The prompts worth keeping in a file
Most prompt libraries are padding. The dozen we reuse weekly, why they survived, and how to maintain a library that does not rot.
Working notes on AI tools, no-code platforms and automation — what they do, what they cost in time as well as money, and which ones survive contact with real work.
For several common AI tasks we priced the subscription against the raw API bill. The crossover point is lower than most people assume.
Most prompt libraries are padding. The dozen we reuse weekly, why they survived, and how to maintain a library that does not rot.
Not every repetitive task deserves a workflow. The volume, stakes and stability thresholds below which automating costs you more.
The unglamorous automations that repay the effort every week, with the trigger, the steps and the failure mode for each.
We ran human and machine text through five detectors. False positives were common enough to make any policy built on them unfair.
Twenty tools whose free plans we ran for a month. Which ones do real work indefinitely, and which expire the moment you rely on them.
Every automation platform added an agent step. Where a model in the middle of a workflow helps, and where it makes failure unpredictable.
What a generated one-page site got right at speed, and the three judgements no generator can make for you. Notes from a fortnight of use.
n8n on a small VPS for a fortnight. Where self-hosting paid off, and the maintenance work the pricing comparison never mentions.
Some generators let the model produce everything; others let it fill slots in a fixed structure. The choice decides how often output breaks.
Eleven AI subscriptions, three months of usage data, and the awkward finding that four tools did almost all of the work.
Credits are a unit vendors define and can redefine. How to convert a credit allowance into a real monthly figure before you commit.
Timed, step by step: from a CV and a photo to a live one-page site, including the parts the marketing pages leave out.