# Scripts & sole proprietorships 1/11/26 I’m still coming off of my share of winter break euphoria, watching Claude Code “do” things for me. The instinct of engineers, PMs, and designers is that AI’s job is to *build* software — to do the work _we_ do.  And that’s true — it is increasingly *great* at that — but I think that understates the impact of “code too cheap to meter.” The mental model of anyone involved in software in the last two decades is what can be ***built*** with code. The model of the “[script-kiddie](https://en.wikipedia.org/wiki/Script_kiddie)”, on the other hand — mostly supplanted by the software engineer, but still alive in grey-area corners of the internet like “sneaker twitter” and “crypto” — is what can be ***done*** with code. To define terms for the sake of this post: software enables you to do, and scripts *do*. --- The software engineer builds for generality and scale. Even the bootstrapped, vibe-coding, indie hacker cares about these things; AI is a means by which to achieve those properties with one engineer instead of ten. The script-kid, on the other hand, wants to solve a particular, momentary problem — and is deploying code as if to employ prosthetic limbs to solve it "by hand." ![[Screenshot 2026-01-11 at 9.09.02 PM.png|StockX, source of "liquidity" for the sneaker kid who jumps through absurd, programmatic hoops to get his hands on new releases.]] The sneaker kid (my younger brother, once upon a time) is buying grey market IP addresses and one-off “bots” — scripts — that bypass (eg.) Nike's latest guardrails. These routinely become useless after first use. I spent my first internship in a Philly real estate developer’s one-bedroom apartment. We were building a personalized CRM around his personal deal flow, turning his ideas — use of public datasets, better valuation via “[comps](https://en.wikipedia.org/wiki/Comparables)”, etc — into hacky software for him and some MBA interns to turn into (millions in) profit, by running the phones & visiting lots all day. Crypto is also full of "scripts." It’s notable that so many of the domains in which this "scripting" spirit is alive, independently of AI, round to finance & arbitrage. You'd only invest in low-use code, if the few uses produced significant payoff. ^49e07a I think this math changes in light of code too cheap to meter. Claude Code works by producing individual command-line commands, scripts, and lines of code that are scoped no more broadly than to the immediate user request. Much of its output isn't giving you *software* for *you* to go and accomplish your goals. Rather, it's code that *does* the things to your precise ends. Most talk of AI-era “personal software” still evokes: * apps and interfaces like the ones we use from VC-backed startups * maintenance burden & software craft which we presume individuals as too busy to deal with -- and are therefore reasons to *not* use your own software Scripts are disposable! Once executed, the thing to do is *done* - there is no maintenance or craft necessary after the fact. The "craft" is in the [tacit knowledge](https://samoburja.com/intellectual-dark-matter/) required to know what you *should* do, not in the codified program itself! It is still good practice - a source of additional leverage - to save these scripts, turn them into software, and allow Claude to compound its efficacy. But to graft the concerns of SaaS onto "personal software" is to translate the wrong thing. And "good practice," that is simultaneously "mostly unnecessary," obsoletes itself and the world evolves beyond it. There are far more people in the world who spend their time *doing* things, than *building* them. That is the language – of the "script" – I anticipate "personal software" to speak in, now that it's this cheap. %% --- I subscribed to Bryan Johnson years before he became the immortality guy and started Blueprint. ![[IMG_4092.jpeg|A screenshot of Bryan's second-ever newsletter.]] He [was](https://www.bryanjohnson.com/articles/the-past-4-years) in his second act as founder (post-acquisition of Braintree-Venmo), working on [Kernel](https://www.kernel.com/), a personal brain-scan device. It's still around — it [assists](https://youtu.be/sJcxQQ0GzyI?si=1ccZlh-lM4BGdk5e&t=2910) his personal experiments via Blueprint — but I find it notable that Bryan’s more durable “product” is the information, individual anecdata, and lifestyle he represents, rather than the machine he helped build.  Bryan's program at large strikes me as a multi-million dollar "script". In initially pursuing Kernel, biased as a "builder", Bryan was a layer removed from his real objective: to _live_ forever and *be* the healthiest human alive. Once he figured out he wanted to *do* that stuff, rather than *build* a particular thing, it clicked.  --- The hyper-production of B2B SaaS can be attributed to something like: 1. The people who find it the easiest to go and productize their itches, are people who know how to "productize" anything at all. Their itches are likeliest to have to do with company-building. 2. They're not their own *customers*, most of the time - so they're zero degrees removed from company-building challenges, and one degree removed from their customer's actual operations. Easier to solve the zero-degree challenge well. Customer "obsession" is a form of _compensation_ for this reality; you have to be _obsessed_, to successfully breach the extra distance. **This is what I think changes.** I don't buy the debate re: if PMs, engineers, or designers are best-equipped to capture the upside of AI writing code. The *operator-entrepreneur*, in the trenches of their own, domain & industry-specific, operational problem, will force-multiply themselves; without intent to resell software, and/or get in the business of "maintenance" or "craft" (at least in the context of software artifacts). I don't think this person, in the long run, is going to approach this as an app-building or software-building exercise. I think they're going to express the intent to *do* particular things, that intent will compile down to code, and the thing they want done will get done in an absurd fraction of the time. I wish I could say something like - this person *becomes* the PM/engineer/designer, all at once - but I don't even think *that's* going to be true! The real estate developer stays a real estate developer - and the computer does *precisely* what he wants. The best users of their computer, will be 10x or 1000x in their roles, and in the right role that *is* the business. There will always be pre-packaged, well-crafted software that will provide better ergonomics – sometimes there are "right" answers to problems – and sure, AI will make its production more efficient. But the sorts of software that *eludes* pre-packaging – because the economics don't make sense – will *continue* to elude both the pre-packaged mode of production and its shape. There will be more scripts! Engineers & PMs interested in starting companies – I'm talking to myself – would do well to reflect on what they'd want to _do_ for themselves and others independently of "company" or "product" building, and try to get their computer to *do* it better for them. I consider us addicted to leverage & generality - but [scratching our own itches](https://paulgraham.com/organic.html) & [doing things that don't scale](https://paulgraham.com/ds.html) were always part of the process! %%