AI agent use cases
Every use case is scored out of 100 for proof, payoff, ease and cost. Filter to what fits you, then copy the prompt.
70 use cases
-
54
Organize a year of receipts into an expense and tax summary
Forward receipts to one inbox and an agent sorts them into categories, totals them by month, and builds a summary your accountant can use.
-
56
Answer customer questions from your help docs, with a human handoff
An agent answers routine support questions using only your help docs, says when it is not sure, and hands everything else to a person with a summary.
-
43
Have a coding agent write tests for code you already have
Point a coding agent at one module, have it list what the code should do, then write tests for each behavior and run them.
-
56
Turn a keyword into a researched blog draft in WordPress
An agent reads the top search results for a keyword, writes an outline and a draft in your voice, and saves it to WordPress as a draft for you to edit.
-
60
Compare prices and reviews before a big purchase
An agent shortlists products that fit your needs and budget, compares current prices and the most common complaints in reviews, and tells you which to buy.
-
61
Check a retiree’s federal tax estimate across ChatGPT, Claude and Gemini
The Finance Buff's Harry Sit gave the same 2026 retiree tax scenario to ChatGPT, Claude, Gemini and Grok; none got it right first time. After he asked them to recheck, ChatGPT and Claude matched his $1,640 figure, while Gemini kept mishandling the charitable deduction.
-
61
Shortlist NYC two-bedroom rentals in tiers with Manus in under 30 minutes
MIT Technology Review's Caiwei Chen asked Manus to find two-bedroom apartments in New York, and after clarifying her criteria it returned tiered picks like best overall and best value in under half an hour. It first read her wish for outdoor space too literally, and she hit crashes and capacity errors during testing.
-
75
Fill an online shopping cart with Comet, then finish checkout yourself
Android Authority's Karandeep Singh had Comet add pasta ingredients to an Amazon cart in seconds, but it looped for minutes on an item that wasn't sold locally. eMarketer's Grace Harmon saw Comet research and pick boots well, then repeatedly error at checkout.
-
63
Find a top-rated restaurant and book it after confirmation with Perplexity Comet
Tom's Guide writer Priyanca Rajput asked Comet to check top-rated restaurants and availability, and it made the reservation after asking her to confirm, which she called a standout. Its more personal outputs in the same test, like a workout plan, felt generic.
-
63
Reserve an OpenTable dinner for two through Copilot Actions in Edge
PCMag's Michael Muchmore asked Copilot Actions in Edge to book an 8 p.m. table for two at a nearby Japanese restaurant on OpenTable, and it completed the booking. He had to supply a phone number and verification code, it first assumed the wrong city, and he found doing it himself was faster.
-
57
Record a rewards-points check once and have Claude in Chrome repeat it on demand
Eric Hal Schwartz recorded himself checking several loyalty programs and logging balances in a spreadsheet, then had Claude in Chrome rerun the whole routine from a one-line prompt, working across tabs in parallel. It did the job well, with a couple of password errors slowing it down.
-
64
Sort nearly 900 loose Google Drive files into folders with Claude in Chrome
TechRadar's Eric Hal Schwartz had Claude in Chrome organize close to 900 stray documents and spreadsheets into six top-level folders with subfolders, set aside about 50 duplicates, and leave unclear files for him to place. He also used it to draft meeting-time emails from his calendar, while noting privacy concerns about broad browser access.
-
64
Turn a week of dinners into a priced grocery list in Google Sheets with ChatGPT agent
Tom's Guide's Amanda Caswell asked ChatGPT agent to plan five dinners, build a categorized grocery list with price checks, flag nearby sales and export it to Google Sheets. She rated it 10 out of 10, though other tasks in the same test, like ordering lunch, needed more hand-holding.
-
64
Book movie seats and send date-night calendar invites with ChatGPT agent
TechRadar's Eric Hal Schwartz had ChatGPT agent pick seats at a theater, create calendar events for the date and babysitting drop-off, send invites and make an invitation slide deck in under ten minutes. It first chose the wrong theater and paused at payment for him to finish.
-
73
Get a three-year parking charge dropped with a ChatGPT-drafted court defence
Essex driver Rafal Gorfal fed ChatGPT the debt-collector letters about a £100 parking charge that had grown to about £300, and it advised him on replies and helped write his court defence. Weeks before the hearing, the claim and fine were dropped.
