A small research outfit pointed the AI at the public 2024 Medicare provider files. It flagged one wound dressing code whose national billing jumped from about $23 million a year to $650 million in 2024, with one Florida supplier taking 62 percent of it, and a pattern of hundreds of family doctors each ordering ostomy supplies for about 175 patients across 40 states while only 15 colorectal surgeons appear in the same data. Some flagged companies have already been caught; others with the same shape are still billing, per the July 2026 enrolment file.
Kimi Code ran two coding tasks on small repos, completed neither, and burned its whole 5 hour quota in minutes, about 1.37 dollars of API-rate usage across 1.59M tokens in the logs.
$1.37at stake
This one went wrong. See what happened before you try it.
His son was lead designer and chief tester on a Smash style fighting game where the whole cast is his real stuffed animals, including a knitted blanket with burn holes as the villain. The game has a story mode set in a 3D model of their actual house, 26 fights, online play, and no ads, and it is out now on iPad and iPhone, posted in Reddit.
Locally hosted agents rebuilt a Conductor flaw in an isolated lab and got root access inside the test container. The team checked the flaw was gone after the upgrade and turned the work into signatures, rules, a playbook, and a paper.
He has the AI edit his YouTube videos end to end. It transcribes them, cuts the dead air, zooms the jump cuts, adds lower thirds, sound effects and music, and hands back a Premiere timeline.
His bookkeeping for a full calendar year was a trainwreck, and he needed to move accounting platforms, tidy the previous year up and prepare for an admittedly late tax return. Fable built a proof of concept migration in less than two days: it set up hosting, automated the spend categorization and got the new platform displaying his previous year’s books. When Opus 5.5 came out, it reviewed the work, caught several discrepancies in the data and corrected them, and also stood up two other self hosted accounting tools for him to try. He still hated the platform he had chosen, because it was over complicated for his needs.
This only partly worked. Worth knowing before you try it.
A user built a custom the AI skill that remote-starts their car, reports EV battery level and monitors charging status from any chat. They published the skill because nobody else had made one for it.
A small business owner runs a few AI bots that share one Google Drive knowledge base. One built and tested his client assessment flow end to end, from fit call to payment, booking and intake, another reads his calendar and preps his mornings, and the bots also fixed his task list and built a customer slide deck, saving him about two hours in one week.
Without being asked, the journalist's Dot (Sam) scanned her email and flagged two things she had missed: a message that had accidentally landed in spam, and a form she had forgotten to fill out.
A question asked for counterexamples to a graph coloring algorithm that reduces coloring to polynomial equations. With AI assistance, a computation in Singular produced a 61 vertex line graph that the method handles wrongly, and the answer posts self contained Sage code that makes the exact call once so anyone can check it.
The AI was fed government forms, UI design and legal requirements and produced the app in 1 hour; the team completed conversion of 150+ forms in 2 weeks with one person versus a year before.
He let the AI UltraCode run overnight on their triggered ad server. It found ways to make the server work better, which it said should add $100K in top line revenue, and it found several long standing bugs they had missed.
His AI Scout built a staged pre-trip checklist ('a few days before', 'day before', 'morning of') for traveling with his dog — pet-friendly lodging, vaccinations, mail hold, packing — then sent reminders that day and a final 8:41 AM morning checklist before they drove off.
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.
A maker had an LLM design a circuit board by running KiCad headlessly from a terminal, checking the design in a loop until it passed every rule check. He sent it to PCBWay for manufacture and it worked the first time, and the board is now in a product many people use.
A father of two walked his house photographing his belongings, including model and serial numbers where he could, and dropped the photos into a folder for the AI. In 20 to 30 minutes it produced a spreadsheet cataloguing the items with descriptions and current replacement values, each with a confidence rating, grouped by room, plus a tab for things needing a second look. He can add photos later and ask for the list to be updated.
A Salesforce developer told his AI agent to unwrap Calculated Insights from a data kit into a sandbox. The agent worked out the config keys for most component types but failed twice on the insight template, then gave up on unwrapping and created the insight directly through the Data Cloud Connect API, copying the SQL from the kit metadata. The created insight showed as active and queryable, but a follow up answer shows API created insights sit in a broken state and cannot become trigger eligible, so it has to be deleted and unwrapped properly.
This only partly worked. Worth knowing before you try it.
His nephew kept asking for a game that plays a note when he types a letter on a keyboard. He used AI to build it and put it online, and his nephew loved it.
His role was cut on Tuesday morning, 22 September. That evening he asked the AI for something fun, gave it one standing instruction to go forward and only pause when his input was needed, and by 6:37 the next morning a three stage beat em up about the job hunt was live on the internet, with his real CV linked from the end card.
A refund of several hundred dollars was still pending, so Cramatata casually asked the AI to call Apple customer service and ask about its progress. The AI made the call, spoke with a live representative who checked the order and its progress item by item, and only transferred the call to her when identity verification was required.
Ran agents under KPIs for his website; agents deployed and monitored it and the CEO agent's report flagged an overnight traffic spike that took his Mac mini down and asked for a scale-up budget.
