AI Engineer World's Fair 2026 — unofficial import demo

Learn more about Ref: https://ref.tools/ AI made writing code nearly free, and on most teams, that's quietly breaking how the team works. Individually, everyone feels ten times faster. Together, the signals point the other way: too many PRs moving in too many directions, engineers throwing away whole agent sessions and starting over ("declaring agent bankruptcy"), and critical decisions getting made inside agent chats that no one will ever see or review. There's a lot of energy, and it's all going somewhere different. I call this velocity sickness: the organizational pain that comes from individual speed. It's the engineering version of an author who ships a book a week: prolific, productive, and completely unreadable by the team that's supposed to build on it. Almost every conversation about AI coding is about making one engineer faster. This talk is about what happens to the team when all of them are. Once implementation stops being the bottleneck, the hard part isn't writing the code. It's tracking it, reviewing it, and keeping a hundred parallel decisions coherent. That's the problem eng leaders are actually being handed, and it's the one this session takes on directly. Engineering has always had three phases: plan, implement, polish. AI collapsed the middle one to almost nothing, so the leverage, and the real work, move to the decision-heavy ends. The fix isn't better prompts; it's changing what our tools treat as first-class. We have to split the decision layer from the implementation layer: humans spend their time at the decision layer, reviewing and making the choices that matter, while agents handle the implementation. That means durable, reviewable plans, not ephemeral chats. Review the decisions before you review the diff. What attendees will leave with: - A mental model for plan / implement / polish and why the decision layer is now where engineering leverage lives, plus the language to explain velocity sickness to their own team. - A concrete shift: how to pull your team's important decisions out of throwaway agent chats and into a shared, reviewable source of truth, so individual speed compounds into team cohesion instead of chaos.