Here I am, stuck w 4G and a church and nothing else for miles.

These AI pets have no future.

They're a poor metaphor for what AI does. They're nothing more than glorified buttons with annoying animations.

They're meant to replicate #Tamagotchi's success, but Tamagotchi is a game designed for fun. AI pets, on the other hand, are just re trying to get things done.

We had enough of Clippy back in the '90s. It's no better now. #usability

I'm excited with @claudeai projects, but I'm turning to the the biomass clicking "Allow" again. Claude admits I've asked the thing; not it wants me to confirm whatever I asked. @bcherny, at least with the 20x plan, can we have 20x fewer confirmations, please?

Here's my feedback sandwich of the relaunched Hard Fork podcast. Also, rate my nickname: Spotify invented it and refused to change.

“Yeah, that’s me” or the role of humans in the future: to be a punching bag for a robot.

Dear U.S. developers: we, international users, don't always want local language or local currency, so please remember our selections. And, looking at you, @Spotify, sometimes we hate reggaeton.

I created this sl*p for myself, but it's useful for all who wants to cut through the noise of 65000 word manifests.

#Astra designs really well on its own, without open design or anything. Somebody on Youtube said it adds noisy detail, so I lectured it before, but boy, look at this.

Apparently, it's possible: beautiful design without 3 bounding boxes and 2 eyebrows for each damn element.

Also note that Claude Design's idea of minimalism is clipping the logo and shortening Español to ES.

If you were looking for an icon of j**king off, here it is. Ladies and gentleman, the electric heating pad!

My $120 GE smart scale’s body fat is just 1.5 × BMI − 17.5.

3 months of data, 40 readings, r = 0.998. Same numbers with socks on.

aiandtractors.com/ge-cs10h-body-…

Cursor is a reminder of 2025: you keep busy micromanaging it, pointing rookie mistakes, explaining the obvious. Hell, I felt so useful those days.

I wonder why every study insists on the cost per 1M tokens even though we’re all using subscriptions. My version: it’s easier to calculate.

With @claudeai workflows, development splits into a poor man’s $20-200 subscription resulting in extra labor, slowness, and suboptimal result. And rich man’s development with multi max subscription, fast, automated, expensive.

Given we already had stagnating price per token and we will spend parabolically more on Claude (per person per month).

The graph illustrates this trend EVEN before workflows.

Spoiler: Postiz launched on May 1st. That is the way!

#buildinpublic #ai

Same prompts to Claude, December vs May. Mind blown.As a designer, in a few months I went through 3 stages of acceptance: from "AI won't replace designers" to "designers will do curation" to "design will be art, not industry."

Being old is when you don’t want iOS Developer Beta on your main phone anymore.

What to do while a reasoning model is thinking? That’s our new form of boredom.

Yeah, I wrote this while a reasoning model is thinking.

Superficial segmentations like age, place of birth, gender, race are useless for predicting people’s character and behavior.

Much more common are occupation, political views, education level, favorite celebrity, etc.

So next time instead of “where are you from” we should first of all ask strangers “what do you do”.