Published
Weekend Reading — The elderly health category
Pixelcats "Opened the mail .. Derpy arrived early! SQUEE!"
Did I just vibe-code an app because I didn’t like the new UI? Why yes I did. I’ve been using Expensify for many years to keep track of my expenses. It was good. There were a few things I would change in the UI, but not enough to get me to switch. Anyway, Expensify just released a brand new UI, which feels foreign, and I have no intention of learning and getting used to it. So I went and vibe coded an app that replaces Expensify for my specific use case — collecting receipts for taxes. Expensify was not showing me good UX skills, thankfully AI makes it easy to build a one-person app.
Tech Stuff
LLM COUNTER-DEFAULTS LLMs default to agreeing with you and yet sounding super confident in whatever claims they conjur. This helpful apps allows you to tune the LLM to your desired behavior — tweak cognitive sovereignty, epistemic uncertainty, reference surfacing, sycophancy, frugality, etc. Use the dials to choose your desired behavior and it will generate a prompt to guide the LLM by your preference. Then copy that prompt into your AGENTS.md or whatever other configuration you use.
degr Screensaver that does one thing really well: it shows the current time and the weather in your location.
The sort of person who liked software development five years ago received a different set of rewards in the process of building things. The tension of “I don’t understand this problem” and the resolution of “I figured it out” are either gone or materially shifted into something unrecognizable to my dopamine receptors.
It stands to reason that the rewards of the new software development process would not necessarily feel as good to the profile of a person who liked pre-agentic development.
Open Minis On-device AI agent that can run on your iPhone or Mac. It has a surprising number of features for what's essentially an iOS app, through the wonders of running a Linux shell, web browser, and access to a bunch of native Apple APIs. I find it super helpful, and I love how just this week they did a release that added even more features. MacStories has a detailed write-up about it.
OpenWorker Open-source agent that can use everyday tools — Slack, GMail, email, Linear, et al — and whatever model you care to use — Anthropic, OpenAI, Kimi, GLM, etc.
I Tried (Almost) Every Major Model Subscription So You Don't Have To In my case, I started with OpenAI then upgraded to Anthropic, then saved money by switching to DeepSeek, found the latest Zhipu (GLM) much better, and now experimenting with Kimi.
The best way to buy AI inference is usually a subscription. It’s well-known that a $20 or $100 or $200 subscription from Anthropic or OpenAI is the most cost-effective way to buy access to their models, by far. One report said a $200 subscription could provide ~$9000 or ~$12000 worth of tokens at API rates from Anthropic and OpenAI, respectively. I haven’t done the precise math on this, but I have tried nearly every AI coding subscription now, as I subscribed to almost everything to run new benchmarks of security vulnerability auditing capabilities of the major models using their own preferred agent (the one made by the model vendor, like Claude Code, or one built for and seemingly having some kind of nod as being a preferred option, like Reasonix) as harness.
tuicr TUI for your code review with vim keybindings.
MailKite Transactional email API for users and agents. It can send and receive emails with a simple API, and also supports REST, IMAP, SMTP, and CLI. You can get started with 3,000 emails at $0/month.
Neodisk Open-source macOS disk space visualizer.
Fancy code can be satisfying for one afternoon.
Boring code keeps paying back for years.
undo This is cool: an undo for destructive commands in bash/zsh/fish. Very clever implementation. Some insights behind the scenes.
Phosphene Apple's video wallpapers: drag a video into the library or the menu bar, pick it in System Settings, wake your Mac. That's the whole app.
Eye for Design
Open-source project fools AI scrapers with poisoned font This is a super cool idea. You make the text visually say one thing, but the underlying text says something else. Mind you, relies on ligatures, so only useful if you want to defeat AI at the expense of confusing some users (this fails with assistive tech).
Peoples
My conspiracy theory is that some people at work are addicted to setting up meetings because no one listens to them at home, and they don't know how to start a podcast.
Basically you sit, do nothing, and notice how it feels to do nothing. (It will probably feel subtly weird and unsettled.) You then see if you can completely embrace these feelings, without the usual squirming and looking elsewhere. But you’ll do it only for a few seconds at a time, using your breath as a measuring stick.
