- Author
- Posted by tosh, with replies from many commenters
- Published
- 29 May 2026
- Kind
- Hacker News discussion thread on a blog post
- Original
- news.ycombinator.com/item?id=48322118
- Subject
- Coding with agents
What it is
The linked post is by Vicki Boykis, dated 28 May 2026. It describes feeling less in control of code after agent sessions. The thread gathers replies from many commenters.
What it says
- Boykis says agent sessions give the signs of writing code without the mental processes.
- Boykis likens the agent interface to a slot machine: pull the lever, get a reward.
- The post lists friction, such as waiting 20 minutes before using the agent.
- Commenter simonw directs agents through extensive refactoring, without typing code, to keep code in memory.
- Commenters such as thfuran reply that IDE tools already refactor instantly and for free.
- Commenter dweekly suggests tracking pushback rate, and found below 5 percent meant too much credulity.
- Commenter K0balt says skills are not atrophying, and treats agents like junior developers.
What evidence it gives
The post offers the author's own experience and a list of practices, with no measurements. The thread is anecdote, opinion and some tool tips. One commenter treats a pushback rate above 50% as a sign of poor context.
Interests and limits
Commenters are self-selected and often work in software. Several admit that their own results differ from the post. The post's memory framing is a brief primer, not a study. Comments about jobs and company incentives are opinion.