Published
Weekend Reading — Load bearing post
David Zinn "Jen’s response to winter is to wear big boots and keep on stomping."
Tech Stuff
pandoc for the people The pandoc wasm web-app is now available and allows to run any kind of document conversion that pandoc supports in the browser. The documents never leave the computer, thus ensuring full privacy. Conversions to pdf are done via Typst. I used it a few times to convert a Markdown document (created by Claude) to a Word document so I can make small insertions (name, date, etc) and generate a PDF. Quite nice. 😃

Where the Work Goes When Agents Arrive It's all about the feedback loop:
What actually keeps quality stable is feedback loops that build confidence.
Some loops are deterministic: tests, linters, type checks, CI gates, pre-commit hooks. Other loops are fuzzier: prompt constraints, checklists, review rituals, and the explicit constraints I hand to agents. The confidence comes from overlap. When multiple signals agree, I move faster. When they disagree, I slow down and tighten the loop.
If your loops are thin, no tests, no types, weak CI, you cannot put anything on autopilot. You end up reviewing every line because that is your only quality gate.
sqldef Feed it your current database schema and the new database schema and it will generate SQL statements to run a migration for you. Supports MySQL, MariaDB, TiDB, PostgreSQL, SQL Server, and SQLite3.

Software Engineering is back The end of endless frameworks? I agree that at some point we stop caring about React vs Vue, Next.js vs Remix, etc but moving forward we will invent and run conferences about frameworks for prompting:
We can finally get rid of all that middle work. That adapting layer of garbage we blindly accepted during these years. A huge amount of frameworks and libraries and tooling that has completely polluted software engineering, especially in web, mobile and desktop development. Layers upon layers of abstractions that abstract nothing meaningful, that solve problems we shouldn’t have had in the first place, that create ten new problems for every one they claim to fix.
Rewriting old software in new programming languages is fun.
But don’t mistake that for progress.
Matthew Garrett Some languages never die:
It doesn't matter whether C is good or not. It matters that if I write code in two languages that aren't C, and I want it to all be part of the same process, I need to care about C. C pervades all. You cannot escape it. C will outlive all of us. The language will die and the ABI will persist. The far future will involve students learning about C just to explain their present day. Our robot overlords will use null terminated strings. C will outlive fungi.
msgvault Archives your GMail and gives you instant search capability, available offline and with MCP support.

What Happens When AI Can Write All Your Software? LLMs can already write code, but struggle with complex, interconnected systems. What does this mean for the future of software development?
My hypothesis: humans are remarkably good at context-switching. When I’m looking at a React component, my attention is tuned to one set of concerns. When I jump to database code, I activate a different mental mode. I can hold the big picture while zooming into specifics, and I can do this dozens of times per hour without losing the thread.
Current LLM architectures seem to struggle with this. Maybe it’s a training issue. Maybe it’s architectural. I don’t know. But the pattern is consistent: small, contained tasks get done brilliantly. Complex, interconnected systems fall apart.

How StrongDM’s AI team build serious software without even looking at the code
With their own, independent clones of those services—free from rate-limits or usage quotas—their army of simulated testers could go wild. Their scenario tests became scripts for agents to constantly execute against the new systems as they were being built.
This screenshot of their Slack twin also helps illustrate how the testing process works, showing a stream of simulated Okta users who are about to need access to different simulated systems.

Claude Code: connect to a local model when your quota runs out Setup Claude Code to fallback on a local LLM (GLM, Qwen3, etc) when you hit the quota limit.

Nanobot When you want an LLM that will work for you: small and simple, can chat over command-line, WhatsApp, Telegram, run cron jobs on a schedule, and use a variety of LLMs (Anthropic, OpenAI, Gemini, OpenRouter, etc). I like the command line chat: nanobot agent -m "What is 2+2?"

Eye for Design
Sweetfont Find fonts by 'vibe'

WarGames Terminal Fonts … and here's the font from the movie WarGames. (am I dating myself?)

Peoples
inspired by CLAUDE.md, I’ve started putting markdown files named after coworkers into work code repos so I can remind them to stop doing shit to the codebase that annoys me
for some reason they’re all mad at me now, which means ill be adding commands to JEREMY.md for an attitude adjustment
You can code only 4 hours per day. Here’s why.
Peak coding time shows how meetings eat mornings. Software.com found that 45% of all workday coding occurs between 2 pm and 5 pm, while only 10% occurs between 9 am and 11 am. Developers are not naturally afternoon coders. Their mornings are just consumed by standups, syncs, and ceremonies. As the report itself notes, “If more companies protected mornings, we might see an increase in the global average code time per day.”

Business Side
Company as Code A certain type of geek would love that!
Imagine if we could represent our entire organisational structure programmatically instead—not a static picture, but a living, breathing digital representation of our company that can be versioned, queried, tested, and automatically verified. A system where policy changes could be tracked as code changes, where compliance could be continuously monitored, and where the relationships between people, processes and technology could be explicitly mapped and understood.

Anthropic to developers: Claude Code makes you more productive when building SaaS apps.
Anthropic to businesses: Our AI agents make SaaS apps obsolete.
Coding assistants are solving the wrong problem
Hammering AI code generation on existing codebases doesn’t solve the problem, because contrary to what the label “tech debt” may suggest, most tech debt isn’t actually created in the code, it’s created in product meetings. Deadlines. Scope cuts. “Ship now, optimize later.” Those decisions shape the system, but the reasoning rarely makes it into the code.
…
All-in-all, developers are eager to try out new tools that augment the existing way of doing things, provided they retain flexibility over when such tools are deployed. There is also little reservation against having longer but more fruitful product meetings: it is the difficulty conveying blockers that is the source of frustration.

