Let an agent handle routine grocery ordering
A parent used Grok Bot with a Samsung fridge and receipt history to automate family grocery runs.
Agent Uses · 1,321 real examples
Real examples, collected from Reddit, X, Hacker News, blogs and the press, of people handing work to AI agents: what they asked for, what happened, what went wrong, and a prompt you can copy to try it yourself.
The day's AI news plus 3 agent uses you can try, with ready-to-paste prompts. Free; unsubscribe anytime.
A parent used Grok Bot with a Samsung fridge and receipt history to automate family grocery runs.
A Grok-based assistant used phone calls to ask local stores about stock and returned the conversations and availability details.
Replit Agent added 61 fresh child-friendly mini-games to raise one app’s total from 40 to 101.
A dot handled separate task threads in the background and sent reminders about tax paperwork and a refund.
They had Scout handle routine errands: bills, inbox cleanup, and a grocery order.
A Meta Muse agent checked investing chatter, followed watchlist names, and turned news and podcasts into a daily macro brief.
They tried to get Claude to finish a Warsaw-to-London booking, but card handling blocked the process and it turned into a travel product project.
They used Claude to scan recent posts for commitments and uncovered a false alarm caused by test content.
Claude Cowork helped outline a full-day drive, from breaks and fuel stops to places where the driver should pay closer attention.
People used Meta’s Muse to search for unclaimed money and to press an insurer for money back on a canceled policy.
A developer used a Dot to spot repeated customer requests, make small fixes, test them, and return draft pull requests with demo clips.
They used Claude to model a retirement portfolio and check whether spending would still hold up through past bear markets and extra home costs.
A developer connected sandboxed Claude Code instances to a virtual file system so they could pass files through MCP without direct access to AWS.
A developer built a cloud email analysis system that lets LLMs pull answers from Gmail and Outlook through MCP.
A poster asked how Claude Code could stay connected to Robinhood through MCP so it could keep checking investments over time.
A user let Comet manage their X follows, and it unfollowed everyone before starting to follow accounts back.
They had Dots organize a trip and start a shuttle reservation, but login checks and approval prompts prevented it from finishing.
A user had Meta Muse scan airline emails and it found a refund that had already been issued.
Muse searched hotel websites for a New York City trip and put the Marlton Hotel at the top.
They used OpenClaw and Linear as a control layer to send tasks to Cyrus, which picked the repo, chose a model, and opened pull requests.
Meta Muse was used to try to buy two plane tickets, but a captcha stopped it and the user had to complete the reservation manually.
Have an agent compare flights, narrow hotel choices, and map a brief layover plan for a short-notice business trip.
A builder used n8n to make a website chat agent that vets visitors and books meeting times without manual help.
A developer swapped token-heavy Jira and GitLab MCPs for Markdown-based skills generated from the Atlassian and GitLab CLIs.