An increasing amount of code I commit is generated by an AI agent. As of pretty recently, when asked to supply my work title, I now reply with my own self-appointed designation of "Code Generation Coordinator". The work I do is still, most definitely, software engineering but I rarely type at a syntactical level. Here are some important lessons I've gleaned from my own trial-and-error as well as others I'm observing:

1. The cost (in terms of time and effort) to produce code has plummeted.
   • This concept is the foundation of everything that now follows:
2. Since test cases are just code, the cost of writing tests has also plummeted.
3. AI agents thrive on acceptance criteria of any kind.
4. Since a comprehensive test suite qualifies as an effective acceptance criterion, the value of such a test suite has skyrocketed,
   • Such a test suite was already worth a lot even before AI came on the scene.
5. What I am calling AATDD (AI-Assisted TDD, powered by agent skills) is now a no-brainer.
   • TDD as a concept has always made sense, but it came with a very real mental and emotional cost which prevented the vast majority of developers from using it consistently.
6. The cost of (re)writing full-blown integration/systems tests has plummeted.
7. The cost of executing large-scale, test-supported, multi-phase refactoring adventures has plummeted.
8. The cost of migrating from one approach/library/LANGUAGE to another has plummeted.

Writing code is just one part of the software engineering process, so without corresponding gains in all other areas we're not going to see a dramatic uptick in velocity across the industry, but so many of the tasks that so many developers have traditionally shunned are now palatable, even fun!

In the last few weeks at work I've seen test coverage in several of our packages go from 0% to 100%, almost overnight. Newer libraries and approaches have been inserted with zero impact to customers. Old, calcified systems we thought would have to be rewritten from scratch are being incrementally migrated across a pretty intricate sequence of events, all envisioned by AI agents, to a more flexible structure and very soon will be perfectly positioned to begin serving new use cases. At home I've been able to make use of small windows of time to improve personal software projects, greatly improving my own effectiveness and efficiency.

There are a lot of aspects of AI's impact in the world that can (and should) be debated (and improved), but I definitely see more potential now than ever before. This was definitely not on my 2026 bingo card.