Just because it's work shaped doesn't make it productive
AI Assisted/Agentic programming are pretty common place at this point. The growing sentiment seems to be that if you can't find some sort of benefit in your workflow, it's more of a skill issue than a problem with the tools. Whether you believe this to be true is really up to you, what is clear is that it's becoming more of an expectation than an option for many engineers.
I think AI/Agent tools come with a lot of gains however those don't come for free. Agent led development comes with (increasingly well documented) hidden costs that developers promoting the hype cycle continue to ignore -
- Poor/weakened skill acquisition - Particularly in junior talent that we should be cultivating
- Skill atrophy - Particularly in more senior engineers, when you don't use a skill you lose it. I wrote more on this recently.
- Comprehension debt - Engineers are supposed to be reviewing code but many aren't. Things slip through the cracks. Systems change at a pace that teams cannot keep up with. In the end they may know what it's supposed to do, but do not why or how it works.
- System bloat - Generating code is now cheap, as a result, generating code that shouldn't exist becomes an easy solution to problems.
- Insufficient quality gates - The speed at which code can be generated will always outpace the speed at which it can be reviewed. Larger PRs decrease the likelihood that the code will be reviewed correctly. DORA associates larger PRs and increased review times with worse outcomes.
These are all subjective and apply differently to different situations, but it's unfortunate to many engineers willing to throw away known good fundamentals and replace them with practices that is showing a productivity heavy tail.
I should be clear that, I'm not arguing against using AI tools, there's a lot of value there, but I'm not convinced that layers of AGENTS.md, SKILLS.md, with dozens of agents generating code in the background is currently the best way to get the most out of them, especially if the goal is delivering safe and reliable solutions to problems.
A known tradeoff with automation is that one with sufficient complexity, the solution evolves into its own system that requires its own upkeep, process and maintenance. It's easy to fall into the trap of spending way more time on the automation than the actual problem.
I think the practice putting together elaborate Agent harness setups that explicitly aim to distance the engineer from the thing they are engineering is heading in the same direction.
Ultimately my point is that these tools are cool but should be used responsibly.
The more I use AI tools, the more I have to admit that I'm not that much more productive... I simply FEEL that much more productive. In reality, the context switching of kicking several things off wipes out my perceived productivity gains. At least in many/most cases!
— Gergely Orosz (@gergely.pragmaticengineer.com) April 3, 2026 at 10:55 AM