Indeed. It's almost impossible to truly know a model before spending a few million tokens on a real world task. It will take a step-change level advancement at this point for me to trust anything but Claude right now.
Claude Sonnet is way better about following up and making continuous improvements during a long running session.
For some reason Gemini will hard freeze-up on the most random queries, and when it is able to successfully continue past the first call, it only keeps a weird summarized version of its previous run available to itself, even though it's in the payload. It's a weird model.
My take is that, it's world-class at one-shotting, and if a task benefits from that, absolutely use it.
So unless you love copy/pasting code, Gemini 2.5 was useless for agentic coding.
Great for taking it's output and asking Sonnet to apply it though.
Gemini also randomly just reverts everything because of some small mistake it found, makes assumptions without checking if those are true (eg this lib absolutely HAS TO HAVE a login() method. If we get a compile error it’s my env setup fault)
It’s just not a pleasant model to work with
