<![CDATA[Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“]]>Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“

I also assume, work for human coders gets more boring and less innovative if you have to spend more time reviewing code in stead of producing it (this is the case w text) - anybody can comment on that?

https://arstechnica.com/ai/2026/10/ai-coding-agents-generate-more-code-but-not-more-software/

]]>https://forum.fedi.dk/topic/cf2a6118-82df-49c0-9402-3ab1ccd6312a/before-letting-human-coders-off-to-ai-coding-agents-consider-this-study-letting-ai-write-your-code-looks-like-a-double-edged-sword-with-increases-in-coding-speed-counteracted-by-similar-increases-in-human-code-review-time-and-effort.RSS for NodeFri, 23 Oct 2026 20:56:15 GMTMon, 19 Oct 2026 05:34:11 GMT60<![CDATA[Reply to Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“ on Mon, 19 Oct 2026 08:55:06 GMT]]>@pernillet It's far worse than "boring & less innovative" - it quite simply isn't possible to do the job that the "human in the loop" is assumed to be doing & demanding people do it anyway will lead to severe stress & other health problems.

BTW these errors can't reliably be found programmatically either - basically introducing AI will produce large quantities of error-riddled code & there's nothing to be done about it.

Said code will then be foisted upon us all & we will bear the cost.

]]>https://forum.fedi.dk/post/https://helvede.net/users/jwcph/statuses/117466733408706898https://forum.fedi.dk/post/https://helvede.net/users/jwcph/statuses/117466733408706898Mon, 19 Oct 2026 08:55:06 GMT<![CDATA[Reply to Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“ on Mon, 19 Oct 2026 07:48:41 GMT]]>@benjaoming Terence Tao is active on Mastodon and writes a bit on the effects of AI on Math.

https://mathstodon.xyz/@tao/117395267721642920

There is also a lot of comments along the lines of "Hey, this result from OpenAI seems to be stolen from someone using OpenAI to write their paper"

@pernillet

]]>https://forum.fedi.dk/post/https://social.data.coop/users/pmakholm/statuses/117466472192619310https://forum.fedi.dk/post/https://social.data.coop/users/pmakholm/statuses/117466472192619310Mon, 19 Oct 2026 07:48:41 GMT<![CDATA[Reply to Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“ on Mon, 19 Oct 2026 07:46:21 GMT]]>@pernillet We're no more reviewing AI code than we're reviewing human written code - and regardless of how much people talk about code reviews that's not something that we have spent a lot of time on for quite a few years now. We rely on automatic tools to do verifications - for humans and for AI.

]]>https://forum.fedi.dk/post/https://masto.sangberg.se/users/troed/statuses/117466463023967995https://forum.fedi.dk/post/https://masto.sangberg.se/users/troed/statuses/117466463023967995Mon, 19 Oct 2026 07:46:21 GMT<![CDATA[Reply to Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“ on Mon, 19 Oct 2026 07:39:04 GMT]]>@pernillet Similarly: Was listening to a mathematician on BBC the other day state that AI can now produce mathematical papers that are 50 pages long and would have taken years to write.

He then went on to say that it would take years to understand those papers and whether they are correct.

Of course no one was asked to give a critical assessment of the benefits here.

]]>https://forum.fedi.dk/post/https://social.data.coop/users/benjaoming/statuses/117466434387752738https://forum.fedi.dk/post/https://social.data.coop/users/benjaoming/statuses/117466434387752738Mon, 19 Oct 2026 07:39:04 GMT<![CDATA[Reply to Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“ on Mon, 19 Oct 2026 07:31:33 GMT]]>And if I may add. As an outsider, I think it may be way more difficult to "test"/review text writing. But perhaps I'm just way more unconscious about what I'm expressing and how I'm doing it, and moving to more intuitive tooling would let a real writer work way better. Martin Thorborg is what I would call an amateur writer in the sense that he has written a lot - but I know that he has embraced AI writing to some extent, so he may have some input to this side of it.

@sorenhave @pernillet

]]>https://forum.fedi.dk/post/https://mstdn.dk/users/Pascal_dher/statuses/117466404843296182https://forum.fedi.dk/post/https://mstdn.dk/users/Pascal_dher/statuses/117466404843296182Mon, 19 Oct 2026 07:31:33 GMT<![CDATA[Reply to Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“ on Mon, 19 Oct 2026 07:17:37 GMT]]>@sorenhave @pernillet
Therefore a frustration becomes a freedom in the sense that things can now be achieved easier and better. In our company we just introduced that the non-programmers can add new fields and make other changes with AI. Then it's sent to the product developer and programmers to ensure the fit and quality. Perhaps it's ok, perhaps it needs to be reworked but the idea is good.
I wrote some more about it in these two danish posts.
https://mstdn.dk/@Pascal_dher/117461661987757835
https://mstdn.dk/@Pascal_dher/117455786380505197

Post 2/2

]]>https://forum.fedi.dk/post/https://mstdn.dk/users/Pascal_dher/statuses/117466350047581136https://forum.fedi.dk/post/https://mstdn.dk/users/Pascal_dher/statuses/117466350047581136Mon, 19 Oct 2026 07:17:37 GMT<![CDATA[Reply to Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“ on Mon, 19 Oct 2026 07:09:40 GMT]]>@sorenhave @pernillet I'd love to.
Things are changing from month to month. The question is what people are doing. Are they builders, in the sense that they are laying the tracks in front of them and have a sense of what problems they want to solve, or are they simply coding up a spec? The latter can largely be replaced by AI, and the person asking for it can even remember the details/context because it returns after 5 minutes, rather than 3 days. For builders, it is an incredible new freedom.
I'm a technical builder with a computer scientist background working on a very complex product, so for me control is extremely important. It has to be correct, otherwise my customers will loose a lot of money. The review has definitely been the bottleneck, but it's coming to an end, because it's turning out that there are a million different ways to automatically test things which I would never have spent time on before, but because code is dirt-cheap now it's a practical solution.
Post 1/2

]]>https://forum.fedi.dk/post/https://mstdn.dk/users/Pascal_dher/statuses/117466318802201539https://forum.fedi.dk/post/https://mstdn.dk/users/Pascal_dher/statuses/117466318802201539Mon, 19 Oct 2026 07:09:40 GMT<![CDATA[Reply to Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“ on Mon, 19 Oct 2026 05:47:39 GMT]]>@pernillet if a person hands me some code they wrote I'm pretty confident that A) they had a strong motivation to write it. B) the motivation of their logic in choice of solution was based on the existing concepts and ways of working in the code base.

With AI code there is neither.

]]>https://forum.fedi.dk/post/https://mastodon.gamedev.place/users/morten_skaaning/statuses/117465996311024807https://forum.fedi.dk/post/https://mastodon.gamedev.place/users/morten_skaaning/statuses/117465996311024807Mon, 19 Oct 2026 05:47:39 GMT<![CDATA[Reply to Before letting human coders off to AI coding agents, consider this study: “letting AI write your code looks like a double-edged sword, with increases in coding speed counteracted by similar increases in human code review time and effort.“ on Mon, 19 Oct 2026 05:40:15 GMT]]>@pernillet Maybe @Pascal_dher can comment?

]]>https://forum.fedi.dk/post/https://mastodon.nu/users/sorenhave/statuses/117465967198101044https://forum.fedi.dk/post/https://mastodon.nu/users/sorenhave/statuses/117465967198101044Mon, 19 Oct 2026 05:40:15 GMT