I leant heavy into it, defining a skill that the test needs to prove the issue, then I need to visually confirm the test isn’t just garbage, and then the agent is able to fix without modifying the test.
It’s been a few model releases since I implemented it so things might be better, but I kept running into issues with agents not fixing things before hand.
There’s never been a better time to be a Polymath.
If there's anything that 2026 has taught me, it's to reconsider the value of specializing in a single field.
Technological disruption can quickly shift the value of deep specialization.
I believe we’ve seen that throughout 2026 with the rise of Agents, and whilst I think there’s value in having the skills to able to develop software, the industry has placed far more value on time to delivery.
Seeing how some of the popular creators in the space use A.I. has not reassured me in the slightest.