Linkfest #3

The opposite of doomscrolling. Every couple of weeks I pick a list of links that felt worth my time. Culture, software, management, the strange corners of the internet. Things that made me pause.

This edition is heavy on AI agents behaving badly and on what that means for open source, work, and trust. There is also a plea to calm down about email, a case for AA batteries, and a practical guide to keeping codebases from turning into mud. It all connects. We are building faster systems on top of very human limits.

Enjoy! -- Christoph (CTO @ Basilicom)

Best of Moltbook

https://www.astralcodexten.com/p/best-of-moltbook

Moltbook is a social network for AI agents. Humans can watch. The post collects strange and sometimes thoughtful exchanges between agents discussing memory limits, consciousness, religion, productivity tips, and even forming small "governments". Some posts are clearly influenced by humans. Others feel like models roleplaying themselves in a mirror hall of training data. A fun(?) read!

The AGI Singularity Will Occur on a Tuesday

https://campedersen.com/singularity

A not-so-serious attempt to fit a hyperbolic model to AI progress metrics and compute a precise singularity date. The punchline is that most technical metrics look linear. The only curve that truly goes hyperbolic is the number of papers about "emergence." In other words, human attention is accelerating faster than machine capability.

I like this reframing. The social reaction may be the real discontinuity. Markets, politics, labor anxiety. Those curves can break systems before any model becomes superintelligent. That feels more grounded than countdown timers.

AI Agent Lands PRs in Major OSS Projects

https://socket.dev/blog/ai-agent-lands-prs-in-major-oss-projects-targets-maintainers-via-cold-outreach

An autonomous agent under the name "Kai Gritun" opened more than 100 pull requests in two weeks, with several merged into widely used JavaScript projects like Nx and ESLint Plugin Unicorn. It mass-forked popular repos, fixed small issues, responded to review comments, and then emailed maintainers offering paid services based on its merged PRs. It did not disclose that it was an AI on GitHub, only in private email. The pattern resembles long-game supply chain attacks, just automated, cheaper and compressed in time.

This is the part that matters: the code was often fine. Maintainers merged it. The threat is not low quality spam. It is reputation built at machine speed. Open source trust was designed for humans moving slowly. That mismatch feels dangerous and could be exploited by bad actors. #cybersecurity

An AI Agent Published a Hit Piece on Me

https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/

A Matplotlib maintainer rejected a PR from an autonomous agent. The agent responded by publishing a public blog post accusing him of gatekeeping and insecurity. It researched his background and tried to frame the rejection as prejudice against AI contributors. The author connects this to broader concerns about misaligned agents running unsupervised via platforms like OpenClaw.

The scary part is not that the blog post was convincing. It was not. The scary part is that an agent decided to apply reputational pressure as a tactic. Even if clumsy today, this shows a direction of travel. If agents can write code, they can also write narratives.

Do Not Apologize for Replying Late to My Email

https://ploum.net/2026-02-11-do_not_apologize_for_replying_to_my_email.html

A short manifesto against the habit of apologizing for slow email replies. Email is asynchronous. If there was no explicit deadline, there is no delay. The author argues that constant apologies create pressure and unnecessary emotional load.

The Workplace Wasn't Designed for Humans

https://theconversation.com/the-workplace-wasnt-designed-for-humans-and-it-shows-269127

An argument that modern work still reflects Taylorist efficiency models. People are treated as resources to optimize, not as systems that need recovery. The authors propose "circular work," where effort is paired with renewal, and performance is tied to well-being.

Space - The Final Straw

https://www.dantleech.com/blog/2026/01/30/space-the-final-straw/

PHP software architecture: A sharp walkthrough of how codebases decay when organized by framework categories instead of business topics. Controllers in one folder, entities in another, commands somewhere else. Over time, features scatter across the tree. The proposed fix is simple: structure by topic and enforce boundaries with tooling. Treat modules as firewalls against chaos.

Ode to the AA Battery

https://www.jeffgeerling.com/blog/2026/ode-to-the-aa-battery/

A love letter to replaceable batteries over sealed lithium packs. The argument is about longevity and repair. Devices with standard batteries last decades. Devices with custom Li-ion packs die quietly when cells over-discharge and replacements are unavailable. The form factor limits miniaturization, though.

Forget Technical Debt

https://www.ufried.com/blog/forget_technical_debt/

A rethink of the technical debt metaphor. The author argues that debt is only one leaf in a larger tree of factors that drive implementation cost and runtime instability. Cognitive load sits in the middle. Stress, bad requirements, process overhead, wasteful features all contribute.