Toots from 2026-05-09#
“For example, AI coding agents are now able to complete some days-to weeks-long software engineer…#
“For example, AI coding agents are now able to complete some days-to weeks-long software engineering projects autonomously”
I see there’s a footnote, I wonder what that says…
“See MirrorCode for evidence on this. MirrorCode tasks involve reproducing existing software, which is a specific task type and not necessarily representative of software engineering in general.”
So AI agents can complete work that isn’t work people are typically responsible for. (Yes, porting is/was a thing). I wonder what they’ll conclude from this:
“Therefore, we should expect people to substitute towards tasks that take advantage of AI.”
I wouldn’t expect that at all from the data provided, although I expect people self-select AI-amenable tasks for all kinds of other reasons.
https://metr.org/blog/2026-05-08-task-substitution-and-uplift/