Today I noticed that cursor have gated the auto mode so now it never tells me what model it is actually using. Instead I now get this generic message “I’m Auto, an agent router designed by Cursor. I don’t have explicit model identification in my context, so I can’t specify the underlying model.”
Previously I was getting the exact model that the auto mode was choosing.
It seems cursor devs are actively reading the forum (or having AI scan it for issues to patch) and they are fixing whatever I point out here. This is a common pattern I’ve noticed 3 times in the last few weeks.
- I uploaded screenshots of the pricing shown for auto mode
- I mentioned I was getting claude sonnet for every auto mode
- I clarified that I had instructed auto mode to tell me the model and I had a rule to only use certain models
Within 1 day of posting each of the above the exact thing I mentioned was changed. Now either this is just a very unlikely coincidence or the dev team are actively using the forum to plug holes (fine but don’t break the app functionality!)
Auto mode has gone from useful to useless in the span of 1 week and now I have no idea if I’m just getting dumb models or if I should upgrade to ultra or what. Cursor is again a black box and I’m back where I was 2/3 weeks ago feeling frustrated because everything was working perfectly and now it is not - now I’m spinning around with dumb models that can’t follow instruction and I’m wasting time trying to hand hold them.
The same commands and prompts with sonnet work perfectly - I tested it directly in claude. But in auto I’m getting some mystery model that can’t even think to use an mcp or cli to fetch database schema.
If cursor are trying to get me to use claude directly and for me to cancel my cursor sub then they are doing a great job at it.
.