Why teams need hands-on AI agent calibration before spec sessions

Short summary

Using the Athenian gymnasion as an extended metaphor, this article argues that teams need a structured calibration phase before spec sessions on AI agents. Without hands-on testing against real tasks, team members carry untested assumptions about what an agent can do, leading to specs that are too ambitious or too conservative. The author contends that shared calibration must happen in low-stakes conditions before it counts, and that getting stuck during testing is the mechanism that builds real judgment.

• •Teams enter spec sessions with untested, divergent assumptions about AI agent capabilities
• •A structured calibration phase — hands-on trial against real tasks — builds shared judgment before specs are written
• •Getting stuck during testing is not a cost but the mechanism that produces genuine understanding

Generated with AI, which can make mistakes.