Scale the Judgment, Not the Model — Andrew Orobator, Reddit

An agent needs access to the judgments that experienced teammates otherwise carry in their heads. This engineering talk argues that the model is not the bottleneck and proposes putting that knowledge into skills (instructions an agent can run), work logs and verification steps. A work log, for example, lets a fresh agent continue from a documented milestone.

The mechanism makes expectations and task state available across sessions. It also creates something the team must maintain: the talk warns that encoded judgment rots. The proposed engineering responsibility is therefore to keep the instructions and checks aligned with how the team wants work performed, alongside improving the agent's underlying model.