An engineer learned to miss his AI usage limits — and how to stop 'snacking' on easy wins
every · x · 2026-09-27
Every highlights engineer Jack Cheng's reflection: after burning through his weekly Fable quota in a day, he realized lighter models made it too easy to throw AI at every idea. Each felt like a small win, but he was doing what Will Larson calls "snacking" — choosing easy, low-impact work over difficult, high-impact work.
His rules to restore deliberate pause:
- Sleep on a feature before asking AI to build it
- Explain to yourself why a hard question deserves an answer
- Write your own tentative answer before consulting the model
A thoughtful first-hand take on preserving engineering judgment in the AI era.
More from AGI Musings
- Economist warns outsourcing writing to AI may be 'cognitively disabling' for PhD students — paulnovosad · 2026-09-28
- Investor predicts saying you made something with AI becomes a faux pas within 6-12 months — StewartalsopIII · 2026-09-28
- AlphaFold database adds 8,000+ virus protein complexes in pandemic preparedness push — AllThingsApx · 2026-09-28
- Could machines become conscious? Calum Chace on machine consciousness — cccalum · 2026-09-28
- Philosopher maps the four problems of consciousness and how brain encoding helps — dioscuri · 2026-09-28
- Researcher predicts powerful computer-use agents will reshape the digital economy in 6-12 months — gordic_aleksa · 2026-09-28