Within a minute Gemini 3 via Gemini CLI had picked up major architectural performance issue. I had it write it to a doc, had Codex review it, codex pushed back saying it's a non issue. Gave the pushback to Gemini 3, and it was insistent. Fed that back to codex and it completely caved, agreed, and pointed out that yes, it's a major issue, yes we need to deal with it right now in this stage of implementation, and yes the entire plan that Gemini 3 produced is rock solid. Implemented it and it's a huge win.

A few minutes later, same thing again. Another huge win from the pure cognitive horsepower that is Gemini 3. Again validation from Codex, which is impressive considering Codex came up with the design and plans that we're working off.

The wins keep coming. It's incredibly powerful. It has some silly bugs, like it'll just YOLO into actually implementing the doc it's supposed to be strategizing about. But these are cosmetic issues that are far outshone by the raw cognitive horsepower of this model. It's like a new kind of super powerful jet engine attached to an outdated airframe. The engine is an absolute game changer, but the entire system needs some work to hum along the way codex does.

Congrats to the Gemini team. I think you've moved the state of the art forward in a meaningful way with this one.