For more than two decades, we’ve tamed software behaviours with living documentation, giving us fewer surprises and leaner code. Antony Marcano shows how to apply that same discipline to developing non-deterministic AI agents — from customer-service bots to coding assistants. Think BDD for AI: taming agent behaviour one scenario at a time.

Each scenario captures one example of what we want from the agent: setting context, providing a prompt, and evaluating the outcome against a scorecard. As we adjust the prompts and guidance that shape the agent, these scenarios build into a regression suite. This living documentation tells us when behaviour drifts — whether from a prompt tweak or a new model release. We can change the agent’s guidance, or adopt a new model, and the suite catches anything that regresses. Test-Driven Agentic Behaviours takes us from prompt-and-pray to deploying agents with confidence.

Antony Marcano has spent more than three decades developing software, and over a quarter of a century applying XP and related practices. He specialises in building high-performing digital product development organisations — either as a consultant through his company RiverGlide, or in-house, as he did during his time as Head of Engineering at Ford Digital.

His work is referenced or acknowledged in books on Agile Coaching, Software Craftsmanship and Agile Testing. He speaks at international conferences and academic institutions, from Oxford and Cambridge in the UK to McGill University in Canada. He has evolved a test-first approach to developing AI agents, and continues to help organisations ship software, faster than the competition, with (or without) AI.

Join this unique conference that is focused on Living Documentation and Behavior-Driven Development. Click here for more information and registration.