A random mix from the whole library. Shuffle for a new set.
At natural gas company Williams, one operations technician built an anomaly detector with AI Cowork. It pulls equipment data reports from an Outlook distribution list, compares them against baseline statistics, and flags signs like rising thermocouple temperatures, falling oil pressure or increasing vibration.
He set the AI loose on entry level machine learning and data roles with hard rules: roles posted in the last 48 hours, pull the job description and required skills, rank matches against his projects, fill out applications, and log every job and status in a spreadsheet. It handled search to form filling and remembered his rules on visas and salary. But search results were full of expired postings and dead links, it made form mistakes, logins and captchas still fell to him, and once it submitted an application he had not approved. He warns against letting it rewrite a resume freely, since it adds things he cannot back up.
During the Apple event she had the AI check the live transcript every five minutes against her client roster. It flagged a Siri announcement as a fit for her voice AI client, and she got pitches out right away.
Food creator Careese (someone) asked her AI avatar 'kwin' to solve the pain of manual cross-posting. The agent found her top 20 reels, checked for duplicates, asked permission, then uploaded all 20 with captions and titles unchanged.
He edits video for a living. His agents handle the non creative manual work around each project: quality checks for missed text, black frames and logos, subtitles, reformatting for social aspect ratios, spreading files to the right products and pulling source material for descriptions. He logs his hours on every project and says agent assisted projects take about half the time with better quality; the weeks spent building the setup paid for themselves in two weeks.
He stopped using Descript and now cuts his YouTube shorts with AI. He says the quality of the shorts has gone up since the switch.
Buried in a thread about application exhaustion, a job seeker describes the workspace he built in an AI app: it searches 300+ job sources, pulls leads and fit-scores each against his experience, preferences, salary requirements and commute; for any lead worth pursuing it breaks down the posting and the organization, compares it with his background, suggests resume adjustments, creates a customized cover letter and drafts the application email - he reviews, attaches documents and submits.
He asked the AI to make a technical diagram with gpt-image-1. After ten tries the version that shipped looked clean but had quietly dropped the exact claim the diagram was meant to show, and he only caught it after publishing.
He gave the AI his CV and a job search brief, and the agent searched Naukri, LinkedIn, and Wellfound for roles posted in the last 24 hours, scored each one against his experience, and tracked the strongest in a Google Sheet. For the top 3 it made a tailored summary and cover note in a Google Doc and drafted an intro email for each, leaving sending and applying to him.
AI agents rewrote a production Node service handling 92M messages a day into Go (Checkly company blog, result stated in the title itself).
He set up an AI for his WP Beacon project to check WordPress plugins every day. After two weeks of daily checks it pinged him about malware in the InstaWP Connect plugin, and wordpress.org closed the listing that afternoon and restored it the next morning.
He runs an online marketplace in Japan that he started in 2004, with around 300,000 members, and he is clear the AI gets no credit for that. What changed is how much else he ships: since working this way he has released four Mac apps and one web product on his own, some with paying users. The AI handles the release scripts, localization into thirty languages, measurement and support drafts, the pile of work that used to stop him shipping the next thing.
Had the AI implement a JavaCard VM from spec and then a proof generator from spec, both statically verified with ESBMC which found bugs; a verifier refactor has been stuck for over a week.
He typed one line into AI asking for a screen recording that shows how to add up numbers in Google Sheets, for his dad who struggles with computers. The AI recorded the demo, did the typing and clicking itself, and exported it ready to send.
A video summary was made for others to use in their teaching. The AI wrote Python code that rendered about 9,000 animation frames, the poster recorded the audio, and the finished movie takes about 30 minutes to draw on a 1,000 dollar MacBook Air.
The AI planned, then wrote a finished Google Sheet into the tester's Drive from one named source file, with the requested columns, sort order, summary row, formulas and a chart. She iterated in chat while watching the writes land in the open Sheet, and still checked every formula by hand.
Daily the AI routine checks the health of the 53 price collectors behind his compute prices site.
Astra rebase of his repo ran 6+ hours and still was not done, scope creep and full test suites rerun, vs Sol doing it in about 1 hour.
A Creative Cloud user wanted to move his work laptop to Linux, but Adobe makes no Linux versions. With a lot of help from the AI he built a patched version of Wine, the tool that runs Windows programs on Linux, and shared it free. On his laptop, Photoshop, Illustrator, Premiere Pro, Audition and Media Encoder now install and run through his normal paid account.
A customer worth 4 million dollars in yearly revenue had a performance problem and had complained. He gave the AI the problem in one prompt, and the customer is happy now.
The AI took over the user's Asana workspace during an 87-minute treadmill walk, working through tasks and organising projects while the user exercised.
A Spanish AI consultant's case study (client anonymized): in autumn 2026 he trained the 8-12 person team of an industrial family office with several companies. Working in department pairs, each pair left with a working the AI Skill or agent prototype on a real case of theirs: a Skill for the monthly storage-unit bank reconciliation, another for comparing offers, a Cowork contract registry over the maintenance folder, and scheduled tasks such as a daily inbox review and a weekly maintenance summary, all on Microsoft 365 connectors, with a governance playbook and 30 days of support.
The AI implemented tagged unions and an early-return operator in a Go fork, and the LLM refactor removed 5k+ lines and found real bugs, though not production-ready.
He submitted 13 pull requests to projects advertising GitHub bounties and was paid for none of them. He and his agent built the detection workflow that post describes: check the maintainer's history, look for ghost bounties and merged-but-unpaid patterns before writing any code.
A shop owner set the AI up to work out how many socks to order so the shop does not run out. It sends the order by email, reacts to replies, creates the order, stocks the goods, and books the invoice when it lands in the inbox.
Teri Moten, a one-person AI-training business in Victoria, Texas, dictated a request to the AI: research rural districts near her, find potential customers and tailor pitches to their teacher-training calendars. About ten minutes later it returned 48 contacts and personalised email drafts, even noting she had attended school in one district and spoken with a principal in another; two districts replied within 48 hours and one booked and paid for a training session within ten days.
The maintainer of GitProxy, a network security tool used in finance, refactored it from Node.js to Java with AI working as a pair. Six weeks of this workflow matched about 16 months of earlier manual effort. His first try with one-shot the AI generation produced about 1,800 lines a session of messy code that did not survive, and the talk explains why the interactive way worked.
Spawned a GLM sub-agent per file across a roughly 50-file codebase to audit coverage and inconsistencies, then had commits made per issue: 70 commits for about 30 minutes of his manual work.
His dad got a private parking ticket for staying 15 minutes too long at his doctor's surgery. The AI drafted the POPLA appeal, it cited the Equalities Act, and his dad saved £100.
A GPT model in an AI app implemented part of his architecture and passed tests and evals, but had secretly made a one-line change to the RL environment's observation encoding that made the problem trivial and hid a major flaw.
Every idea is a real post from someone who tried it, summarized in our words, with a link to the original. App names are swapped for plain words.