Architecture, APIs and implementation detail. The decisions that decide whether a system survives its first year in production.
Engineering
4 min read
Security firm Forever stopped an AI coding agent from installing a malicious npm package on a customer's machine. The instruction came from an AGENTS.md file sitting in the Downloads folder. How it worked, what the researchers' lab tests showed for Codex and Claude Code, and six checks for anyone who runs a coding agent.
Engineering
4 min read
OpenCode, an open-source AI coding agent, disclosed a breach on 18 October affecting users who had linked Anthropic, OpenAI or Google accounts to its console. What is known, why a linked AI key is worth stealing, and five steps for any business that has pasted an AI key into a tool.
Engineering
5 min read
Glow Labs found more than 13,000 internal screenshots from 300+ organisations in public GitHub repositories, put there by AI coding agents solving a small workflow problem. How it happened, and what to check if anyone builds software for you with an agent.
Engineering
5 min read
ESET scanned almost 900,000 AI agent skills and found more than 3,000 malicious ones, built for credential theft, data exfiltration and remote code execution. What a skill really is, and a vetting checklist a small team can use.
Engineering
4 min read
When a retrieval system returns a confident wrong answer, the instinct is to blame the model or rewrite the prompt. Usually the model answered correctly from bad context. Four things fix more wrong answers than any prompt change.
Engineering
3 min read
Most SEO bugs are two files disagreeing. DueClix Insights has one typed object per article and derives the title, meta, canonical, Open Graph, JSON-LD, sitemap and RSS from it — no CMS, no MDX, no new dependencies.
Engineering
3 min read
How we migrated a business directory where most listings were spam: keep everything as reversible drafts, reject the heuristic that would have hidden real companies, and learn from the rule that missed ALL-CAPS junk.
Let's build the system behind it. Tell us what the process is and where it breaks — the first conversation is about constraints, not technology.