Published
Weekend Reading — Sunlight with ads
ConnectionWithWonder "Knitted bollard-warmer Weymouth"
Tech Stuff
UNF* Records every version of a file and can easily rewind back without having to manually commit changes. Perfect if you're working with AI agents and don't trust them. Also perfect if you're just prone to accidentaly deleting or modifying files. I messed so many of my files, so I'm giving UNF a try. 🙏
pcvelz/superpowers I'm using this plugin to write code, and it's surprisingly good. It might take 30 minutes or 2 hours to finish a run, but it does all the planning, asks for clarifying questions, creates a roadmap, saved as Markdown file, and then launches several agents to complete these tasks. A modification of obra/superpowers, designed specifically for Code Claude. ⭐⭐⭐⭐⭐
Two weeks later with OpenClaw and this is what I’ve learned Various tips & tricks for using OpenClaw successfully, many of which also apply to other claws, for example:
Setup: Do This or Regret It
1. Start with the TUI, Not the Web UI
The terminal interface shows you exactly what the agent sees during setup. Way less debugging than the browser dashboard.
2. One Channel, Perfected
Connect WhatsApp OR Telegram OR Discord—not all three. Debug one integration completely before expanding.
MarkViewer Your LLM is working primarily with Markdown files, which is why we're seeing an explosion of people using Markdown. This simple (and free) app offers a Markdown preview and plain WYSIWYG editor.
Marked 3 Related, quite the shophisticated Markdown preview app, including live preview, advanced proofreading, export to PDF/HTML/DOCX, and more. Version 3 is currently in beta and launching soon.
Writing code is cheap now - Agentic Engineering Patterns 💪
For now I think the best we can do is to second guess ourselves: any time our instinct says "don't build that, it's not worth the time" fire off a prompt anyway, in an asynchronous agent session where the worst that can happen is you check ten minutes later and find that it wasn't worth the tokens.
ast-grep What if you could search and replace using the programming language's own syntax? ast-grep -p '$A && $A()' -r '$A?.()'
difftastic A smarter diff that makes it easier to see the key differences. You can easily configure git to use this as the default diff.
Don't trust AI agents From NanoClaw which wants to be OpenClaw but with a more secure architecture, and they do make some valid points for any codebase you're working on:
OpenClaw has nearly half a million lines of code, 53 config files, and over 70 dependencies. This breaks the basic premise of open source security. Chromium has 35+ million lines, but you trust Google’s review processes. Most open source projects work the other way: they stay small enough that many eyes can actually review them. Nobody has reviewed OpenClaw’s 400,000 lines. It was written in weeks with no proper review process. Complexity is where vulnerabilities hide, and Microsoft’s analysis confirmed this: OpenClaw’s risks could emerge through normal API calls, because no one person could see the full picture.
Little Libraries Tiny JavaScript libraries the author tries not to update. They're mostly under 1 kB and available directly on this website, no package manager required. (via Caolan McMahon)
Shuru macOS local-first sandbox for AI agents.
Building an AI Blog Editor with Claude Skills One of the more interesting features of Code Claude is how easily you can extend it without any programming experience, just write a SKILL.md file:
I've been using Claude Code for development work and found it significantly better than other AI coding assistants. So when I learned about Claude Skills, I wondered: could someone have built a skill to make Claude function as a technical blog post editor? I searched and found several skills for blog post writing, but nothing for editing. So I decided to build one myself.
Eye for Design
I once told a technically illiterate and very aggravating graphic designer in our marketing department that she should stop using large fonts because they were filling up her hard drive and we didn't have any suitable replacements. Sorry, not sorry.
Font Review Journal A great source of lovely fonts and reviews packed with helpful knowledge. (via Piccalilli)
Peoples
After Vibe Coding: What Developers Who Succeed With AI Do Differently Developers who succeed stop executing steps and start managing feedback loops:
Developers succeeding with AI treat code as a tool for revealing truth. They write code to:
- Test assumptions
- Observe real behavior
- Expose edge cases
- Learn where their mental model is wrong
This shifts the goal from perfect code to clear understanding.
Child’s Play Related, are you … "highly agentic"?
The future will belong to people with a very specific combination of personality traits and psychosexual neuroses. An AI might be able to code faster than you, but there is one advantage that humans still have. It’s called agency, or being highly agentic. The highly agentic are people who just do things.