-
67
Cut a $195,000 hospital bill to under $40,000 using Claude’s line-item audit
After his brother-in-law died, Matt Rosenberg used Claude to compare the hospital's itemized charges against Medicare billing rules, finding unbundled duplicate charges, wrong inpatient codes and inflated supply prices. Claude helped write the dispute letter, and the hospital cut the bill from $195,000 to roughly $33,000-$37,000 (reports differ).
-
73
Win back a denied prepaid-spa charge with a ChatGPT-written dispute
Charlotte business owner Suuz Martines had been denied a refund for months on prepaid services at a spa that closed. ChatGPT gave her wording, document lists, possible legal grounds and a phone script for her card issuer, and about five days later the dispute was approved and she got her money back.
-
73
Audit a $1,200 dental bill and draft insurance appeals with ChatGPT
Writer Jamie Phillis used ChatGPT to spot questionable charges on a dental bill, then draft an appeal, a second-level appeal and a state insurance complaint. The first appeal was denied and the case never formally closed, but the office never billed her the roughly $1,200 balance.
-
64
Cancel three streaming subscriptions by letting ChatGPT agent click through each site
Tom's Guide AI editor Amanda Caswell had ChatGPT agent log into Peacock, Disney+ and Netflix and work through each cancellation flow, and all three were cancelled. She needed to nudge it in a few spots and warns some sites may block or flag the agent.
-
56
Move a company website off Webflow to a fast static site in under two weeks
Very Good Ventures used Claude Code to move its site, including 204 blog posts and about 1,200 images, from Webflow to Astro on Cloudflare in under two weeks, with agents also writing and running the QA tests. StackOne reported a similar Claude Code rebuild that took new pages from days to 1-2 hours.
-
66
Replace a $40K CRM contract with a custom CRM built in an afternoon
Atonom, a 25-30 person AI sales company, built its own CRM in Lovable with a first working prototype in about 3 hours. It replaced a Salesforce contract, cutting annual CRM costs from $40,000 to about $1,200.
-
66
Build a newsletter summarizer that cuts weekly reading to 30 minutes
Consultant Justin Tan, a self-described low-coder, used Replit Agent over three days to build an app that pulls newsletters from Gmail, removes duplicate stories, summarizes them and archives the emails. His weekly newsletter reading dropped from 2-3 hours to two 15-minute sessions.
-
52
Let a coding agent write most new features and fix defects 10x faster
Cisco rolled out OpenAI Codex across engineering and reports it wrote over 95% of the new features in its AI Defense product. Defect-fix throughput rose 10-15x, and the company says it saves more than 1,500 engineering hours a month.
-
62
Ship a solo-made roguelite to Steam Early Access with an AI coding agent
AI researcher Przemek Chojecki used Claude Code with the Godot engine to build Catvivors, a cat-themed survivors-style auto-battler, and released it on Steam Early Access. He describes about two weeks to a fun first level and a few more weeks for content and store setup.
-
53
Build a full roguelite bullet-hell game in 10 days with parallel coding agents
Adrian Bigaj of BigDevSoon ran up to eight Claude Code agents, linked to the Unity editor through MCP, to build the 2D roguelite Void Balls in 10 days. The project reached about 29,000 lines of C# with 88 test files, for around $50 a month in AI tools.
-
64
Build a retro space shooter with a child in about 20 minutes
Product writer Peter Yang and his 7-year-old used Claude Code inside Cursor's terminal to make a 2D vertical space shooter with three levels, boss fights and music, then published it on Vercel.
-
52
Split a giant legacy codebase into modules with an agent doing the grunt work
Nubank used Devin to help migrate an 8-year-old ETL monolith with over 6 million lines into smaller sub-modules, running many sub-tasks in parallel while engineers reviewed. The company reports an 8-12x engineering efficiency gain and over 20x cost savings on that work.
-
67
Have an agent code a complex open-source feature unattended for seven hours
Rakuten ML engineer Kenta Naruse gave Claude Code the job of adding an activation vector extraction method to the vLLM library. It worked on its own for seven hours, he wrote no code, and the result matched the reference method with 99.9% numerical accuracy.
-
60
Let an AI model play and finish Pokemon Blue live on a stream
Software engineer Joel Zhang built a harness that let Google's Gemini 2.5 Pro play Pokemon Blue on a Twitch stream. The model finished the game after about 37 days, which Google CEO Sundar Pichai shared publicly.
-
56
Turn repeat WordPress hosting chores into one-command agent skills
Austin Ginder of Anchor Hosting built 22 Claude Code skills that run tasks like site migrations, backups, malware cleanup and plugin patching across about 2,900 WordPress sites. A migration that took 20-30 minutes of hands-on terminal work now starts with one command and runs while he does other things.