Claude Code can feel daunting, and most people's problems are not software-shaped, but it is a high-agency AI tool and the home-cooked app renaissance is great
are your problems software-shaped? — ∙ Paid — If you tell a friend they can now instantly create any app, they'll probably say “Cool!
@jasmine's substackJasmine Sun
Context & Ripple Effects
Claude Code’s appeal has already been demonstrated in developer-led creation, including a native macOS app built almost entirely with Claude Code. This post shifts the lens from coding productivity to whether ordinary users’ needs can be usefully expressed as software at all.
The post reframes Claude Code’s immediate value around high-agency users who can turn a concrete workflow or annoyance into a small custom application, rather than around universal app creation.
Its warning about daunting interfaces and non-software-shaped problems sets a clear limit on the addressable audience: capability alone does not make agentic building useful for every task.
Second-order effects
Tool makers seeking broader adoption will be pushed to reduce the translation work between a user’s problem and a software specification; the Claude Cowork approach is one response to that gap.
As more users build narrowly tailored tools, packaged horizontal software faces pressure mainly where its features can be replaced by a bespoke workflow—not across every consumer need, despite the sharper warning for horizontal software companies.
Third-order effects
If custom, disposable software becomes easier to produce, differentiation may shift from shipping a generic interface to owning distribution, trusted data connections, and workflows that users cannot readily describe or rebuild.
The durable constraint is likely to be problem formulation: agentic tools can expand software creation, but their mass-market impact depends on interfaces that help users identify and safely delegate software-shaped work.
The trend: This is one data point in the shift from software as a product people select to an agentic work surface where some users assemble purpose-built tools for themselves.
Cadillac tasks: I believe many estimates of LLM productivity boosts are over-estimates because people are using them for cadillac tasks: things that would take you a long time unaided, but have only marginal additional value.
claude code is a cursed relic causing many to go mad with the perception of power. they forget what they set out to do, they forget who they are. now enthralled with the subtle hum of a hundred instances, they no longer care. hypomania sets in as the outside world becomes a blur
One reason vibe coding is so addictive is that you are always *almost* there but not 100% there. The agent implements an amazing feature and got maybe 10% of the thing wrong, and you are like “hey I can fix this if i just prompt it for 5 more mins” And that was 5 hrs ago
This is spot on: “If you tell a friend they can now instantly create any app, they'll probably say “Cool! Now I need to think of an idea.” Then they will forget about it, and never build a thing. The problem is not that your friend is horribly uncreative. It's that most people's
@TheStalwart yeah I made a list of Claude Code starter projects and “visualize a random CSV” is #1 or you can export and visualize your own data — iMessage, Goodreads, etc! [image]
“Here's what using Claude Code initially felt like: cooking with ingredients from a stranger's fridge, the blankness of a page before you start writing, solo traveling in a country where you don't speak the language.” https://jasmi.news/...
If you're intrigued by Claude Code but also feel overwhelmed, I can recommend this more personal account of trying it out by Jasmine Sun: jasmi.news/p/claude-co...
“the browser is the sandbox” - as a connoisseur of sandboxing this article taught me all sorts of useful things, especially around the intersection of iframe sandboxes and CSP
i've been using claude code on a couple small projects — mostly organizing and assessing exported data from other sites. in both cases, i caught 3-5 errors in its display or analysis. i'm sure opus is a real capabilities jump, but it's hard to feel euphoric about a tool that's
just met a guy who cancelled a $18B contract with lockheed martin after he ran claude code with “design a 6th gen fighter, include a Long-Range Engagement Weapon, make sure to use uv and not pip”
fun niche genre of tweet on here lately is people saying “I just discovered an amazing new workflow using claude code and glorp” and 100% of the time you check their profile and they're the founder of glorp
.@garrytan is an example of an inveterate tech twitter poster. In the last 24 hours, he's found time to comment on admissions standards for colleges, drug use in San Francisco, and how terrible a wealth tax would be. Want to guess what he hasn't mentioned?