I never see anyone else bragging about this but in the past 2 years, I've had *five* engineers in my team take multiple months off, either as new parents, or for their 5-year sabbatical all on the most fast-moving & innovating engineering team in the company. These things are not disconnected.

it remains true: to get better at doing something, you have to do it a lot, preferably in a row, and see how it turns out writing. playing music. teaching. sports. doing dishes. grant proposals. any human skill you care to name

Happy pride — remembering when the US was progressing rather better than recent days (2004)

I admit I would like to see a bit more variety in the "gotcha" questions people use to show gaps in AI abilities I mean, I have never once in my life asked any program of any kind how many [letter] are in [word] — can we mix in some more useful-feeling gotchas?

"[Chart type] of [type of data] that shows [reason for including chart]" is a start but alt text that leaves out any *conclusion* you can draw from the chart is still common. "A line graph showing EV, hybrid & ICE car sales over time to show large recent shifts" includes a reason.. but is useless

So here's my thesis re: "AI" at this point. 1. Not a "strong" AI 2. Not a "soft" AI 3. Questionably general 4. Still a useful tool and one that's going to permanently change the technology landscape

I'm not a philosopher but it does seem obvious that ethics apply to actions people take; we are the responsible actors. AI ethics would just be "are people deploying / using AI in ways that are morally good or not," and it's up to us to evaluate outcomes & regulate as needed.

slightly increase a band 23cmИ

It's curious — I know enough of how LLMs work, there's no consciousness that'd be upset if I experimented to see what kinds of abuse would be rejected vs tolerated by different models but I just don't want to. It'd feel repugnant. Alarm bells going off.

Yeah, this just changed my view. Starting from scratch risks building the wrong thing, whereas other companies can be a good proxy for what you want to do. simonwillison.net/2026/May/6/v...

There's an amazing communication hack where you share your interpretation & next actions, so the other person can see where you're headed and maybe help course-correct ... this was around before Claude, I swear it

Trying to identify the helpful animal in this graphic? Here's a little rule of thumb to help you figure it out: Q: How do I distinguish a stoat from a weasel? A: A weasel is weasily wecognised and a stoat is stoatally different credit: Game & Wildlife Conservation Trust

AI-run business experiments are interesting and fun up to the point where they waste the time of humans who haven't opted into the experiments - I think they need to keep their own human operators in the loop for outbound actions that affect other people simonwillison.net/2026/May/5/o...

With AI, we can now create tech debt at unprecedented speeds ✨ Super curious if we can get *out* of it again, also at unprecedented speeds. For a start, if you have a working product you're WAY ahead of people building "good architecture" for a product they're still inventing with zero users.

I've been using Android TalkBack for a while as a way to 1. Read articles, stories etc without needing my eyes on a screen 2. Remind myself how terrible accessibility is, despite the fact that we figured out this tech decades ago TalkBack does slowly improve; but overall experience gets worse

Trying to figure out why I'm utterly uninterested in this but others are I've never wanted a dustpan or paint rollsr tray that held a large bucket worth of dust / paint, for one — too heavy, disaster if it spills, more paint than I'd ever paint in one session (and if I have that much dust: ouch)

We are apparently inventing a way to generate media that just remixes what you already recognize & how you already think. No need to encounter other lives, mindsets, cultures, even other characters; no risk of accidentally enriching your perspective; just eating your own output, over & over, great

First principle of data privacy is: what you don’t have, can’t be stolen from you, and no one can compel you to hand it over.

I'll need to write this down as a blog post, thanks for the prod. off the top of my head: - moving away from a task flow as the dominant paradigm - reducing noise in the flow - constructing delegation differently - much more recoverability - many more layers of abstraction on change/work history

In an interesting thread about vigilance exhaustion errors, and why Claude Code UX feels like it could be much better -- my reply here was super-compressed and not worth going into depth in thread, so I'm pulling some of it out to my own. Firstly, Anthropic has pressures that most of us don't:

Recurrent pattern of emails wanting to sell me new AI-powered services that'll help my team with a bunch of things we already built our own tools to do ...they aren't lost causes, though, imo; I think *most* teams do not want to rebuild their processes from first principals, over & over

Cynicism: needle but no thread May have a sharp point, not much use

I keep resolving to start regularly posting creative work publicly as a way to push back against AI music/art (I'm angry hearing AI music in cafés, and it's growing) but my music never meets my taste, nor does the app I made to make bsky videos from audio And oh, right: creative work is SO HARD.