Business Side
Silicon Valley loves young founders. Until it doesn't. If you raised a round but am not posting insane growth metrics on LinkedIn are you even an entrepreneur?
There are more funding opportunities than ever, regardless of age — accelerators, incubators, pre-seed funds. But that money comes with strings attached: Founders … flush with millions in cash, are expected to deliver growth in months, not years.
VC-backed startups commit more fraud, and researchers think they know why Related, don't forget that mere growth is very last year. Grift is where the real money is:
In fact, the UT report found little evidence that alleged fraud prevents founders from raising funding for new startups, even when those fraud cases received major media attention.
55 Percent of Leaders Regret AI Layoffs—and a Major Hiring Reversal Has Begun (via Anna E. Cook)
The reasons those cuts were made range from AI not automating work as well as expected—or doing so at far higher costs than planned—or managers realizing apps use and content they produce must be overseen by experience, trusted humans. At Ford, that means hiring or bringing back hundreds of engineers to work on improving the tech’s performance.
Machine Intelligence
I’m curious. Why haven’t we heard any of the AI players promise that their products will once and forever get rid of spam? Get rid of malicious ads? Get rid of malware in general? Inquiring minds want to know.
The Art of War, Applied to Litigation I have an AI agent that helps me handle legal cases. It's not a lawyer, but it's quite useful and the hourly rate is around $1.20, that's 100x cheaper than a human lawyer. I figured I'd train it on The Art of War, which is a classic book on how to win a war by not fighting, minimizing your losses, and working smart not hard. I asked it to come up with core principles and here's that list. Enjoy.
Insecurity
"residential proxy network" is what you call a botnet when it's run by a corporation that will never be prosecuted for doing something that would land a 13 year old in jail for the rest of their life
Everything Else
CheapFoodMap Find affordable meal options in your area. This is great when I want to go out to eat, but don't care to spend $20+ on a plate.
Is it nostalgia to remember when a lot of this stuff was called "fraud" and there were consequences?
sometimes I say "no worries!" when I am at least 33% worries
The Brand Ledger Tracks 345 brands, from the tools in your garage to the pans in your kitchen. Who owns them, what they used to be, whether they're still worth buying. (via Lou Plummer)
Today, we made a difficult decision to lay off 7% of our AI agents
"Do you hear something?"
"I have tinnitus, I always hear something."
We used to be able to buy things in this country Locking up toothpaste was a warning sign that our executive class was crazy:
Instead, the great lotion lockup was a corporate moral panic, which has been a disaster for their profits and for the rest of us a degrading enshittification of running basic errands.
I have adopted an "always close the tabs" approach to my web browser and it is liberating. Either I'm actively reading something or I'm not. No minimized windows full of tabs. No huge tabs list on my phone. It's great.
AI companions are mainstream (a third of Gen Z report romantic use), and the alarm around them follows Cohen’s moral panic cycle: rare tragedies become the representative case, thin correlations become causal headlines, and a dozen-plus jurisdictions have passed control measures, some of which the evidence says backfire. The real risks – data sensitivity, manipulative design, youth protection – go under-researched while panic forecloses the studies that would settle the question.
Inside an AI TikTok Shop Slop Factory That Shills Supplements Recalled By the FDA “the elderly health category”
Jan Wildeboer Speaking of enshittification:
Hearing aid hacker needed! My father‘s hearing aids are being repaired, so the shop gave him a pair of Signia Motion C&G SP 1IX as temporary replacement. Nice. BUT. It seems they were configured as trial devices, this wasn’t mentioned to my father. The trial period has ended a few hours ago, so now they play an annoying melody every 2 minutes and we cannot stop this. The hearing aid specialist is on vacation until end next week. So. Anyone knows how to fix this?
1,741 “informed” consents with one click?! GDPR complaint against dict.cc filed
Reading all of their privacy policies would at least take 170 hours (even if you just scan each policy for 6 minutes). That is more than an entire week for one single consent request. In addition, these “partners” often say that they forward your data even further. For the complainant, as well as for other website users, this makes it practically impossible to understand the consequences of their “consent”. This common practice (see e.g. www.repubblica.it, www.bergfex.de, www.fifa.com) makes obtaining an “informed consent”, as required by the GDPR, essentially inconceivable.
Hans-Cees "It is called the elephant"
