Spec-Driven Development on a Higher Level: Guardrails for AI Agents

Agents now write more code than you can carefully read. The common answer is "read the plans, read the diffs, write a better prompt."

I don't think that's the answer — and I've got the receipts.

A good spec pins intent. It doesn't pin the solution, so you end up reviewing everything anyway — no certainty, just diligence.

This session is the story of how I stopped doing that: adding deterministic guardrails — behavioral (BDD) tests and enforced C4 architecture — on top of the spec, so I could hand an agent a bigger task and read less of the result by hand.

What you'll walk away with:

• Why "a better prompt" is not the fix — and what a deterministic guardrail actually is
• How a spec becomes an enforced contract: visualized architecture + behavioral tests the agent can't quietly violate
• Where this goes next: the tooling, enforcement, and the frontier — ontologies and neurosymbolic AI

Who it's for: Engineering leaders owning AI adoption, and the senior devs / tech leads / EMs doing the work. If you're trying to move faster than review can keep up, this is for you.

Format: ~50 min talk + live demo, then open Q&A.
Remote, recorded.

Wed Sep 9, 2026 · 18:00 CEST (12:00 ET / 09:00 PT), on Google Meet — join link sent when you register.

Save your spot → registration is on Luma (link below), not just Meetup RSVP — that's how you get the recording and the deck afterward: https://luma.com/wlf8j75t

_______
Part of Software Engineering for the AI Era — a recurring workshop series on agentic coding and AI-native delivery.