The Model Picker Problem: Why Too Much Choice Makes Copilot Worse for Most Users

Is model choice useful?

Both GPT-5.3 and GPT-5.4 have hit Microsoft 365 Copilot as options under "More". My understanding is that you're not going to get those by choosing "Auto" so, even if you have access to them, you have to hunt them down and choose them.

But is this idea that choice is good flawed?

For most purposes, the default of Auto is good enough. And if the message we want to get across is that it's beneficial to make a manual choice, then unless you are someone who likes exploration and experimentation, surely the smartest model is always the best choice. Sure, smarter equals slower, but on most tasks even a slow AI response is competitive with the quickest human, but those dumber models will often let quality slip below what is needed. On average, it is safer to sacrifice speed than smarts.

We are in the phase of mass adoption of AI tools. The excited experimentalists are getting outnumbered by the mildly disinterested masses, and one of the biggest takeaways I want to convey to that profile of user is that Copilot is the option that just works. A seven-option model picker doesn't help my case!

Why not let us geeks have our complicated AI tools in other parts of the platform? Can we just make Copilot as straightforward and boring as possible for everyone else?

First posted on Linkedin on 03/07/2026 → View Linkedin Post Here