We May Not Have Nice Things

17 Aug, 2026

In December 2025, I started filling my Home Screen with personal apps, replacing all software I couldn't trust with clean, simple, local-first apps.

In addition to private never-ruggable bookmarking, a personalized exercise program, and a crypto wallet, I've kept on my Home Screen for at least 6 months my own secure messenger, a habits tracker, a personal finance app, and several AI orchestrator front-ends that allow me to Claude on the toilet while commuting.

Two obvious limits when I started: I would never replace Google Maps, with all of its access to algorithms and usage patterns. And I would never replace a Todo / Tasks manager with my own greenfield build. There's man-decades of R&D embedded in Linear, ClickUp, Notion, Asana, Jira, etc. A CRUD-slop productivity app should pale in comparison¹.

Recently, my friends made Stupid Social and Casberi with an entirely different approach: composing around powerful existing building blocks. They don't do the heavy lifting themselves, but they aggregate, filter and route information from other places.

So I'm retiring my personalized exercise program, replacing it with Arnold Schwarzenegger's Pump Club, which offers benefits that personal software can never offer:

• vast resources to provide fresh experiences
• capital backing to provide breadth and quality, above what an LLM can give
• editorial focus from the team of a former Mr. Olympia winner and life-long fitness "influencer"
• proximity to celebrity glamour
• an optional community

The cost of these valuable benefits is:

• upfront monetary outlay, albeit with an offer to recoup it via commitment over months
• muh data

Taking a hard look at the limits of AI-enabled personal software, there are three clear effects that everybody needs to juggle:

• Pareto principle. To get to high quality, you will spend 4x the effort that it takes to get to good quality

• Jevons Paradox and Hedonic Adaptation. As powerful agents enable higher quality one-shot software, the goalposts shift

• Efficiency. One entity spending resources to achieve quality, and distributing it to thousands, should be more efficient than thousands of entities independently spending resources to achieve similar quality. This overlaps with the economic theory of Gains from Trade due to Comparative Advantage

Two possible equilibria

In game theory, there is the idea that actors can land in a stable equilibrium, where there is no Pareto improvement available, which is nevertheless less utility-maximizing than an alternate equilibrium (which might be less stable).

One. We might get enough people paying for quality software, to allow everyone to get the quality they desire, at reasonable utility-improving price points. Think "Microsoft global enterprise software dominance era" and "Golden Age of Apple UX Permeating Into Tesla Dyson Nest Android and Xiaomi".

Two. Or, AI capabilities might destroy the market for quality software, which leads to everyone using their own shitty good-enough software, at quality points that individuals have the skill / resources to reach².

We might end up living like Wild West pioneers, where everyone eats butter they churn at home, before refrigeration and railroads could induce factory farms to produce at 100x efficient scale, competing with local dairies on many dimensions of cost and quality. We might also wear durable burlap.

But not due to lack of technology. This outcome of ubiquitous, free, just-good-enough software would come from everyone's individual preference to undercut paid software pricing, to below the price point where quality software can be produced sustainably.³

Personal software eating the world may bring immense deadweight loss: aggregate utility-improving outcomes the market can not reach on its own. Economists understand well the sets of remedies for DWL, and they are balanced against the danger of entrenching non-free-market dynamics, upon a production technology that has not yet ossified.

If we end up in 10 years with all our current software business models defunct, I will have already learned to produce welfare-improving software for myself. And I will be riding the wave of capabilities, assisted by AI, in a way that was much more expensive before the advent of AI.

But we should all spend less time reinventing existing software. I actually think it's better, for humanity as a whole, for talented developers and designers to contribute openly to libraries and open source projects, than to spot a good idea half-executed, and copy it for personal value and clout.

I hope the more open / less gate-kept Open Source projects succeed, due to quality contributions congregating their valuable contributions with them, rather than in closed projects. For now, this is not an obvious outcome, as the more open projects are being drowned in slop.

This is a little like the entropy problem, and Heat Death of the Universe pulled forward.

Also, current agentic coding harness business models are incentivized to produce novel patterns and more volume of code to maintain, rather than to extend existing patterns to produce quality, maintainable code.

I fear the trend towards YOLO stop-reading-code-ya-dinosaurs it's-just-like-what-we-did-with-reading-assembly-code-after-all will drown all software in slop, even as LLM coding agents sing Siren promises of productivity gains.

Previous Next