Published
Weekend Reading — Invisible as someone else's problem
Connect to Expense There's a debate going on whether MCP is better than API/CLI access, since modern LLMs have access to the CLI and can do REST over HTTP easily. I think it's a pointless debate. APIs have their own design guidelines, MCP is different beast and much better suited to fulfill end-user needs. I think some products are better served with MCP and shouldn't bother with an API. Anyway, my point in all of this is that MCP is not for direct human consumption, but you do need to set it up somehow. Sentry does that quite nicely, and so I borrowed that idea, and created a descriptive MCP landing page for Expense which explains why you'd want to use MCP and also instructs you how to connect it to different AI harnesses (Claude, ChatGPT, OMP, et al). Check it out.
Tech Stuff
OpenClip Floating action bar for macOS. Highlight any text and it will pop up a floating action bar with commands to copy, calculate, transform, run custom scripts, etc. Also AI assistive features to rewrite, summarize, proofread, translate, and more. Open-source, quite easy to use, and very handy, if you ask me.
Stuff This looks very interesting: a playful, focused to-do app. Well designed UI, scan mode (take a photo of your written notes), lists you can share, to-do dependencies, adventure log, CLI, and more. I think every person needs a different to-do app that is perfect just for them, and this might be your pick.
neat-annotations I’m digging this. Already using it in various places in the Expense app to add a little bit of nicely styled info. For example, here I'm showing you how expenses are automatically processed. And I didn't draw or paint anything, this is all plain HTML with some wisely selected CSS classes.
No more specs On the road from specs to consistency:
We're spending a lot of our time getting AI to program like humans instead of figuring out the optimal way to that AI wants to program. Each role in the SDLC is thinking on how that role fits into AI-driven programming. We do a lot of "spec driven" AI development.
I'm also on the road to discard my SDLC practices of the past that were rooted in running a human team, and to explore the new and different ways to develop code with the help of AI.
pdfcn shadcn but for rendering PDF documents. I find this quite useful because sometimes I do need to generate PDFs that are designed for clarity and there are not many libraries for doing that.
emailcn shadcn components for rendering emails. I love that it can use React Email, MJML React, or JSX Email. I used React Email before on a few projects, since it came out of the Resend product, which is great or automating email sending.
ogimagecn shadcn components for rendering OG images. I love Open Graph images. Sometimes I share links on social media or SMS, and I like seeing them reveal a proper Open Graph image, but when you're sharing an app or dashboard, what would the image show? This library allows you to render textual content into a properly sized image.
Hister Hister is a self-serve search engine that can index your files and browsing history. You install it on your machine and point it at the files you want it to index, or tell it what to import. You can then search from the web, command line, or MCP.
I mentioned Hister before, because I find it very useful. The one thing I didn't like was having to remember which port it runs on. Well, turns out you can use portless to proxy your own hister instance behind https://hister.localhost:
$ portless alias hister 4433
Portless, which I also mentioned before, allows me to run all my local dev servers behind HTTPS and proper DNS names. I find Portless quite useful and one of the first thing I'll install on any new macOS.
Worker Threads vs Cluster vs Child Process in Node.js Clear benchmarks on startup, memory & IPC. When to use worker threads, cluster or child_process, plus a production-ready worker pool pattern.
Made with Jev And of course we now have a directory of apps, AI agents, and dev tools written with Jev. Not to mention Kev, a Jev-like decision model built from Qwen 3.5, which you can even use to play chess, and Jevchat which uses Jev's next word classifier as a chat model (because of course).
I like the concept behind Jev because sometime all you need is a quick decision or classification and chatty LLMs talk too much. I don't currently use it, though, but I do expect this capability will roll into general purpose LLMs soon enough.
SemIf (formerly OpenJev) Run a decision model in your browser.
That problem gets harder as you go from five agents to hundreds. Keeping up can’t mean reading every conversation and trying to prompt an explanation for each code change. Agents need to identify the fewest places where human attention can have the greatest impact and surface enough context to make that attention useful.
My personal process is to direct OMP to write code with a vague description of what I expect to see, but sometimes I do tell it to switch into plan mode, so I can quickly review the plan. I want to see what conclusions and ideas it comes up with — it often adds things I didn't explicitly ask for but are obvious from looking at the code. Twice I asked it to correct its plan, but typically I tell it to follow the plan. Most plans execute perfectly. For reference, OMP is the harness I use, and the LLMs are either GLM-5 or DeepSeek-4.
Catsup The reusable grocery list. Essentially you slide a knob for something you need, go buy it, and slide the knob back. All toggles, buttons, and the font are hand-drawn. (by Zhenyi Tan)
I've got two systems hooked up to my KVM and they have the same master password except with different capitalization.
