this seems to be an interesting result although the math is far beyond me. I'm definitely curious to hear if anyone looks at the methodology for this. it does seem to indicate that you can deliver more features at lower cost with agents, and I don't want to reject that out of hand
some salient takeaways before a deep dive, though:
- it's a … monte carlo simulation? where they're comparing against a "human baseline"? how does *that* work
- on average, 61% of the agent budget is for bug fixing
- "at realistic velocities, most money is spent on developers"
- "Monitor technical debt, since agents tend to increase it"
- "When agents are capable, developers spend 88–90% of their coding time on bug fixing, while few (if any) agents are used for bug fixing" wow, *miserable*
this does seem to encode a lot of … I would say "falsified" waterfall-era ideological assumptions about how software development works, like you can measure "delivering features" separately from "bug fixing". and yet it routinely relegates humans to "bug fixing"
Oct 15, 2026, 7:09 PM@glyph If it did turn out to be true then my question would be "how can we, as users, impose monetary penalties on a team for producing software this way?"
@mcc I like to have efficacy arguments available because they provide the opportunity to co-opt capital to make mutually beneficial decisions. a purely ethical argument may be correct but it's also a retreat that reduces the possibly space of allies.
and also just from personal experience, as annoying as the efficacy arguments are, the moral arguments are surprisingly *way* harder
in the grim darkness of the far future, there are only bugs
Oct 15, 2026, 7:25 PM@glyph </Starship Troopers>
Oct 16, 2026, 5:33 AM@glyph i assume the "human baseline" is represented by some numbers they had clod make up