Diana ran her own Website Builder Skill in an AI app inside her site files and asked for an idea finder for her article listing 25 ways to use AI. The AI read the article and her live stylesheet, worked out her colours and fonts on its own, placed the finder right after the article links, and kept all 25 ideas searchable and filterable. It tested the result in a headless browser against her live styles and found no style changes across the 453 elements on the page.
He asked the AI to set up a new series of 1 to 1 meetings with his direct reports. The AI created sixteen calendar invites, all to the same team member.
This one went wrong. See what happened before you try it.
Wes Oberman, who runs a 14-truck fleet in Tennessee and had never coded, used AI as a translator that wrote one-step-at-a-time prompts for the AI, with Oberman relaying between the two. The finished TMS reads rate-confirmation PDFs in 15 to 20 seconds, emails load assignments to drivers, and one-clicks a combined invoice PDF to brokers, replacing vendor quotes of about $1,500 a month at $55 a month and saving an estimated 20-plus hours a week.
Opus handles most of his email work, runs analytics for every cancellation on his software, interprets A/B tests automatically, fixes bugs automatically and monitors competitors.
He is not a developer; he is a management trainee at a company made up of five entities. Every quarter, full financials go out to the management team: a batch of at least 25 PDFs from their SAGE accounting software, several of them over 20 pages long. He spent weeks reviewing test files and batches and debugging formatting mistakes to build his own converter, after finding that the AI models themselves could not do the job faithfully on larger documents: they would combine rows and categories or produce totals that added up but were not identical. The AI itself told him the task would need a hard coded deterministic app. His converter can now batch format the whole set at once, with the bolding, blank rows, sum lines and column sizing he is picky about.
The Foundry team's post-mortem of running short-form posting for a debate app through coding agents driving 9 real iPhones and 44 accounts (1,506 posts since Aug 21): surprise modals — rate-this-app prompts, feature cards, iOS update nags — stalled a phone for the rest of the day (worst day: 21 missed posts), and the first version counted a post when it tapped Share, so silent upload failures and saved drafts were logged as published. Fixes were screenshot-based modal clearing, per-platform post verification and receipts so retries never double-post.
This only partly worked. Worth knowing before you try it.
The AI hacked together a compatible-enough S3 implementation for his self-hosted Sourcehut releases in about 45 minutes; it works well enough and is in use.
Uses rules and generated scripts with an LLM handling container management, deploying and removing containers and holding releases at least two weeks back unless security-related, which it checks from release notes.