Sean Voisen

Sean Voisen

@seanvoisen.com

Trying to make computers better for human thinking and creativity. Opinions own. Interwebs: seanvoisen.com Pronouns: he/him

166 followers148 following148 posts

I'm pretty sure one of the levels in Dante's Inferno involved sitting through corporate training videos on repeat.

Went to have my I-9 notarized today and the notary was required (by their employer) to use an AI chatbot to babysit them through the identification process. Why?! Completely unnecessary use of AI. Fortunately no PII went into the bot.

Tuesday was my last day of work at Adobe, after more than 13 years total working on creative tools and design systems. It’s a bittersweet moment and I’ll dearly miss many of the people I worked with. I have a month off before my next adventure, which I’ll use to read, write, relax. (1/2)

Here’s a stupid question I’ve recently become obsessed with: Why the heck does Cookie Monster seem to appear so frequently in tech culture and major tech “marketing moments”? 🧵 (1/7)

View full thread

And all this is just from some cursory web searches. Maybe there’s more. After reading through everything, though, I still don’t know why the Xerox team chose Cookie Monster. (Alan Kay seems pretty active on Quora and I’m very tempted to ask .) (7/7)

Here’s a stupid question I’ve recently become obsessed with: Why the heck does Cookie Monster seem to appear so frequently in tech culture and major tech “marketing moments”? 🧵 (1/7)

View full thread

Here’s a stupid question I’ve recently become obsessed with: Why the heck does Cookie Monster seem to appear so frequently in tech culture and major tech “marketing moments”? 🧵 (1/7)

I started digging into it the other day when I learned that in 1973, when the Xerox team working on the Alto (the first PC) flipped the switch to turn it on for the very first time, they were greeted with a bitmap of the muppet’s grinning mug. I couldn’t help but wonder: Why?! (2/7)

Great shower thoughts in the history of computing: Alan Kay's idea for overlapping windows (invented at PARC for Smalltalk): "It occurred to me (in a shower, my favorite place to think) that FLEXtype windows on a bit-map display could be made to appear as overlapping documents on a desktop."

The other night my wife and I watched the new Mary Oliver documentary and were moved to tears. I have one of her books (Devotions) and enjoy reading from it now and then, but something about hearing her poetry read out loud really hits differently. Highly recommended. https://youtu.be/B7bfYkgotBM

youtu.beMary Oliver: Saved by the Beauty of the World | Official Trailer | American Masters | PBSOfficial website: http://www.pbs.org/americanmasters | #AmericanMastersPBS #maryoliver #poetry #poem #writing Celebrate the life and career of Pulitzer Prize...

Starting today I'm off work for an entire month! (Mini sabbatical). Plans: Head to a cabin, read a lot of books, wander the local trails, journal and maybe blog a bit. Mostly focus on being a human being instead of a human doing.

Watching my 9-year-old make a game in PICO-8 by painstakingly copying Lua code character by character from examples. Just like how I learned to code when I was a kid. He knows AI exists, but doesn't know it can write code, and I'm not telling him yet 🙂

Best thing I've built for myself with Claude Code: An Apple Watch app I use for capturing ideas while out on long walks. I talk to my watch, it takes transcription, runs it through LLM, and spits out a cleaned up version in Markdown. Saves to iCloud. All on-device AI using CoreML.

Remember when you used to pay a one-time license fee for great software, and then owned it for life? Wondering if we'll see a resurgence of that pricing model in the coming years.

Could be wishful thinking, but my hypothesis is that AI changes the economics of software production such that for any app which is mostly fixed cost (i.e. no service dependency) one-time licensing per major version release is viable (albeit not as attractive to makers).

Of course, most tools bake in *some* kind of service to justify a subscription. But a lot of tools can be "bring your own service." E.g. Obsidian, where you can pay for sync, or just use Dropbox or iCloud. Small, local-first, well-built software you *own*—I hope we see growth in that space.

Local inference remains largely ignored. You can run 35B Qwen on an M5 Max today, and consumer hardware will improve. I suspect in a few years the default will be local-first inference with escalation to frontier models as needed. Anyway, worth a read: backofmind.substack.com/p/tokenalysi...

backofmind.substack.comtokenalysis and john henryI have a few things to write about, but am now conscious of what ought to be going here and what ought to be going into the manuscript… so, instead of more stuff about governance and the problem facto...

This should be a golden age of robust, reliable software. But so far that doesn't seem to be happening. I pointed Fable 5 at a personal Rust project last night. It identified 2 possible crashes and a few points of possible failure related to data corruption and DB writes I hadn't even noticed.

AI can be an amazing tool for improving software quality. But instead, what I often see going to market are slop imitations and excessive new features nobody asked for. The pursuit of speed over quality is going to bite us.

About half-way through "On the Calculation of Volume" (Book I). Quiet, slow, meditative, repetitive, deep focus on seemingly mundane details ... a welcome reprieve to the pace of *everything else* in the world right now. Great way to wind down an evening.

Guess I need to write a blog post about why "prototyping in production" using AI is not the panacea everyone seems to think it is for the supposed problem of "designer-developer hand-off."

Hello #AcademicSky #ResearchSky, I'm building a new, local-first library/citation manager (for fun). Zotero is great, but I've long wanted something more modern. I have a lot working: auto metadata hygiene, local LLM, full-text search, semantic similarity, plugins. But, I'm not a researcher! (1/2)

I would love to talk to more academics who use these tools daily for real work. I built this tool for myself—to scratch my own itch as someone who loves reading research papers. But I'm hoping to shape it into something *you* would use. DMs open if you'd be willing to chat & want a demo.

Been increasingly curious how writers and academics manage reading and sources across long projects. Heard everything from "pristine Zotero libraries" to "a folder of PDFs I'll organize someday." If juggling sources is part of your work and you're up for it, I'd love a 15-20 min chat.