'Capability Gaslighting': A New Mental Model for Over-Trusting Frontier Models
Summary
Maggie Appleton (Staff Research Engineer, GitHub Next) coins 'capability gaslighting' — frontier models impress users into over-trusting them before failing the same task later. She's stopped reading agent-generated PR code for prototypes, instead writing detailed upfront specs with explicit verification criteria and trusting the agent's own checks; also flags 'decision fatigue' breaking agent planning when too many sequential choices are presented.
Why this signal
A new, named mental model for the human-AI trust problem — matches Neo's confirmed-high-signal calibration for novel framings (like 'cognitive surrender') that are useful for team conversations even without a specific technical fix.
Original email
The Pragmatic Engineer · 9/23/2026
Design Engineering with Maggie Appleton