The age of agentic coding requires a greater focus on the baseline of code quality.
In the past, great developers often had a mindset of leaving something better than they found it. In the slack time between ticket estimates, engineers would often dedicate time to refactor something that had grown stale, alongside pushing new development forward.
Today, code is generated quickly and cheaply, with more focus given to pushing new development forward at a blistering pace. This leaves less time to evaluate the quality of existing code for opportunities to improve it. I have found that the biggest impact on agentic code quality is the existing code quality, with AGENTS.md, prompts, and skills having much less of an impact than what is already in place.
I have observed a notable improvement in code quality after initiatives to improve less-than-ideal patterns when agents replicate the better examples. Trying to include instructions into your skills or prompts of producing a higher standard will always be a losing battle compared to the surface area of what is already in place. Refactor what is being perpetuated and you’ll see the quality continue.