Machine Intelligence
Miss One Weekend, Fall Behind One Month 🤦♂️
My reaction? The same as every time: excitement mixed with mild panic. I haven’t even finished learning
Claude Codeproperly (out of 5 levels1 I am at level 2) and now there are agent teams? I was supposed to tryCodexlast week, but I was busy debugging a YAML file like it’s 2022.If you work in or around AI, you know this feeling. This post is for you.
I use this simple repetition trick with ChatGPT when it misses the point — and it works every time Or said another way, the easiest AI upgrade is saying the same thing twice:
After testing multiple models, researchers found that repeating a prompt improved accuracy and quality without affecting response length or latency. The improvements weren’t tiny either. For many tasks, the models produced stronger reasoning and steadier detail on the second or third repetition.
What Claude Code Actually Chooses DevRel circa 2026: make sure LLM appreciates your project, as newer models tend to pick newer tools.
Let that sink in. She had working software that proved the concept, and the response was “nah, that’s not how we do things.” The objection wasn’t technical. It was psychological. People had mastered assembly language and machine code. That was “real programming.” A compiler was cheating. It was a crutch. It would produce inferior output. The machine would never be as good as a human at writing machine code.
Following the launch of Claude Code, US GitHub code pushes were 30% above the pre-2025 trend.
Similarly iOS app releases were up 55% vs last January, and new website registrations rose 34% year over year globally after years of flatlining.
AI productivity gains have hit software in a major way.
Everything Else
InsiderTreat "I had this same patch in flight school. My classmate sitting behind me didn't always find it as funny as I did."
the cross-platform social media protocol is called screenshots
Can I go back to the "sunlight with ads" package? I'm only using about 4 hours a day, hardly seems worth it.
Wen "A brilliant bit of snow sculpture, Central Park New York"
It's probably a bad thing that the "Turing test" entered popular culture but the "ELIZA effect" didn't.
Scientists have detected another possible Monday arriving as early as next week. If true, this will be the 9th Monday this year. They warn that there are a little over 3 days left to prepare.
Mondays can be bad. Hug those you love and help them through it. Spend these remaining days wisely.
2026 Winter Olympics: The most striking photos from the Games
In corporate advisories, you don't say "I love you." You say "We take our users privacy seriously," which means "we were storing your social security number in plaintext." I think that's beautiful.
the same people demanding you get back to the office to better collaborate in person as only humans can are the same people desperate to replace everyone with some code that runs on a server in ohio
Aphrodite "I have a hunch @[email protected] would like this video from Norway."
Won Gold Medal in Taco Eating 👿
Sometimes when I pass by a stranger I like to whisper "I was just thinking the exact same thing."
I'm just a girl, incrementing the counter on the number of times I have been sent a plaintext email from a Protonmail user telling me that the message is encrypted.
This teddy bear waved to me this afternoon, as I passed its barge while walking along the tow path. No really. The owner has a rope set up that runs from the arm of the teddy bear at the bow to the stern where he was stood steering. It was kinda lovely and heartwarming 🙂💕
Why Your Brain Becomes Unreliable When You Need It Most Also, why your LLM coworker is making smarter decisions than your human co-worker:
Your brain has two modes. One is slow, thoughtful, reads the fine print, weighs the options.
The other is fast, impulsive, and mostly just trying to get through the day without spending too much energy.
Guess which one makes most of your decisions?
The Convenience Equation aka The inverse coefficiency of enshittification:
When something is incredibly convenient—effortless, seamless, always available—we become willing to overlook consequences we'd normally consider unacceptable. We accept terms we don't read. We hand over data we wouldn't normally share. We lock ourselves into ecosystems we know reduce our freedom. We do all of this because the convenience is genuinely good, and the risks feel abstract. This is the phenomenon that behavioral economists call "inattention"—when convenience is high enough, people often fail to fully evaluate the costs and risks involved.
Magnetic Modular Fidget Pen You know you want one!
The tram conductor has a special button on the ticket machine that prints a ticket with a big smiley face on, for the under 5’s who don’t need a ticket, but want one.
Velocity Is the New Authority. Here’s Why A culture optimized for first takes, not best takes:
The system rewards whoever speaks first, not whoever lives with it long enough to understand it. The “review” at launch outperforms the review written two months later by orders of magnitude. The second, longer, more in-depth, more honest review might as well not exist. It’s not that people are less honest by nature. It’s that the structure pays a premium for compliance and levies a tax on independence. The result is a soft capture where creators don’t have to be told what to say. The incentives do the talking.
This seems like a good time for a reminder that when COVID hit in 2020 and everyone was living via Zoom Studio Ghibli released a bunch of images from their movies for free use as virtual meeting backgrounds, and those backgrounds are still up and free for you to use
