Sign in Subscribe

Topic

Claude Code

A collection of 33 issues

"The draft is ready" is not a status: what an agentic routine owes you when it reports on its own work

A daily scheduled routine carried its own claim that two replies were drafted. Neither draft existed. Why an agent must verify the assertions it holds in state - and the API trap that nearly confirmed the bug.

Claude Code mods vs skills: a skill is a suggestion, a mod is a guarantee — and what I would hard-block first

Mods run inside the Claude Code harness and can refuse a tool call outright, where a skill can only advise. The scoping rule that follows, the four I would actually build, and the security posture they demand.

The daily automation that never calls a model: local scheduled jobs, a receipt printer, and designing for the device being offline

Claude Code built it; nothing in the running loop calls a model. The useful lesson was not the printer - it was that the output device is the flakiest link, and coupling the email to the print cost me both.

I replaced my Loom subscription with a single app I build myself: what Claude Code builds in an afternoon moves the buy-versus-build line

Screen capture, a camera bubble, live transcript, YouTube upload with real subtitles and chapters - all browser APIs, one local file, built with Claude Code. The constraint that looked like a bug became the best feature.

Nineteen days of ad spend with no measurement: the Google Tag Manager snippet was gone from the page, not broken in the console

A purchase conversion sat at zero for nineteen days. The container was fine - the snippet had been removed by a deploy. How to pin the hour without backend access, why you must tell Smart Bidding its data is wrong, and the fourteen-day cap that makes fast detection worth money.

I built a client performance dashboard in a day with Claude Code — the hard part was not calling enquiries revenue

A lead-generation business that takes no payment online. Four APIs disagreed with their own documentation, but the decision that mattered was what to call the biggest number on the page.

My scheduled Claude Code routine failed three weeks running: what agentic workflows need that a written lesson doesn't give them

An AI automation that couldn't unblock itself, scheduled in a context where the one thing it needed was impossible. Why agents must know what their execution environment can do — and why a well-written lesson with no owner changes nothing.

The audit found eleven missing receipts. Nine of them existed.

An automated gap report compared bank lines against an index that only knew one of several folders. The fix wasn't a smarter matcher — it was an index that could prove it saw everything.

OpenAI built a whole browser for the agent. The agent didn't need one.

Claude Cowork just got a built-in browser, seventeen days after OpenAI switched off Atlas. The improvement isn't the browser — it's that there are finally two modes, and knowing which job belongs in which one.

I scheduled a weekly check that could never run.

An automated SEO check that needed a browser on my laptop, scheduled in the cloud. Designing a measurement window that can honestly report nothing was the real work.

Work with me →×