Skip to main content
Choosing the Right AI Model

LESSON 3 OF 4

When to switch, and when not to

BY THE END OF THIS LESSON

Decide whether a better tool is worth the disruption of moving.

The cost of moving is not zero

The weak version and the strong one.
AvoidA tool that is slightly betterthe cost is paid immediately
Do thisRemoves a problem you keep hittingworth rebuilding for

Switching means rebuilding standing context, re-saving prompts, and relearning where things are. That cost is real and it is paid immediately, while the benefit is spread thinly over months.

A tool that is slightly better is usually not worth it. A tool that removes a problem you keep hitting is.

Reasons that are actually good

2 of 5 to check carefully.
  • Safe: The input is always too long
  • Safe: It will not follow a constraint you need
  • Safe: It cannot reach the place your work lives
  • Check this: Something is new
  • Check this: A comparison was posted

Switch when you keep hitting the same wall: the input is always too long, it will not follow a constraint you need, it cannot reach the place your work lives, or it cannot give you a source and your work requires one.

Do not switch because something is new, because a comparison was posted, or because someone whose work looks nothing like yours prefers it.

Keep the door open

A document you own, with 0 lines flagged.
A document you own
Your standing context
Your saved prompts
Moving is a paste rather than a rebuild

Write your standing context and your saved prompts somewhere that is not inside a single tool — a document you own. Then moving is a paste rather than a rebuild, and the question "is it worth switching" gets much easier to answer honestly.

This is also what stops the sunk cost of a long setup from making the decision for you.

WORKED EXAMPLE

SWITCHING ON NOVELTY

Something new came out this week and people are posting about it, so I moved everything across.

SWITCHING ON A WALL

Three times this month I have had to split a report in half because it would not take the whole thing. That is the same wall each time, it costs me twenty minutes, and a tool that takes the whole report removes it. Moving my standing context takes one paste, because it lives in a document.

The first is a reaction to attention, and will happen again next month. The second names a specific recurring cost and a specific thing that removes it — which is also how you tell afterwards whether the move worked.

YOUR TURN

Find out whether you have a wall or a preference.

RUN THIS

I am wondering whether to switch AI tools. Ask me what has actually gone wrong in the last month, and how often. Then tell me honestly whether I am hitting a recurring wall or whether I have been reading about something new. If it is a wall, name the specific capability that would remove it.

What a good result looks like

A wall shows up as the same complaint three or four times. If you can only name one incident, or none, you have a preference — which is fine, but it is not worth a rebuild.

KNOWLEDGE CHECK

Answer all 2 correctly to complete this lesson.

  1. 1. What makes a switch worth its cost?

  2. 2. Why keep your standing context in a document you own?

REMEMBER THIS

Switch for a wall you keep hitting, not for something new — and keep your setup portable so you can.