The AI assistant job matrix
The job matrix: writing wants structure, code wants instruction-following, research wants citations, analysis wants long-context recall, customer work wants cheap reliable latency.
Which assistant for which job
- Long-form writing - a strong writer with long context. You trade speed for prose quality.
- Code - an instruction-following model with a big context window and a fast iteration loop.
- Research - a model with live search plus citations. No citations, no trust.
- Data analysis - a model that can run code against your files instead of narrating them.
- Customer chat - a small fast model with tight system prompts and a kill switch.
No single assistant wins all five. The people who get value run two: one for thinking, one for speed.