it's built-in to the tool, you create a rules/ directory and fill it with these yaml-formatted rule definitions ast-grep.github.io/guide/rule-c...
Latest Posts by Tom MacWright
id: yieldable-errors message: Directly yield errors instead of wrapping them severity: error language: tsx files: - app/*/*.ts - server/*/*.ts - app/*/*.tsx rule: pattern: yield* Effect.fail(new $A($B)) fix: yield* new $A($B)
another refactoring done efficiently with ast-grep in a way that can be enforced forever ❤️
haha, sorry - I saw your talk! Should have said hi
speak for yourself, i've been hitting the gym
great work hosting! sorry i didn't swing by to say hi during the thing
the good: pretty easy to make super lightweight components with inline css and js. a+ for speed-obsessed web dev.
the bad: super super hard to learn, very easy to mess up, generally quite bad error messages.
under the hood, been using webc for a few things on my blog, for the @observablehq.com plot chart on macwright.com/2026/04/05/i..., the theme selector, and youtube embeds.
i'd say it's 2/10 DX, 9/10 magic, technology that fills me with both joy and fear www.11ty.dev/docs/languag...
finally posted thoughts about the atmosphere conf (along with an illustration and a bunch of tweaks to the blog) macwright.com/2026/04/05/i...
just to serve mode.css to do light/dark/auto theme switching without having to put js on every page :/ it's super tiny but alas gist.github.com/tmcw/2015bda...
the same watercolor, on a scanner
it's satisfyingly large art irl, too
watercolor illustration
illo for an article i'm trying to publish. been in writing/editing/watercoloring/making-a-chart mode for three hours unexpectedly tonight. felt good to get in a ""flow state"", though the article still isn't what i want it to be.
ai agents filling out quarterly 360 reviews
stack-ranking subagents
doing bigger watercolors is better because you spend more time putting the paint on the paper
dark mode has been slowing my website down by more than a hundred ms 👿. netlify edge functions: not as fast as i'd hoped.
i really liked their original product but it ended up being kind of like replay.io, amazing tech that people don't value.
the all-time startup pivot champion is quirk i think
- [[Fiberplane]] - 2023: Debug your infrastructure in collaborative notebooks - 2024: Test and debug your Hono API - 2025: The API playground for Hono - 2026: Build better MCP servers - Later 2026: Agent-Native Issue Tracking - [AthenaHQ](https://athenahq.ai/) - January 2025: Streamline Your Clinical Trials End-to-End - June 2025: Empowering Next-Gen Marketing Teams in Enterprise - January 2026: Dominate AI Search Results - Basedash - 2020: Edit your database with the ease of a spreadsheet - 2022: The CMS for your database - 2024: Connect your database. Get an admin panel. - 2026: AI-Native Business Intelligence - Textile: - 2018: Building new digital experiences by focusing on user privacy, transparency, and decentralization. - 2019: IPFS [Textile](https://web.archive.org/web/20190702164516/https://textile.io/) - 2020: IPFS [(archive)](https://web.archive.org/web/20200805133501/https://textile.io/) - 2021: IPFS [(archive)](https://web.archive.org/web/20210629152552/https://www.textile.io/) - 2022: Tableland (crypto product that let you query data in tables) [Welcome to Tableland](https://blog.textile.io/welcome-to-tableland) - 2025: Recall labs (crypto/ai product that lets you bet on which ai models are good) [(writeup)](https://flat-agustinia-3f3.notion.site/Our-mission-is-to-accelerate-the-exchange-of-information-across-society-28ddfc9427de8148973cc33aa49bd3f2)
a weird thing in my @obsidian.md vault: a note that keeps track of all the startup pivots i see
i think it’s going / they achieved a bunch of stuff - @schuyler.info is one of the main folks
i haven't because i'm fastmail 4 lyfe, but it's probably pretty good in the specific ways that most 37signals stuff is
oh, and to deliver that data aesthetically and cheaply, protomaps has basically solved that problem, anyone can get great looking maps for very little $ protomaps.com
openstreetmap's hiking trail coverage is pretty decent, strava has been backing their updates into it for years because they use mapbox so that's where their map data comes from.
i bet 'alltrails' is the next thing to be rebuilt on atproto if it hasn't already
sure, i guess my personal prejudice is that i love everything that isn't natural language for creation and despite the manager-employee dialog that vibecoding operates on
sure, i didn't mean to do a descartes, we have to affect the world, but i don't think that the mediums of guitar/painting/writing/etc are similar to the medium of llms
when beginners fail at painting or guitar, they sometimes blame the materials but mostly are aware that they need to become better. when people fail at vibecoding, they are just disappointed that the artificial intelligence product cannot translate their vague thoughts into concrete representation.
being alone with your thoughts and struggling to produce an output is kind of the polar opposite of trying to produce an output by begging an all-powerful god to do it.
aesthetically, if it matters, seeing vibecoding workflows firsthand, the struggle of creation takes the form of begging, whining, arguing, more like a supplicant than a creator. it looks and feels pathetic, and when it is paired with a disinterest in learning, i think it is
the arc from "popular software with the worst code quality i've ever seen" to "seeing that software in subway ads" is unstoppable recently