This only partly worked. Worth knowing before you try it.
Before he bought a used graphics card, the AI went through eBay listings and seller histories, flagged a scammer quoting $1,450, and steered him to the right 24GB card at $770 instead. He says he would have bought the wrong card, and possibly from the scammer, without the check.
He had the AI create a website where people at work send in requests, which keeps him organised. Then he made it read the requests and carry out the fixes as they come in.
AI fixed a Home Assistant integration restart bug with a one-liner that works for him, and now maintains his personal fork with the patch against upstream.
Used nit-picking history-teacher agents to evaluate a few hundred generated stories for accuracy, and pointed the AI at HN comments to aggregate issues and fix them with his OK; reported issues mostly fixed.
A business user asked the AI to duplicate a template; the request bounced between chats for an entire day, each claiming the project belonged elsewhere or that it lacked access to make the change. Contradictory explanations piled up and no usable result was produced, so the business is migrating away.
This one went wrong. See what happened before you try it.
An angler who designs fishing games used AI Build to make Bank and Boat, a free fishing game for phone and browser. It covers day tickets on UK canals, rivers, lochs and piers, and it is live now with no sign-in.
Anthropic uses Fin, the customer-service agent Salesforce acquired, which autonomously resolves 79 percent of the conversations it sees, per Salesforce's release.
He wrote lyrics with AI, made the song in Suno, then had the AI plan a stills list and assemble the video. GPT rendered 42 stills, and the AI put them together with the lyrics and rendered the finished video.
A CMO at a business software company designed a marketing system of about 130 AI agents, one per job: researcher, writer, fact checker, critic, publisher. It runs her blog in three languages, LinkedIn for five executives, outbound and SEO, ticking every five minutes and starting work only when something is due. After a year in production it has published more than 630 articles and blog traffic from search is up 3.9 times. Her worst failure was a scheduled job that failed silently for a month; now a weekly audit checks the scheduler itself.
A developer asked the AI to batch-generate images through fal.ai in budgeted chunks with checks between runs. The script the AI wrote ran all 34 items with no account or balance check between the chunks. The account locked partway through, 11 more items were charged and failed, and the balance ended at -$3.68.
$11.67at stake
This one went wrong. See what happened before you try it.
He uses AI to check his email hourly, mostly overnight, handling customer mail and marketer pitches while it does not act during the day. It also watches his ad campaigns and gives him a summary of how everything is running, and checks his school email into a weekly digest.
He used AI to build his own to-do and budgeting apps that work the way his brain works, after never finding apps that suited him. The AI also taught him how to sideload them onto his phone, and he calls it a game changer.
Her job is Teams messages and email all day, and brain fog makes finding things hard. She asks the AI to find a file by who shared it with her, and it finds it.
He had the AI log into his carrier account and go through his phone lines: three lines, four counting his watch. The AI worked out that moving his wife and son to a different plan while keeping him on his grandfathered plan would keep his unlimited tethering and their free HBO, and cut the bill by $543 a year. It laid out the options, he confirmed the plan he wanted, and the AI did everything after that. He says he now throws almost all his form filling at it.
A Sol coding agent and an Opus review agent spent half a day on a new tool and produced only scaffolding, admitting they worked on unrequested things despite a scope-creep checkpoint.
This one went wrong. See what happened before you try it.
He and his wife play in a Magic: The Gathering league where they open new product every month and keep rotating their deck builds and strategies. The tracking used to live in spreadsheets and note apps. He built a custom Android app that holds everything they need, including built in life trackers that log wins and round data automatically. He also built a recipe sharing and storage app, because he cooks a lot at home and never found one he liked.
He built one skill that does his company credit card expenses and another that files his email in Outlook. Both were manual jobs he hated. Now he mostly watches, answers the odd question and finds an invoice when asked.
A seller listed an exercise machine on Facebook Marketplace using her Cue agent Luna's own phone number instead of hers. Luna answered and screened the buyer calls all evening and handed her a list of who called and what they wanted, so she could choose the buyer without answering a single call.
A small contracting business owner, replying to a day to day uses thread, says he has the AI updating his QuickBooks from his online receipts. Bookkeeping from scattered receipts is the classic end of week chore for a small contractor; his agent pulls the receipts and keeps the books current.
Hitachi built AI agents for more than 200 use cases in its global power and utilities business, including verifying photos of every tool and wire before and after a job. Frontline worker productivity rose 400 percent and safety improved.
A dad who had never built more than blinking lights and a dog-bark detector decided the commercial live MLB scoreboard was too expensive for what it did and built his own as his son's 10th-birthday gift - a physical product, not an app. Over months of nights and weekends with AI/the AI and the AI he went from concept to web preview to a parts list and a 3D-printed case screwed into the back; the finished board shows live MLB and NFL scores, pulls his son's own baseball games and practice schedule from GameChanger, runs arcade games, and has a quiet mode that shuts it down after bedtime.
