In a screen-recorded benchmark run October 16–17, 2026, the AI was given the same CSV reconciliation task three times in fresh chats and graded against a hidden answer key. It passed all three runs at a median cost of about $0.48 per task at list price.
AI agent looped 21+ hours on his rendering engine streaming problem, tuning motion prediction and caching, fixed minor bugs, main problem unsolved, quota burning.
This only partly worked. Worth knowing before you try it.
Forum founder MJ DeMarco needed a text update across 12 pages of one of his websites, but the platform that built it demanded a new subscription just to edit. He gave the AI access to the site folder instead; it made all the changes in about 3 minutes, avoiding an estimated hour of manual editing and a $20/month subscription.
The Verge reported the AI paid creators to run Instagram, YouTube and TikTok accounts promoting its agent as an easy-money gig: find local businesses with no website or a bad one, have the agent build one in under ten minutes, then call the business and sell it, with claims of a potential $5,000 a month. The creator posts often obscured their paid ties to the AI, and the featured TikTok accounts were taken down after The Verge inquired.
He told his AI to come back with 20 cold email accounts paid for, set up and warming up. It sorted Namecheap API access, signed up with an inbox provider, bought 20 inboxes on his saved card and wired them into their system.
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.
He asked his bot to sign him up for the side events around All-In Montreal. He woke up registered for 10 events, including some he did not know existed.
A store owner's month-end runs itself: QuickBooks ingests bank transactions while the AI works the store's API transactions, reconciles the two, scrapes vendor invoices out of Google Drive to update COGS batches, and files a PDF audit trail with each month-end journal entry. Four minutes a month after about an hour of setup, in a thread where the original poster had just found $3,500 of mishandled HST doing the same job by hand.
A solo operator in Texas runs DeedSpring, where people get paid $13 to $75 for small real world deeds like leaving art in public or helping a stranger. His agent Vesper runs the desk: recruiting, acknowledgments, checking photo evidence against the posted rules, and payouts through escrow. Six deeds have been paid so far across five countries, and every receipt sits on a public ledger.
Smarsh's internal Agentforce agent Emmy hit 65 percent adoption among support reps, resolves 31 percent of cases with AI, and saves an average of 7.5 hours per case.
In the same week-long GrokBot trial, the reviewer found every wake (a routine, a webhook, another bot, finished background work) costs a full model turn even when the bot decides the news is not worth sharing, and his bots burned about $100 on that silence. The idle-cost failure, plus memory being rewritten out of sight, is why he stopped using it.
$100a week
This one went wrong. See what happened before you try it.
Beef, sheep and cotton farmer Andrew Gill dictates voice notes and takes photos in the paddock, and the Farm Assistant agent files them alongside his paddock sensor data, livestock market prices, weather and cash flow. Automating the admin alone saves him and his wife about five hours a month.
A developer says he now just tells his AI to find the bugs overnight and fix them by morning. The AI works on its own computer, cloning the repo and installing the toolchain, and his Puppeteer script clicks approve so pull requests are waiting for his review.
An AI routine with only Slack and Sentry MCP access reads Sentry errors, compares them to the app source, and posts a daily digest to Slack; reports very accurate.
Someone broke into his web host and deleted everything behind Now I Know, his daily newsletter of sixteen years, 3,409 published stories, and there was no backup service on the server. His AI chief of staff, set up three months earlier to run his calendar, inbox, job search and projects, helped draft the emails to his host's support team, scanned thousands of his emails and Internet Archive links for surviving copies, and recovered almost 100 percent of the writing and most of the images, then connected to beehiiv and uploaded a test draft from the recovered archive for his review.
This only partly worked. Worth knowing before you try it.
For four months, the AI has handled the faxes that come into his company. He says it now spots junk faxes every time, and the ones from real estate firms, investment firms, and staffing agencies no longer get through.
Opus 5.5 clean-room reimplemented a proprietary legacy grid control from observation of its public API, producing 500 source files, several thousand tests and pixel-comparison reports in a few hours, avoiding a licence over 1,000 dollars per developer per year.