I've been shipping code for 8 years. In the last 6 months, I've completely rebuilt how I work — and it's not because I wanted to be trendy.

Here's what actually stuck:

1. AI-assisted code review — I stopped catching 90% of my own bugs manually. The models are genuinely better at spotting edge cases in large PRs than I am after a 10-hour day.

2. Natural language → SQL — I haven't written a raw query in months. For anything that isn't hyper-optimized production queries, natural language gets me there 5x faster.

3. Documentation generation — The most underrated use case. Point an LLM at your codebase and generate docs that actually reflect what the code does, not what you wish it did 6 months ago.

4. Test generation — Coverage went from "we should really write more tests" to 85%+ across three repos. The key is treating AI-generated tests as a first draft, not finished work.

5. Architecture brainstorming — Before spinning up a new service, I describe the requirements and constraints to a model and ask it to poke holes. It's like having a senior engineer on call 24/7.

The pattern I see: AI doesn't replace developers. It replaces the parts of development that developers secretly hate.

I write about this every week in Signal Stack — frameworks, tools, and implementation patterns that are actually worth your time. No hype. Just signals.