If you want to see the potential value of Claude, pick a somewhat isolated function in your code and tell Claude "write unit tests for this". Watch it write them, compile your code, validate the test, fix the ones it did something dumb on, and iterate. Laugh as you get 10 awesome tests and 2 really dumb ones you have to go in and clean up by hand for 5 minutes of work. Half a day of iterating later you'll have finished the extremely tedious test writing that would have taken you a day or two without Claude.

When doing a code review, tell Claude "review commit xyz". It finds things humans are not finding in reviews at this point.

Not sure about what you want to do for some design? Use Claude as a rubber duck - tell it "I'm deciding between implementing this with A or B and here's why... is there any thing I'm missing that would make the decision more clear, or any potential solution I've missed?" It has the context of your existing code, which can be incredibly helpful for stuff like this.

Opus 4.6 is good enough that it's undeniable that it's going to change the industry at this point. It's just a question of how big the change will be. The earlier models were shit. Everyone I worked with spurned them, including me. Everyone around me using Opus 4.6 is saying "well, shit, this is real now" with varying degrees of excitement and unease.

AI is not some magic pill. A lot of software development is not writing code, it's requirements gathering and design. That part doesn't magically go away, in fact it becomes way more important. But AI is speeding up the writing code part of things - if, and only if, you put down a good plan first.

People will structure their code where AI can churn on it (specs in markdown, and TDD all over again), it's useful enough to be worth it. We're going to get segmented into engineers who can use it and engineers who can't.

For this technology I'd not wait. Start learning it.

We're all going to be more focused on being architects instead of developers.