AI and some new division-of-labor problems

Software designer Nate Meyvis argues that AI-powered services face a new division-of-labor problem: they have more reason to perform tasks for users and more reason to offload work to users, making the balance a newly important skill in software. Meyvis notes he has implemented automatic LLM-driven updates in TurboTree while writing his blog post, and the feature is working well so far.

Here are some of the many software-design questions on my mind: I've had analogous problems in other domains.<sup 2 fn-2 </sup The forces making it more attractive to implement AI-powered tools as part of the service often simultaneously push us to simply give our users the means to plug in their own AI tools: When I think about this, I'm reminded of an old question poker players like to debate: does pot-limit Omaha https://en.wikipedia.org/wiki/Omaha hold %27em have more luck or more skill than no-limit hold'em? The excellent Ben Morris https://skepticalsports.com/author/benjaminmorris/ liked to answer that it had more luck and more skill. AI-forward services, it seems to me, have both more reason to do things themselves for the user<sup 4 fn-4 </sup and more reason to offload work to the user. If this is correct, implementing this balance,<sup 5 fn-5 </sup and choosing what goes where, is a newly important skill in software. As of this writing, I haven't actually implemented this yet. It could happen this hour, today, or not at all. Time estimates are hard https://www.natemeyvis.com/time-estimates-in-software/ , and the AI era makes some of them harder: easy but low-priority in the context of my whole life tasks are tough to reason about. ↩ fnref-1 E.g., recipe management https://www.natemeyvis.com/what-might-an-ai-first-recipe-site-look-like/ , but note that I've deprecated that project, so please don't try to sign up. ↩ fnref-2 It's of course possible to pass charges through to users, offer subscription tiers, and so on, but this all adds complexity, and model capabilities and costs are changing so quickly that it's hard to implement these in a way that aren't immediately obsolete. ↩ fnref-3 Update: I have now implemented automatic, LLM-driven updates in TurboTree, while I was writing this blog post. The process of writing the first footnote shamed me into doing it. The feature seems to be working well so far. ↩ fnref-4 ...whether that's with MCP https://www.natemeyvis.com/mcp-in-2026/ or by other means https://www.natemeyvis.com/more-on-mcp-in-2026/ . Note that those posts were written a long time ago, and MCP is treating me just fine these days, but it's still a complicated subject I hope to revisit soon. ↩ fnref-5