Lesson 06 · 10 minutes
Choose the right kind of AI
Match the capability and the effort to the task, rather than assuming the newest or the deepest option is automatically the right one.
Why this matters
Fast, deep-reasoning, research, image, voice and coding tools trade cost, speed and capability against each other in different ways.
The product wrapped around a model often matters as much as the model itself.
The practical bit
Use fast, general help for low-stakes drafting and tidying. Reach for a deeper reasoning mode only when a problem genuinely needs working through step by step, and the extra wait actually buys you a better answer. Use a research mode when current, sourced information is the point, since these return the address and title of each page they leaned on so you can open and check them.
Specialist image, voice and coding tools tend to fit their medium better than a general chat does. Voice is handy for capturing ideas on the move. For a software workflow you will run again and again, look at an API only once you understand the process, the checks, the volume and the economics.
What it looks like
A quick internal rewrite does not need the slowest, most expensive reasoning mode. A weighty decision memo might justify one, along with multiple sources and a proper review. The deepest modes can also sit behind a top-tier plan and are sometimes labelled experimental, so do not build a routine around one until you know you can rely on having it.
Dictating notes by voice works well while you are walking. Comparing this month's travel rules and prices is a job for a research tool that can cite where each answer came from.
Picking a tool off a leaderboard before you have defined the task, the product you will use, the data you can share, the budget and how you will check the result.
Use your judgement
Try it now
- Describe the output, inputs and consequence of error.
- Choose the most important constraint: speed, depth, freshness, privacy, cost or integration.
- Select one capability to test and name a simpler alternative.
Use this in a real situation
Open the Tools guideCompare current products by task fit rather than provider hype.Open the practical →Keep this
Use enough capability for the task, not the most that exists.