Summary
Matt Dailey, CEO and founder of Ref, delivered this conference talk on what he calls “velocity sickness” — the paradoxical stress and diminishing returns that emerge when AI makes individual engineers dramatically faster but the team as a whole fails to benefit proportionally. Drawing on firsthand experience building Ref’s product and observing engineering organizations at scale, Dailey identifies four specific failure modes: PR queues that collapse under the volume of AI-generated code, engineers and agents moving in incompatible directions simultaneously, “agent bankruptcy” (repeatedly restarting agents and re-solving the same problems across sessions), and the most serious issue — agents making critical architectural decisions without meaningful human review, effectively transferring code ownership from engineers to AI.
Dailey’s diagnosis is that teams are applying implementation-optimized tools — ephemeral chat sessions and IDE-centric workflows — to work that is fundamentally about decisions, exploration, and expressing engineering judgment. His proposed solution is a distinct “decision layer” built around persistent, shared documents rather than isolated chat threads, where key choices are captured and visible to the whole team before implementation begins.
The talk concludes with practical steps engineering leaders can apply immediately, including separating planning and implementation into distinct workflow modes and treating agent-session decisions as artifacts to be reviewed rather than ephemeral context to be discarded. This is a substantive, practitioner-level perspective on organizational scaling challenges specific to AI-native development.
📺 Source: AI Engineer · Published August 09, 2026
🏷️ Format: Keynote Launch
