AI Copilot
An AI copilot is a model-powered assistant embedded in a workflow—code editor, email client, design tool—that suggests next actions while a human retains control. GitHub Copilot, which autocompletes code in VS Code, is the canonical example. The copilot pattern works because it keeps the human in the loop: the AI proposes, the human disposes. This is a deliberate design choice that sidesteps the hardest problems in AI deployment (accountability, error correction, trust) by never removing the operator. The limitation is also the point. Copilots augment speed on tasks the user already understands. They are less useful when the user cannot evaluate the suggestion—a junior developer accepting copilot code they do not understand is not augmented, they are automated without knowing it.
Related terms:
Fuzzy Interface
A fuzzy interface is AI’s adaptive translation layer between rigid organizational systems and human intent, interpreting context and adapting to various...
Gravity Wells
Gravity wells describe economic dynamics where scarce resources flow disproportionately to entities with the greatest ability to pay and deploy, creating...
RLHF
Reinforcement Learning from Human Feedback (RLHF) trains a reward model on human preference comparisons and uses reinforcement learning to align language...