So many people join team "AI is evil" or team "AI is good" and then fight to score as many points as they can off the enemy ... why do that? It seems like a poor way to figure out the best paths forward

Think of building a detailed spec as you go (mostly not in advance): high-level vision & testable design, then more detail about "why" at all levels. Really annoying, actually! Quickly stale and misleading, half the team does it wrong... Useful then tedious & bad. But suppose AI keeps it accurate.

UIs fully customized to the user, made on the fly, are suddenly possible and sound very cool but I remember the *many* ways people interact with software (and with each other, about software) that rely on consistent & learnable UI ...we're still building _some_ on-the-fly UI, but thoughtfully

In zero G there's actually no limit to a densely-packed cuddle pile; no one gets crushed

Super interesting — I also notice that solar power generation & battery tech research is NOT adjacent to horrific-scale weapons research, does not involve creating many of the materials for weapons, etc. That seems like a wise shift.

"The meek shall inherit the earth," they say. Inherit... huh. So basically after the people currently in charge, the non-meek, have taken everything they can from the earth and eventually all died out THEN the meek get whatever withered husk remains Okay, yeah, that tracks

Se7en Layer Dip

Four interviews for adding an AI product manager in the past couple of days, and I'm hugely impressed by the candidates but also it's so hard to map people's abilities over to our style of working, just so dramatically different from classic PM roles, where everyone's staying in their lane.

Ah, not just me

So many people don't notice it, but cellos & upright basses are *constantly* wound up for a fist bump Rude to just leave them hanging (but be gentle) And violins/violas have little baby fists, act accordingly

i believe everyone should make music in some shape or form you don’t even have to be very good at it

One thing that's been making me aware of how social media algorithms are hard to do well: the bittersweet posts about people who've just lost a beloved pet (with photos) ... I'm seeing about 10 of these a day in my Discover feed.

LLMs are not like human actors in plenty of ways BUT in terms of handling security & safety failures... think how you handle actions a person takes. People get training, sure, but you still do not assume your SRE (or pilot) will always follow training. Good systems assume human error.

I am running twenty four The Onion subscriptions in a gastown-inspired reader insight injection rig, each subscription taking on both Mayor and Hound roles in a Ralph Wiggum loop I have lost track of the *precise* goals but we are on the verge of a breakthrough and I'll sleep only then

More from the code craftspeople about the pain they're feeling here: andrewmurphy.io/blog/the-fiv... via @jimbobbennett.dev

It's early on a cold, sunny Saturday morning, and I have a contribution to the "can an AI be conscious and have feelings?" debate: the strong empathy I have for this recycling bin that partied too hard last night, and is undebatably conscious of that fact right now

Got this new keyboard, lmk what ya’ll think.

In fact, good behavior change research expressly finds that helping people to believe that they are "good parents" and that part of being a good parent can include changing their minds about things like health choices is far more effective than shaming and tying messages to "you are a bad parent"

How do we line this up with: • resilient systems tolerate human (or agent) errors • if you have not tested your disaster recovery, you have no disaster recovery • the chaos monkey goes around breaking random things & your service must tolerate that That POC agent is a distracted new hire; so...

I love hiding jokes like this in the house, but no idea if anyone will ever notice (to be read in the voice of a movie mob boss)

When I see companies cutting jobs because "with AI we can do the same with fewer people" I instantly think about the quality of their products, customer service, everything. They're dropping people with years of understanding of their business & customers — to stand still. As if that were success.

I keep getting whiplash between hyperbolic takes on AI — the developers who are convinced their jobs are gone because it's so good, mixed with the non-devs who are outraged because it clearly has no value, to people 100% sure it cannot be stopped destroying humanity, to– Seeking the cold takes

...for your own learning, for homework assignments, and for the final project. So on the board I put a list of different levels of AI use from one to 10..." 💭 - @analog-ashley.bsky.social in our latest: www.changetechnically.fyi/2396236/epis...

My daughter & her school friend were stuck & upset — their tech teacher gave them a file to upload to scratch.mit.edu (with sprites) and a little program to write... but it only half worked. I got to play tech superhero — fixed the teacher's file (failing scratch 2 to 3 auto-conversion). Fun!

Well, shit. This is batshit. This is bullshit. Even *posting* that I think this is bullshit is... I can't even. (I have a green card) www.brennancenter.org/our-work/ana...

NBC ran a piece where Mary Carrillo went to Norway to find out why they're so good at the Winter Olympics and i was yelling BECAUSE IT'S COLD THERE then she went to a little tiny ski jump for kids and was like "it's free, equipment is provided, & they don't emphasize results" and i was like. oh!

The drunk uncle theory. You don’t argue with the casually homophobic uncle at Thanksgiving dinner to change his mind; you argue so that the closeted cousin at the kids table knows there’s safe people and better possibilities out there