Velocity Sickness

Ref CEO and founder Matt Dailey calls the gap between higher agent output and unchanged team impact “velocity sickness.” He argues for moving the hardest review earlier, keeping consequential decisions with people, and using durable shared plans as agent state.

This is a founder presenting both a workflow argument and the problem his product addresses. The talk offers useful operating guidance, but it does not measure a 10x productivity gain or compare shared documents with other coordination approaches.

Key Points Covered

• More output can overwhelm the team around it: Dailey connects coding-agent adoption to unmergeable pull-request volume, merge conflicts, and fragmented work. He also connects it to “agent bankruptcy,” where developers abandon sessions they can no longer reconstruct [00:01:00]-[00:03:35].
• The central risk is losing decision ownership: His strongest concern is not wasted tokens but engineers allowing agents to make consequential product and code decisions they no longer understand or own [00:02:41]-[00:04:26].
• Impact, not generated volume, is the useful constraint: A newsletter writer producing a book's worth of material each week illustrates his broader point. Output that exceeds its audience's capacity can feel productive without creating value [00:04:26]-[00:06:29].
• Agent work shifts the human role toward planning and polish: Dailey contrasts implementation-centered IDE work with a workflow where people explore the system and make decisions. An agent then implements before people inspect and refine the result [00:06:29]-[00:09:11].
• Chats hide decisions that teams need to share: Dailey argues that implementation-oriented chats are isolated and ephemeral. By contrast, a durable document can expose the key technical decisions for discussion before code exists [00:10:01]-[00:12:52].
• Separate action from state: In his proposed model, the shared document holds project context and decisions while agents perform actions from that state. New agents can start from the same source, and people can rebuild their own context by rereading it [00:12:52]-[00:14:30].
• Move review before implementation: Dailey says exploring several plans and deliberately not building some turns code velocity into idea velocity. He says early agreement on important decisions can reduce later code-review ambiguity. It can also prevent parallel work from diverging [00:14:30]-[00:17:10].
• Keep a durable decision log, then share it: Dailey closes with three practices. Distinguish planning from polishing, and treat the plan as a view into the software system. Get a teammate's feedback before giving the plan to an agent [00:17:10]-[00:19:40].

Full video: https://www.youtube.com/watch?v=Kz4QJmNrVXU(opens in a new tab)