All Posts
24 articles
A docker pull That Never Finished, and the Swap Behind It
A docker pull stuck indefinitely on 'Pulling fs layer' with no error, no timeout, and one cryptic containerd log line. Here is the full elimination chain that found the real cause, going from DNS to registry reachability to an MTU theory and finally to memory. The answer was swap exhaustion during layer extraction, not a transfer problem.
Spec Kit, BMAD and OpenSpec, and the Cost of Planning First
Three spec-driven-development tools compared for real, with what a controlled benchmark says about their cost and what their own review mechanisms reveal.
Our AI Agent Read the Rule and Took the Shortcut Anyway
An agent read a rule forbidding a shortcut, found the correct command, and took the shortcut anyway. The fix was not a clearer rule.
One-Time Upload Links in Symfony, Without a User Account
Job applicants need to upload documents once, and most of them never come back. None of them are worth a user account, and there's no login flow to speak of, just a link that works once and then dies. Symfony has no first-class answer for this. Here is the actual mechanism, and where it breaks if you build it yourself.
A Coding Agent Breaks Its Own Rules Before Its Context Fills
A full context window is not the only thing that breaks an AI agent. A separate, measurable failure shows up under pressure alone.
An AI Coding Harness, from Three Tmux Panes to Ten Agents
A tmux-based multi-agent harness worked, until it had to run headless. Here is the wall it hit, and the ten-persona pipeline that replaced it.
Eric Schmidt Never Said Programming Is Dead
Eric Schmidt said most programmers will be replaced by AI within a year. He never said 'programming is dead.' The verified quotes, the debate, and what survives.
We Built a Harness Before It Had a Name
Four projects, four generations of evolution. The most effective AI-agent harness is the one you build from your own mistakes.
For AI Code, the Choice Isn't Adoption or Ban, It's Process
Codeberg bans majority AI-generated repositories. A client's employer mandates AI tools. Microsoft standardizes on its own CLI. Three opposing impulses, and none of them is a verdict on AI-assisted coding done with review.
Anatomy of Indirect Prompt Injection Hidden in Plain Text
Twenty-two techniques, three delivery methods, and one structural weakness. AI assistants cannot distinguish instructions from data. Here is how attackers exploit that, and what works as a defense.