And yet, the latest generation of models is surprisingly underwhelming.

Hard disagree. The latest generation (Astra 6/Opus 5.5) is the first one that doesn't produce slop IMO. They basically fixed Claudish, the code quality (at least in Rust) is close to what I would write most of the time, they don't spew a gazillion useless comments through the code, etc.

I don't think you can talk about agentic coding without saying which models you're talking about - there's a huge difference between the latest frontier models and those from 6 months ago.