I’ve always been more interested in making something than selling it. Give me something to solve and a sketchbook or blank screen, and I’m happy to spend the next few hours working the problem. It could be a logo, a layout, or a complex workflow. But asking strangers to care about what I made is...hard, and doesn’t come naturally.
Andy Budd, who spent five years working with early-stage founders, describes how heads-down work on improving the product is really ignoring the hard work. Which is selling and getting rejected. Whether it’s getting hung up on, cold emails that go nowhere, or repeating the same demo over and over, it’s grueling.
I get it. Building is my comfort zone. I built and launched AI for brand strategy and had trouble selling it. Persuading others to try something new is way outside that zone.
Vibe coding tools have given DIYers the same capabilities that 3D printers gave to hobbyists. With a little persistence and know-how, folks can make things, software that they couldn’t before.
Adam Waxman built five tools around his household’s own rules and data, including a sleep app for his newborn that took a week to make and lasted about four months. Once his son slept through the night, the family retired it. His personal software didn’t need a market. It served a single household until that need ended.
That changes what deserves to be made. A useful app can serve one family, a tiny community, or even one person.
About a month ago, I went to Comic-Con in San Diego. I’ve gone for the last dozen years and grew accustomed to the shitshow of a schedule their website and app present. It’s impossible. Over 3,000 events across 51 rooms and venues. It’s a nightmare. But the weekend before the con, I just wondered if it would be possible to build a better scheduling app. I dispatched Claude Fable and GPT-5.6 and had a working app by Monday. (I’ll write more about the making of the app in a future blog post.)
Who knows if the app will find its audience, but I don’t care because it helps me, once a year. The software can last until the problem goes away. I don’t have to write a growth plan for it, and as long as I used it during the con—and I did!—is evidence that it worked.
Traditional software economics filter ideas through a business case: can the expected audience justify a team and budget, followed by distribution and years of maintenance? Niche audiences seldom survive that calculation, even when the idea would be useful to them.
The calculation changes when a capable person—and designers are immensely capable—can build the first version in days. Lifespan and reach become design choices rather than prerequisites. Brian Lovin built an app earlier this year to chat with his tax returns. It’s great! The maker can ask whether the thing helps and what would tell them it has finished its job. Revenue can be irrelevant.
That freedom is especially valuable inside homes and narrow professional communities, where shared context is abundant but demand is limited. In enterprise software, the rules, trusted data, and processes are what make something useful, especially so for an AI agent. A generic one has to accommodate thousands of different routines. But build it bespoke, and it’ll be many times more useful and gain a lot more adoption.
Ease and lower production costs don’t remove the work around an offering. They let us decide whether a tool needs a real business or simply a person who needs it now. A company taking something to market still has to earn attention and trust. A small tool for five colleagues can begin with the people, their data, and the way they already work.
My Comic-Con app and Waxman’s data-heavy tools continue to depend on access to larger platforms’ data. I used an open calendar feed from Sched and he brought together FatSecret and Withings data with feeds from Oura and Strava since all four offered usable APIs. Joost de Valk makes a similar case for one source of truth with multiple interfaces, so smaller tools can use the same reliable information without rebuilding the services underneath it.
This looks like the future HyperCard once suggested: software assembled around the particulars of our lives. General-purpose products hold the data and provide the connections. Products built for a market must persuade buyers to adopt that foundation. Smaller tools can begin with the people who already want them, serve those people for as long as they help, and disappear when the need is gone.
After Comic-Con, my app did its job, but I’ll certainly use it again for next year’s con.
What I’m Consuming
ChatGPT vs. Claude vs. Grok vs. Gemini: The Best AI Across 10 Real Use Cases. Peter Yang compares the four major AI products through live demos covering personal agents, difficult decisions, coding, design, writing, and five other everyday tasks. The useful part is seeing the choice made by job rather than pretending one model wins everything. (Peter Yang / Behind the Craft)
More models, more choice: Comparing 11 different AI models. If you want a different methodology comparing models, Netlify ran the same coffee-shop brief three times through 11 coding models and published every result. Claude Opus 5 averaged 519 credits while DeepSeek V4 Flash averaged 2.4, but cost and visual quality didn’t move in a straight line. It’s a useful reminder that model choice is a trade among initiative, validation, design quality, and how much guidance you want to provide. (Netlify)
How AI text watermarking works. James Padolsey built a wonderfully clear visual explanation of how a language model can hide a statistical signature in its word choices. A secret key gently favors one set of otherwise plausible words, and the key holder later checks whether that preference appears too often to be chance. The guide also explains why a detected mark means “processed by” a model rather than “written by” one, and what different levels of editing do to the evidence. (James Padolsey / NOPE)
Susan Kare: Designing Icons & Graphics for the Original Mac. Susan Kare recounts joining Apple with an art history PhD and limited typography experience after preparing graph-paper icon sketches for her interview. She tells the stories behind the Command symbol, Chicago, and Happy Mac. It’s lovely to hear how a monochrome grid only 16 pixels square helped produce a visual language that made computers feel approachable. (Susan Kare / Y Combinator)
The 37signals Manager Playbook. 37signals published its 16-chapter operating manual for managers, covering authority, hiring, onboarding, performance, feedback, difficult conversations, coaching, recognition, and promotion. The first chapter links directly to the company handbook, career frameworks, review forms, org chart, and old job postings. It’s unusually concrete about what managers are expected to do and which artifacts support the work. Great reference for current or would-be managers. (37signals)
The Internet Is Still Fun
From time to time, I will link to cool websites or apps.
The AI Interface Museum. A visual timeline of how interacting with AI moved from Siri and predictive cards to image canvases, coding assistants, and agents that act on our behalf. Search the collection or wander chronologically through fifteen years of interfaces.
The Vintage Beauty of Soviet Control Rooms. Room after room of physical switches, oversized buttons, analog dials, and control panels built before screens swallowed everything. The gallery ends with Chernobyl Reactor 4’s control room.
TypeStax. Build a harmonious type scale using an interface that looks like a piece of vintage audio equipment. It supports Google Fonts, per-element adjustments, live previews, and Figma export, though you’ll need a desktop-sized screen to enjoy all those lovely knobs.
WhatCable. This free Mac app identifies what a USB-C cable can actually do, then explains whether the cable, charger, device, or Mac is limiting charging, data speed, or display performance. There’s also a command-line version for anyone whose cable drawer demands automation.
Wiki Spy. An endless I Spy collage made from objects cut out of Wikipedia images. Search by color or concept, or click an object to summon more things like it, and promptly lose track of whatever responsible adult task you meant to do.