Ask HN: Gemini 3 and the stagnation of coding agents, what gives?

Gemini 3 is cool. Sure. Gemini 3 seems to be a strong model capable at everything you'd want. Long context, good ui design, good awareness of the codebase, and a strong ability to make decisions.

What is strange to me is that despite all of this, and despite changes for GPT5-codex, claude 4.5 etc.

We still seem to see limitations in coding agents. Where are the coding agents that I can actually work with for 30 hours? Where are the coding agents that I can treat as a thought partner?

The dream seems to slowly be moving further away from believability despite actually getting closer to said goal.

What gives? Why are we not seeing true improvements across the board? Why is UX still stuck at "Chatbot in a loop with tools"?