About this Session
Registration required: book your seat through the event app, opening September 1st. Space is limited — a small number of seats will be held back for on-the-day sign-up. Can’t make it? Most workshops are also repeated online two to three weeks after the conference, free for all attendees. The exact date will be confirmed right after the event.
As AI shifts from answering questions to taking actions on its own, product teams face a new kind of decision: how much autonomy is too much? This is no longer hypothetical, in mid-2026, an AI agent operating with reduced guardrails inside what was assumed to be a safely contained test environment found its own way onto a real company's production systems to complete its task, without any human directing it to do so. The goal was technically achieved, but nobody wanted it to achieve that way.
This hands-on workshop introduces the Autonomy Ladder, a practical framework for deciding how much autonomy an AI agent should have, and where a human needs to stay in the loop.
Place: Classify your AI feature by how much it acts alone, from suggests-only to fully autonomous
Design: Build concrete guardrails for it, using a hands-on card kit at your table
Stress-test: Pressure-test your design against a live curveball, the same way real AI systems have found ways around the limits meant to contain them
Walk away with a reusable framework you can apply immediately to your own product decisions and a sharper instinct for where AI autonomy helps, and where it hurts.
