Twenty years ago, Michael Feathers defined legacy code as code without tests. That definition is no longer enough. The version that matters now is code without recoverable intent, and by that measure, most of what any organisation has ever shipped qualifies, because the reasoning behind it lives in old meetings and departed colleagues’ heads where no agent can reach it.

In this talk, recorded at CTO Craft Con: Europe 2026, Simon Bostock, Managing Director Benelux at Equal Experts, shares a working approach to making tribal knowledge machine-readable as a byproduct of building, not as a separate project bolted on afterwards.

Access the talk recording

Fill in the form to receive instant access to the full session recording.

What You'll Learn

• Why the definition of legacy has shifted from “code without tests” to “code without recoverable intent”, and what that means for every system your teams have ever shipped
• What AI agents actually need to maintain software, since, unlike a new team member, they can’t ask around
• The anatomy of a working semantic layer for software delivery, built from real meetings, specifications with unknowns flagged rather than guessed, and stories traceable back to the decisions behind them
• How drift detection, built to catch where code and intent pull apart, doubles as a rescue path for the legacy systems you already own
• A practical way to pilot this approach, starting with a single team rather than a full re-platform

Who Should Watch?

• Technology, engineering, and delivery leaders working with AI agents across their SDLC
• Teams responsible for legacy systems or long-running decommissioning projects
• Organisations exploring how to make existing knowledge usable by AI, not just new work
• Leaders asking what governance and platform thinking look like at agent speed