Start With The Job, Not The Model
The central habit is simple: do not open the strongest model by default. First ask what kind of thinking the task needs. A stronger model is worth it when the answer depends on judgment: choosing a strategy, spotting risks, weighing tradeoffs, planning a complicated week, writing a sensitive email, or turning messy notes into a clear argument.
A cheaper or faster model is usually enough when the job is routine and easy to inspect. Ask it to shorten a paragraph, make a grocery list from a recipe, clean up grammar, draft three subject lines, or turn meeting notes into bullets. If the result is slightly bland, you can fix it quickly. Paying for the strongest model on that kind of work is like hiring a senior advisor to alphabetize a folder.
Use The Strong Model For The First Hard Turn
Once you know the task needs judgment, spend the stronger model where it matters most: the first hard turn. That means asking it to understand the situation, choose an approach, and define what a good result should look like. For example: “I need to ask my landlord for a repair without sounding hostile. Here is the situation. What tone should I use, what facts matter, and what should the email include?”
This is different from asking, “Write the email.” The better first prompt asks the model to plan before writing. You can then reuse that plan with a cheaper model: “Using this plan, draft the email in 180 words.” This keeps the expensive thinking focused on judgment and lets the cheaper model handle the typing.
Hand Off Clear Tasks To Cheaper Models
After the stronger model has made the important choices, the rest of the work is often execution. Execution means carrying out a clear instruction: rewrite this more warmly, make this shorter, create a checklist, produce five variations, or format this into a table. Cheaper models are good at these jobs because the path is already set.
The failure mode is handing off too early. If your instructions are vague, the cheaper model may produce generic work or miss the point. Fix that by carrying over the strong model’s best context: the goal, audience, tone, must-include details, and any “do not do this” rules. Instead of “Make this better,” write: “Rewrite this for a busy manager. Keep it direct, remove defensiveness, and preserve the three dates.”
Save The Strong Model For Review And Risk
The strongest model is also useful at the end, especially when the output matters. Use it as a reviewer: “Read this email as the recipient. What could be misunderstood?” or “Which parts of this plan are unrealistic?” This is where stronger judgment pays off, because the task is not just polishing words. It is finding weak assumptions before they cause a problem.
For low-stakes work, skip the final review or use a cheaper model. For high-stakes work, such as a job application, client proposal, complaint letter, travel plan with tight timing, or health-related question you will verify with a professional, do the final check with a stronger model. The point is not to use the best model less. It is to use it where better thinking changes the result.
Key takeaways
- Use stronger models for judgment, planning, sensitive writing, and final review.
- Use cheaper models for summaries, cleanup, formatting, drafts, and variations.
- Ask the strong model to make the plan, then give that plan to a cheaper model to execute.
- Do not hand off vague work. Include the goal, audience, tone, key facts, and constraints.
- Bring the strong model back when the cost of a bad answer is high.