Vibe coding security: what to check in an app built with AI
Six checks the owner of a Lovable, Bolt, Replit, Cursor or Base44 app can run without reading code, each tied to a public incident and a primary source.
vibe coding security
Read // articles
Writing from the work.
Longer reads on building things, making decisions and working with developers.
This archive is for work with something real behind it: a build, an experiment, a failure, or a position earned by doing the work. Sources stay close to factual claims, and uncertainty stays visible.
Looking for experiments and evidence from actual builds? Read the field notes.
Questions to ask a web developer before hiring, and checkable answers
Six questions for a business hiring a web developer, the difference between an answer that is a claim and one that can be checked, and who owns commissioned code.
hiring websites
Read Claude Code subagents: what they isolate, and what they share
What a Claude Code subagent is, what it can and can't see, and why its requests draw on the same usage limit as the main session, with a dated record of 37 subagents in one day.
claude code ai tools
Read Is ChatGPT safe for client data? What regulators and OpenAI say
What legal and accounting regulators in Britain, the US and Europe say about client data in AI tools, set against OpenAI's own terms for each ChatGPT plan, read 3 October 2026.
private ai confidentiality
Read Claude Code vs Codex vs Cursor: the AI subscriptions I pay for
What I pay for in September 2026, what each AI subscription is actually for in my work, and why I don't pay API prices.
ai tools pricing
Read Hiring someone to build an AI system: ‘done’ is a claim
What published measurements show about AI coding agents claiming work they didn't do, and how one lab keeps the claim apart from the check.
hiring ai systems
Read LLM benchmarks: the benchmark was wrong before the models were
What an AI coding benchmark score measures, how the public benchmarks corrected their own tests, and twelve times one lab caught its own instrument wrong.
benchmarks verification
Read Is vibe coding bad? A viral maintenance statistic, traced
A viral post put a number on the maintenance cost of AI-written code. Traced to its sources, and set beside what research actually measured.
ai coding sources
Read CLAUDE.md vs AGENTS.md: four layers of instructions, and where they drifted
Where Claude Code loads CLAUDE.md and AGENTS.md, why both are context rather than configuration, and one rule that gave two agents opposite answers.
claude code instructions
Read Claude Code hooks: a write guard, and the folder where it never ran
Where Claude Code loads CLAUDE.md and hooks, the subfolder where a write guard never ran, and what the guard still cannot see.
claude code verification
Read Got a question these don't answer? Send it. If it comes up twice, it becomes the next one.
As a guide, a website rebuild starts at €1,250 and a 3D website at €3,500. Software, automation and AI are scoped and priced on a free intro call. Every project is priced on its own complexity and the work actually in it. See the offers