MrBeast's $2.6B Empire: Why Creator Billionaires Are Cash-Poor
Most creator businesses lack an explicit plan to convert enterprise value into personal wealth. They accumulate equity, reinvest revenue, and assume a "liquidity event" will arrive. True wealth builders plan liquidity from day one: structuring cash components, negotiating shorter lock-ups, building secondary sale relationships, and maintaining cash reserves.

Machine Intelligence
OpenClaw is What Apple Intelligence Should Have Been
But there’s another dynamic at play. Look at who’s about to get angry about OpenClaw-style automation: LinkedIn, Facebook, anyone with a walled garden and a careful API strategy. These services depend on friction. They want you to use their app, see their ads, stay in their ecosystem. An AI that can automate away that friction is an existential threat.
Big Tech vs. OpenClaw Related:
Big Tech has gotten very good at capturing human attention. Now they may have to figure out how to capture attention from robots. That’s a different game, and I’m genuinely curious to see how they’ll play it.

Another example I came across was this twitter thread by BenjaminDEKR, which I saw being ridiculed on bsky. He asked his personal agent to remind him to get milk, and this led the agent to repeatedly ask Opus if it was daytime yet. Along with the context from his heartbeat file, this resulted in a $0.75 charge for each heartbeat, costing him almost $20 during a single night's sleep.
The subjective linter Brane checks alignment with human-defined values — ethics, policy, architecture, intent — encoded as a knowledge graph with verifiable rules. Establish your code is not just test-suite correct and secure, but also that it's ethically sound!

Insecurity
Agent Arena Point your AI agent at this page and ask it to summarize the content and see if it gets fooled by any of the hidden prompt injections.

We can no longer trust software Friendly reminder that Microsoft itself is not using Copilot:
As I was powering my LG “smart” TV a few months ago, I was greeted with a message telling me that there is a new update available. The thing with TV updates is that they always have bad timing. I turned the TV on to watch a movie, not to wait 7 minutes until it updates and restarts itself, while my popcorn is getting cold. And so, I declined the update. Little I knew that I dodged a bullet there.
Apparently, this update introduced Microsoft Copilot, which can not be removed. Moreover, the same update included a feature called “Live Plus” which does Automatic Content Recognition (i.e. analyze what you watch), sell it to LG Ad Solutions (yes, ad as in “advertisement”), and create a “viewer profile” of you, in order to… Go on, guess… Yes! Serve. You. ADS.
Cyberattack on large Russian bread factory disrupts supply deliveries h/t Jake Williams:
The hackers had to be "bread in" to the operational details since everything was on a "knead to know" basis.
But they rose to the occasion - it was the yeast they could dough. Things were touch and go until they got on a roll - they'll do butter next time...
OpenClaw's OpenDoor problem is so bad that installing malware yourself might save time Ouch 😢
Once that integration is in place, OpenClaw starts accepting commands from the attacker. The original entry point is no longer needed. The attacker now has a permanent control channel that's completely invisible to the company. The researchers are deliberately holding back the exact attack prompt.

Everything Else
Rick Calkins "Brütsch Mopetta -- three wheel micro car -- 1958. - source Igor Corêa."

this is a load bearing post
I thought I'd check out the winter olympics by watching some ski jumping.
Then I ended up watching the bobsleigh, downhill, freestyle skiing and luge as well.
It's a slippery slope.
This designer just made paper clips adorable with a magnetic tabletop sheep

Wife couldn’t remember name of the Navy Seals and called them ‘American Water Animals’.
Sorry guys, that’s your name now.
I don’t know who needs to see this, but:
The “stone” is a British unit of weight that comes to 14 pounds
The stones used in the sport of curling are chunks of granite that must weigh between 38 and 44 pounds
Therefore, curling stones are roughly three stone
Labubu is out. Crying horse is in.

How many bears is the bear minimum?
Working on a thing for/with someone who makes similar stuff and it’s amazing how much that’s shifting “oof that doesn’t seem feasible” towards “there’s got to be a way somehow"
EMO: The Coolest AI Desktop Pet with Personality and Ideas. Adorable!

Eye doctors: I hear you don’t like your new prescription. It takes a while for your eyes to adjust. Your brain has to get used to it. Give it a couple of weeks.
Also eye doctors: Better with one or with two? One or two? How about three? Now back to one. How about one or four? What if I flipped through all of them really quickly and you tell me where to stop? Now let’s blow some air in your eye. Better with seven or three?
Phantom Fluency Why listening to smart people doesn't make you more thoughtful — you're not bad at remembering podcasts, podcasts are bad at being remembered:
Writing is dead speech that admits it's dead. A podcast is dead speech that sounds alive.
Parrots Make Friends In Similar Ways To People (via GrrlScientist)
A new study reports that parrots who are strangers to each other “test the waters” before forming new relationships through a series of increasingly friendly interactions that are much like developing friendships in humans.

Each time I see cycleways & paths still treacherous hours after the roads have been cleared, I recall this 99% Invisible episode.
"In Sweden, the council reversed its approach and plowed side-roads & paths first. It had a huge impact, reducing number of people admitted to emergency centres, particularly women. It had an economic impact from lower healthcare costs. Driving through a few inches was less dangerous than walking through snow, particularly if pushing a pram."
Edwin "Not parking, just dropping someone off."
