Use AI Where It Improves the Work.
Find recurring tasks where AI could save time, improve quality, or create useful value. Start with a small test and check whether the work improves.
← Back to Tech & AIAI is useful task by task.
AI is not a capability you simply “have.” Its usefulness changes task by task. A strong draft does not establish that the same tool can reliably check a spreadsheet or make a customer commitment.
- Understand and prepare
- Research, information retrieval, summarization, document analysis, and meeting preparation can become easier when sources are available and checked.
- Draft and organize
- Drafting, rewriting, classification, reporting, and customer-support preparation can reduce effort while a person reviews the result.
- Analyze and build
- Spreadsheet analysis, coding, small utilities, and workflow support can be useful. Validate important calculations, test changes, and include review time in the comparison.
Use AI first where mistakes are visible, reversible, and cheap. Increase controls as consequences increase.
Find one to three recurring tasks worth testing.
We agree a small set of workflows to examine. Bring examples, current effort, and what a good result looks like. The review compares possible interventions rather than assuming every task needs AI.
- Leave alone / Improve the process
- Keep a working task as it is, or remove an unnecessary step and clarify who does what.
- Existing feature / Simple automation
- Use a capability you already have, or rules for stable, predictable steps.
- AI assistant / Integration
- Use AI to help a person interpret or prepare material, or connect systems when the handoff is the actual problem.
- Mini app / Bounded agent
- Consider a small tool for proven friction. Consider an agent only when variable, multi-step work justifies AI choosing the next step.
Take away a short opportunity list, the simplest suitable approach for each task, and a test plan: inputs, baseline, review owner, success measure, and stopping condition. Finding no justified AI use case is a valid result.
Count checking and correction as part of the work.
Human judgment owns consequential decisions. Drafting a response is different from sending it, changing a record, or promising a refund. Any implementation needs its own scope, permissions, and testing.
If the job seems to require multiple steps and tool access, first check whether an assistant or ordinary automation would be enough.
Read: Do I need an AI agent?Where does useful work keep taking too much effort?
Bring a recurring task and an example of a good result. We can assess whether AI or a simpler change fits.
A 20-minute conversation by phone or video. Any paid work has a separately agreed scope and price.