In an on-screen walkthrough, one Telegram instruction set three the AI profiles working as a pipeline: a read-only scout and an ops profile gathered and summarised the week's topics in parallel, a verification step dropped anything vague, and a writer profile produced the final post with sources. The finished post was delivered back to the chat with every artifact saved as a local file.
For about 25 years he finished a monthly financial statement in Excel by hand, and it took hours. He connected the AI read only to all his accounts through Plaid. The AI now builds the whole statement and asks him only about things it cannot see, like cash spending. The job takes 15 minutes instead of hours. He is having the AI build him a custom money dashboard next.
An economist set up an email bot to bring him any message tagged #handcarry right away. The bot misspelled the tag as #hancarry in its own memory file, made a new folder, and filed a week of urgent emails deep in it where he never saw them.
This one went wrong. See what happened before you try it.
His mother has run a resume writing and career coaching business for almost 20 years, but bookings had been falling and her site sat on an old WordPress setup. He moved her to GitHub and Vercel, rebuilt the whole site with AI in a few hours, added online booking through Calendly, and moved the domain over. She now runs the site herself with AI's help, and her revenue has quadrupled in the last two months on better conversion and more bookings.
As part of shopping research Harry McCracken threw at his Dot, it surfaced a $100-off deal on a Magic Keyboard for his iPad that he says he would never have located on his own.
A Barron's editor gave the AI full access to his credit card account for the review. The AI flagged a suspicious four-figure expense from August that the editor did not remember; after back-and-forth, the agent admitted it had accidentally identified a mortgage payment from August 2025 as an expense from August of this year. Schafer's dry takeaway: most things coming from AI right now still require human verification.
This one went wrong. See what happened before you try it.
After fake bot orders started swarming his Shopify store and blockers failed, a store owner worked with AI to run checks every half hour: any order under $10 is auto-blocked, known fake addresses like 123 Main St are blacklisted, and he gets a report every 30 minutes. The bots still try, but far less often.
Ekaterina Li reports on a week of delegating to the AI, including processing two refunds totalling $1,000 that she had long postponed. Her thread also notes the AI cut her subscription costs by $200 in the same week, and that the $20 plan covered all her tasks.
He says his business got serious about Power BI reporting about twelve months ago, then three months ago threw it out and replaced it with AI licences. A third party warehouse tidies the data with an MCP on top, and the finance team now builds its own reports quickly instead of waiting on a reporting team.
Three independent AI audits went through his Cashu payment server and wallet library line by line for a new release. All of them found real bugs that previous audits and models had missed, and one model found a higher severity bug nobody had caught. The release fixes what they found.
Spent about 10,000 dollars running autoresearch agent loops over the top 200 GitHub repos for frontend speedups and distilled the wins into a public leaderboard and wrapper tool; no per-repo numbers stated.
$10,000
This only partly worked. Worth knowing before you try it.
New dad Crawford Collins (verified, 1.1M) personalized his agent as 'Atlas'. Voice commands while caring for his infant son: it planned efficient 4-day-a-week workouts synced to his calendar, found and ordered fragrance-free size 3 diapers (approved, delivered), and when grandma watched the baby, found and booked a sushi restaurant for two (Yasu Beverly Hills / Sushi Note).
Every week the AI takes his pile of missing work reimbursements, files them, and submits them for him. He says it is the real time saver for anyone who travels for work.
An AI runs a small shop in public — writing posts, publishing articles, replying to readers and checking its own health — and must log a 'debt' whenever it does a job by hand, then turn that job into a tested command. After the owners made an open debt fail the agent's own test suite, median time to close a debt fell from 8.2 days to about 6 minutes across 48 debts.
He and his friends share their Hádej slova and Wordle results in a WhatsApp group. Their AI, named Klepeto, watches the group, builds leaderboards and statistics from the results, and also plays under its own account.
Josh Dubowsky connected the AI to his Gmail and asked what he needed to deal with that week: six items came back in about nine seconds and five were right, but the sixth claimed he had confirmed attendance at an October event he had explicitly declined — the summary had dropped the word 'unfortunately'. Weeks of relying on summaries also meant a failed-payment line buried in a dull email went unnoticed for four days.
This only partly worked. Worth knowing before you try it.
A bearded creator (Medo, 6.1K followers) showed screenshots of the AI and a text conversation where he asked the agent to check his email for uncancelled subscriptions. The AI drafted a cancellation email, he tapped 'Yes, cancel it,' and two days later he received confirmation of a $24.99 refund.
He rebuilt vdbx.io with AI as a static Astro site pulled from his own Obsidian notes, in about 4 days. The new site replaced his Framer site and his GitBook pages with one site.
Had the AI transpile his old frameworkless PHP site to Rust and ran it behind the same nginx config; Rust was slower until the AI added micro-caching, reaching comparable rps and latency.
This only partly worked. Worth knowing before you try it.
His budget app and his bank balance were $50 apart. He exported the data from each one, had the AI compare the two files, and the AI found the mismatch, saving him about half an hour of checking lines by hand.