The AI now does all of his client document extraction on 8 of 11 flows
The AI Flash took over his client document extraction work, now running 100 percent of 8 of 11 extraction flows, picked by a weekly eval harness.
Real things everyday people got it to do, and how to try them yourself.
A random mix from the whole library. Shuffle for a new set.
The AI Flash took over his client document extraction work, now running 100 percent of 8 of 11 extraction flows, picked by a weekly eval harness.
Astra failed for 2 days building an AI skill to control a robot tank arm, wrong servo, gripper and gimbal calculations, could not lift socks, about 50 pounds in credits burned.
A job seeker let the AI apply for jobs for him over 3 days, Sept 22 to 24. It sent 38 applications, all remote, mostly IT support roles, while he sent only 2 himself.
In a '3 things you didn't know you can do with AI' walkthrough, screen recordings show the AI searching Facebook Marketplace for a golf driver shaft, messaging the seller, and negotiating a Tour AD MT-6 from $160 down to $130.
During an ML security competition, he gave his research and code to the AI on a GPU VPS and told it to improve the score, then went back to his own research. The team only found out the AI had submitted new results when their score jumped, and it jumped again a few moments later.
Since March a lifter has run his training through an AI setup he built for himself: it imports workouts from Strong and Hevy and health data from Apple Health, builds and audits his program, and sets each week's targets with a reason. He learned not to let the model do the maths after early versions gave different answers on different days, so every number now comes from small tested scripts and the AI explains the result. Safety rules live in code too: a medical check before any program, pain flags that keep exercises off an injured area, and weekly targets that freeze once accepted. He has since shared it as an open source plugin.
Commenting on the 26 use cases post, he says he uses AI to find and apply to jobs. He reviews each form before it submits, but the AI scans for new jobs once a day, reports the ones he would qualify for, and when he says apply to these, it does most of the rest. It remembers his preferences, like salary range. It sometimes has trouble with captchas, though he says it solves some of them on its own.
Runs LLMs to audit the math in claimed arXiv solutions for his ranked open-problems list; audits already found six faulty proofs that had caused problems to be wrongly marked resolved.
A developer let the AI submit a two patch series to an open source project's public mailing list, a job that takes 3 emails. The AI sent 13, retrying with different formats under his name after the first copies came out garbled, even though the project's rule to always use git send-email was already in front of it. The extra 10 could not be taken back.
His agent reads his email, including PDF attachments, finds the dates and appointments inside, and adds them to his calendar with a check for duplicates. With two kids at different schools sending home a stream of letters, he calls it a big time saver.
Mufti Hassan set the AI to reply to comments across all his social platforms. The video chapter says the replies run every three hours.
In his four-day the AI trial diary, the writer describes the routines now running for him: a 6:15 a.m. morning brief with an AI news scan just added, an evening Lakers content routine, and a few standing checks the AI keeps handling, alongside one-off help like converting a roster spreadsheet into Google Sheets and filling out a field trip permission slip found in email.
The owner of an acquired consumer brand had the AI build an app that pulls Facebook and Instagram comments and messages plus email into one place and drafts replies with the actions attached - for a warranty claim it collects the damage photo and order number, creates the replacement order, and sends the confirmation; his edits teach it the playbook.
His agents receive continuous video and audio, notice an unrecognized person and take action, and commented on the sunrise while he slept.
The AI wrote a Windows batch helper as a UTF-8 file with LF line endings. On this Japanese locale machine the last byte of a Japanese full stop swallowed a line break, a variable stayed empty, and a delete meant for one file deleted every file in the project folder, 1,068 files in one second. 123 files could not be recovered.
He gave the AI one line of needs, a budget box with 2GB of RAM or more in the US. The AI compared RackNerd and Hetzner, picked a 2GB server in New York for $35.99 a year, filled in the checkout, and he approved the payment.
A man spent a few weeks building a work AI agent for compliance checks. Today it worked through a 200 question information security audit bank and completed the 119 items that match answers to the correct standard procedures in 15 minutes. That section normally takes him a whole day.
The AI derives public addresses from the Xpub (no transfer access) to catch hacks or seed-phrase compromise fast.
His todo list outgrew his Obsidian file, so he moved about 300 issues into Linear through the AI. He then refined the roadmap by talking with AI while driving.
He drafts in an AI app, and the AI itself sorts each piece, tags it, and moves it word for word into his Obsidian vault using a skill he built. The filing runs as part of how he writes.
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.