Note
September 18, 2026
1 min read
Put the Strongest Model Where Errors Spread
By Cristiano Pierry
Model choice should follow how far an unnoticed error can travel through the work, not how impressive the task sounds.
If an agent gets a button's spacing wrong, I can usually see it. If it gets a calculation wrong and every later step uses that value, the finished product may look convincing while teaching the wrong thing.
That is where I want the stronger model, tighter tests, and more careful review: where an error can travel before anyone notices.
The assignment is not a compliment to the hard-looking task. Some difficult work leaves an obvious failure. Some modest-looking work sits at the base of everything else. I want the agent plan to tell me which is which.
Model names and prices will keep changing. The cost of a hidden mistake is a more durable way to decide where to spend attention.
This writing reflects my personal perspectives on product management, AI, and content discovery. It does not represent the official position of my employer or any affiliated organization.