Scope down to scale up: Lessons from agentic AI adoption

Getting everyone to use AI was the “easy” part. The real lesson came after we pushed toward full agentic workflows and it fell apart on three fronts at once, unreliable output, eroded trust, and blast radius too complex to debug with everyone in the company wants to ship code to production.

This talk is about pulling back, rebuilding trust and doing less in order to go bigger. A journey of how a 100 large engineering team adopt AI in our product development. And it is not all unicorns and butterflies

Who is this talk for?

Engineering leaders and senior engineers who've gotten past the "convince the team to use AI" stage and are now deciding what to build next. Especially useful for anyone currently pushing, or being pushed toward, autonomous or agentic workflows before the basics are solid.

What will attendees take away?

• Why full agentic workflows broke for us on three fronts at once. Output quality, trust, and blast radius all failed together, and we'll unpack why that combination is common, not bad luck, and what it means for how you should be testing your own rollout.

• Getting people to try something new is one kind of leadership. Standing in front of the same team and saying the big bet didn't work, here's the smaller thing we're doing instead, is a different one, and it's the one nobody prepares you for. We'll talk about what that actually looked like.

• We'll walk through exactly how we cut the unit of work down, why that specific cut worked when others didn't, and what it now lets us attempt that we couldn't before. The scoping change that got us from stuck to actually shipping."