For me it's a force multiplier but only for specific tasks in which I clearly define the requirements and outcomes. Maybe I'm doing it wrong but it's not even close at replacing even part of what I do. And I've been hearing how it's gonna replace me in 3 months for the last year. I think we would need to see order of magnitude improvements for that to happen.

It seems like the biggest improvement has been from the tooling around it (Claude Code, Zed's agents) that wrangle it into something useful. And that probably has a closer ceiling.

I haven't seen it. I also didn't see any production ready AI deployment that isn't supplemental at best.

The one exception I noticed sales supports for potential customers.

AI has changed my programming workflows a bit, although it really depends on the type of project. Students seem to employ it heavily. Any harder question is first put against their AI of choice. I wouldn't have done it differently if it would have been available to me at the time.