How to think about AI with sound judgment
A simple framework for evaluating AI tools, focused on real capability, risk, and the conditions for adoption.
Generative AI arrived with a difficult mix: real capabilities, impressive demos, anxiety about jobs, overblown marketing, and tools that change every few weeks.
The most useful question is a concrete one: What work can it help us with, within what limits, and under what review conditions?
Three layers for evaluating a tool
1. Real capability
What it does well today, on a task similar to yours.
Examples: summarizing documents, comparing options, writing first drafts, classifying information, generating ideas, writing helper code, or preparing questions.
2. Cost of a mistake
What happens if the tool gets it wrong. A draft email leaves room for review. A critical legal, medical, financial, or operational decision needs a different level of control. AI can take part, within a workflow where human accountability is clear.
3. Conditions for adoption
What a person or a team needs to use it well: access, time, examples, criteria, an internal policy, secure data, review, and permission to learn.
A quick map
Ask these questions before bringing AI into a workflow:
- Which task do we want to improve?
- How is it done today?
- Where does it hurt: time, quality, coordination, cost, or risk?
- Which part can AI assist with?
- Which part must a person review?
- How will we know it improved?
The sign of sound judgment
An organization gains clarity when it can say:
"This is where we use AI. This is where we need human review. This is where we should move more slowly."
That clarity is worth more than the most ambitious presentation.