I can HIGHLY recommend this if you want to feel like you're going slowly insane
VibeCoded AI-Slop License v1.0
Eye for Design
Skip the Tools, Make the Outcomes Are you writing code or are you helping people achieve their desired results?
Now I ask myself (and pester others with): are you building another tool, or are you delivering the outcome the tool was supposed to produce? I've found just asking that question inspires new ways of solving problems.
Grit your teeth and ship it Being good at building and being good at shipping are two separate skills:
Gifted programmers often freeze up. I've often seen them retreat to smaller domains where they can safely make the code "correct": tweaking dev-environment setup, or refactoring tests. Sometimes they just do nothing, and spin in shame and guilt (plus the compounding shame of not achieving anything) until they implode and quit. If they had worse taste, they wouldn't be as good at programming, but they'd be a lot more useful. You can typically improve a bad diff with time and effort. You can't improve no diff.
When features go wrong We should all care about this and design our software for a better user experience:
We do this constantly in software. The session timeout that logs you out partway through the form. The autosave that faithfully preserves the version you didn't want. The notification that's genuinely useful right up until it fires in the middle of a demo. All installed for good reasons. All of them serving somebody who isn't the person sitting in front of the screen.
Impeccable Gives your AI agent the design expertise to build from scratch, or improve what you have, and stay true to your design system. This project has SKILL.md, plugin, and npm runtime, for different needs. There's a gallery that shows what different LLMs produce (Opus 5.5, GPT-6, Gemini 3.8, etc). If your doing any DX and your project is AI-assisted, I suggest you give this a try. You might just like it.
Peoples
Why do we listen to people with no skin in the game? On taking software advice from managers and influencers who don't have to build or deliver it themselves.
Money – like software – creates abstraction. Distance to the friction of the world. The ability to never be touched and to never have to touch anyone or anything. To have nobody else’s life, their existence as a person with needs and rights and hopes and feelings, matter. That probably is why rich folks – and those how aspire to that lifestyle – love “AI” so much.
I don't want the details Why reasonable explanations are one of the easiest ways to avoid changing anything:
If your corrective action depends on people remembering a conversation from six months ago, you don't have a corrective action. You have organizational folklore. If everyone involved in the incident left the company tomorrow, would the fix still work? If the answer is no, the people may have learned something while the system is still destined to fail.
Business Side
AI CEO: "I left the water running, put the stopper in the tub, and then after a while the water went rogue and poured out of the tub."
the reason management is so keen on genAI is that it lets them (maybe) produce productivity improvements without having to consider the fact that many productivity problems are the result of poor management
Machine Intelligence
I really maintain LLMs are like the microwave. There are some narrow specific uses where it's good, but holy shit we're still in the "True Believers trying to make Thanksgiving dinner with them" phase.
Which LLM is worth it Every model on the Artificial Analysis Intelligence Index plotted against its blended API price. Find the LLM that's right for your price bracket.
Everything Else
Book Corners Find neighborhood book exchange spots where you are.
When interviewing for a job, ask about their "yogurt policy" and chuckle derisively at whatever answer they give you. Now you have the upper hand and they don't even have a coherent yogurt policy.
there's nothing as invisible as someone else's problem.
I was not prepared for how much of my job has boiled down to staring at a screen and clicking ‘Allow’ or ‘Accept’
But is it really free? A Browser extension that marks search results as really free, free with a catch, or costs money. Stop wasting time on "free" that isn't.
@pndc Ouch.
Q: How many Open Source maintainers does it take to change a lightbulb?
A: None. Being burnt-out is the new normal.
The best parental control ever invented was a 56k modem and a mum who needed to use the phone.
I just went downstairs, walked three minutes to a boulangerie, and got a warm baguette. Tell me again why 15-minute cities are bad.
Give Me Some Good News The daily news from nonprofits, co-ops, family and community papers, and such. This site avoids chains, hedge-fund papers, and metro dailies.
Our society would be better if we assumed that celebrities probably have takes and ethics in the poor to average range. There’s no good reason to think someone who is good at kicking footballs or writing songs about breakups or acting out characters would be good at thinking clearly about social issues, especially if they spent their formative years obsessing over developing at what they’re famous for
Asked one of my kids today who really loves history to name anyone dead or alive who they thought was a genius.
I was expecting Leonardo Da Vinci, Albert Einstein, Sir Isaac Newton or maybe Alan Turing.
But no. They said Dolly Parton….and then said “not a lot of people know this, but she didn’t have kids, but she loved them all with her Imagination Library”
We then later sat down and watched her Glastonbury set as it happened to be on. We watched her play guitar, banjo, fiddle and saxophone
Ret "one of those days"
