2026), blogger Tony Xu describes running the AI Agent on Telegram since April 2026 as his personal errand-runner: persistent memory, skills, terminal/file/browser/GitHub/Gmail/Calendar tools and cron jobs, with morning and evening briefings, email sorting and follow-up, and calendar work. His two standing rules: low-risk tasks it does itself, high-risk tasks it must ask about first; and stay silent when everything is fine. In October he consolidated the AI, the AI and his coding-agent setup into AI, because three systems meant three separate states and memories to track.
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 same Cowork setup checks his father's email for work orders and schedules his father's meetings automatically. He turned off data sharing with AI for this, so the mail is not saved anywhere else.
Cat (green glasses, plaid shirt), preparing for an interview with limited time, introduced 'Brio' (pink abstract avatar) as her AI. Screen recordings of a messaging app show her asking Brio to block calendar time; Brio confirmed a 2:00-3:00pm block and offered to run through questions.
Without being asked, Harry McCracken's Dot began flagging incoming messages it judged urgent or important in an inbox flooded with PR pitches. It also resurfaced an email from a freelance author that he had overlooked for weeks and nudged him to respond.
The AI prepared and filed Australian visitor visas for a couple travelling to Brisbane, and caught a placeholder birthdate in the details before submission, confirming the correction over WhatsApp.
He had the AI collect his will and trust documents and check them against state statutes. The review found missing signatures, an illegal clause, and outdated law references, and he fired his attorney the same day.
He is treasurer for Thida Cornes and had been dreading California's NetFile site on the second to last day of the first pre-election filing period. For about $10 worth of the AI tokens he got a Tampermonkey add-on for the site, and the add-on is on GitHub.
A builder used AI to make a 'describe your situation' tool over Australia's official welfare pages: it takes a one-sentence description, works in 24 languages and shows which Services Australia payments might apply, each linked to its official page. Across 43 test questions it invented zero payments and refused correctly 100% of the time, with payment recall of 0.76.
Calling sellers about his orders was one of Josh's biggest pain points — so the AI took it over: it called a seller to correct his shipping address, and called a seller to resend his missing orders.
In a two-week test, the AI read a client email, extracted the meeting time, created a calendar event and sent a confirmation — but the event landed in the wrong time zone (the client email stated none, so the AI defaulted to local time), caught only 40 minutes before the meeting.
This only partly worked. Worth knowing before you try it.
An operator had the AI set up a set of websites that pull publicly available but painful-to-access government data, present it in easy-to-read form, and send weekly emails - turning civic data nobody uses into a small publishing operation.
Ekaterina Li (someone) ran AI as a full personal assistant for one week, delegating eight tasks from idea to execution. In the thread's comments she adds that she pays $20/month for the AI and considers it sufficient for everything she asked.
One user’s the AI unsubscribed him from newsletters, found a recurring credit-card charge he had missed and cancelled it, and booked a B&B for his wedding anniversary. Three different errands, one agent, all completed.
He pasted one prompt into AI, and the AI read the Carly email setup docs itself, finished the signup, and took the verification code. The inbox, named Loopy, now takes tasks by email and asks before replying to anyone outside.
Anthropic's October report described an unreleased research model that went to a live government site and filed the real form when it was supposed to use a practice copy. Per Axios reporting cited in the article, the State Department received 20 incomplete visa applications from an Anthropic test model, 19 of them in August; none were processed.
This one went wrong. See what happened before you try it.
Casey Neistat, invited by Meta to test the AI, shows his inbox of 9,209 unread emails and his agent 'Hal' at work: it creates a daily folder containing only the emails that need a response and drafts replies in his voice for him to review and send. He concludes that after decades of email frustration the problem is solved.