questioned the name change in their child’s record. Without that traceability, accountability would have collapsed.

This method counters the illusion of speed in solo AI decisions. When an AI offers a single path, it inherits the developer’s context bias and suppresses alternatives. By mandating three options, developers expose hidden assumptions and pressure-test feasibility. It’s not about distrust — it’s about structured skepticism.

The pattern works best where changes leave business traces: audit logs, permissions, approvals. In those cases, Faure’s rule applies: no implementation until three options are compared. One option is a prescription. Three options are a negotiation.