# Agent Casebook

> Real things people are doing with AI agents. Steal an idea.

Outcomes are reported by the linked sources. This directory does not independently verify them and does not add dollar figures that are not already in the case note.

Current sponsor price: $200. One week on the site bar + Daily 5 email. Price rises $100 after every sponsor. Sponsors sold: 0.

## Pages
- [Directory](http://localhost:43123/): 40 reported cases
- [Daily 5](http://localhost:43123/daily): email signup; sending is not wired
- [Sponsor](http://localhost:43123/sponsor): price ladder and Creem checkout
- [Submit](http://localhost:43123/submit): suggest a case
- [About](http://localhost:43123/about): how to read the reports
- [Feed](http://localhost:43123/feed.xml): RSS

## Topics
- [Money, Bills & Benefits](http://localhost:43123/topics/money): 14 cases
- [Coding & Shipping](http://localhost:43123/topics/coding): 4 cases
- [Research & Knowledge](http://localhost:43123/topics/research): 6 cases
- [Writing & Content](http://localhost:43123/topics/writing): 4 cases
- [Ops & Admin](http://localhost:43123/topics/ops): 4 cases
- [Sales & Growth](http://localhost:43123/topics/sales): 2 cases
- [Personal Life](http://localhost:43123/topics/personal): 2 cases
- [Building Agents](http://localhost:43123/topics/agents): 4 cases

## Use cases

### Repair corrupt files for a flat fee

- URL: http://localhost:43123/use-case/repair-corrupt-files
- Topic: Money, Bills & Benefits
- Agent: Custom agent
- Summary: An AI agent offers paid repair of corrupted images, documents, audio and archives for about $5 per file, refunded if it can't fix them.
- Outcome: Paid micro-service with refund-if-fail pricing.
- Source: Agent Use Cases (https://agentusecases.ai)
- Prompt: Build a flat-fee file-repair agent: accept upload, attempt repair for images/docs/audio/archives, charge ~$5, auto-refund if repair fails.

### Grow a bespoke agent that builds its own MCP server

- URL: http://localhost:43123/use-case/claude-code-mcp-server
- Topic: Building Agents
- Agent: Claude Code
- Summary: A builder started from bash + Python + Claude Code CLI; the agent created its own MCP server so they could chat with it from any Claude session.
- Outcome: Reusable MCP-backed agent reachable from Claude sessions.
- Source: Agent Use Cases / Reddit (https://agentusecases.ai)
- Prompt: Using Claude Code, turn this workflow into an MCP server with tools for [tasks], then connect it so I can call it from any Claude session.

### Scrape marketing agency leads into a Google Sheet

- URL: http://localhost:43123/use-case/scrape-agency-leads
- Topic: Sales & Growth
- Agent: Perplexity Comet
- Summary: Comet extracted founders' names, emails and company names for a niche and updated a Google Sheet.
- Outcome: Lead list written straight into Sheets.
- Source: Agent Use Cases / Reddit (https://agentusecases.ai)
- Prompt: Find [N] [niche] agencies. Extract founder name, email, company, website into a Google Sheet. Skip incomplete rows.

### Plan trips and compare grocery prices with Operator

- URL: http://localhost:43123/use-case/operator-trip-groceries
- Topic: Personal Life
- Agent: Operator
- Summary: Users had Operator research a trip into Notion and compare prices across nearby grocery stores into a Sheet.
- Outcome: Trip plan + grocery price comparison, though often slow.
- Source: Agent Use Cases / Reddit (https://agentusecases.ai)
- Prompt: Plan a [N]-day trip to [city] into Notion. Separately compare [grocery list] across [stores] into a Sheet with cheapest store marked.

### Read blocked web articles through a browser-control MCP

- URL: http://localhost:43123/use-case/browser-mcp-blocked-articles
- Topic: Research & Knowledge
- Agent: Claude + browser MCP
- Summary: Claude with a browser-control MCP navigated, clicked, and read articles it otherwise could not access.
- Outcome: Access to pages behind soft blocks via real browser control.
- Source: Agent Use Cases / Reddit (https://agentusecases.ai)
- Prompt: Using browser MCP, open [URL], dismiss cookie banners, extract the full article text and a 5-bullet summary with quotes.

### Book flights, groceries and restaurants with ChatGPT Agent

- URL: http://localhost:43123/use-case/chatgpt-agent-bookings
- Topic: Personal Life
- Agent: ChatGPT Agent
- Summary: Commenters report ChatGPT Agent booked a flight, ordered Instacart groceries, and made restaurant reservations when given details.
- Outcome: End-to-end booking/ordering with human-provided details.
- Source: Agent Use Cases / Reddit (https://agentusecases.ai)
- Prompt: Book [flight/restaurant/grocery order] with these constraints: [details]. Confirm totals before paying and pause for my approval at checkout.

### Clean Gmail and find old Reddit comments with a browser agent

- URL: http://localhost:43123/use-case/comet-clean-gmail
- Topic: Ops & Admin
- Agent: Perplexity Comet
- Summary: Comet summarized tabs, found an old Reddit comment, marked Gmail read, and unsubscribed from newsletters.
- Outcome: Inbox cleanup + cross-site lookup in one session.
- Source: Agent Use Cases / Reddit (https://agentusecases.ai)
- Prompt: Mark promotional Gmail as read, unsubscribe from the top 20 senders, and find my old Reddit comment about [topic].

### Run a small business's recurring workflows with scheduled agent tasks

- URL: http://localhost:43123/use-case/scheduled-smb-workflows
- Topic: Ops & Admin
- Agent: Claude Cowork / Claude Code
- Summary: An author moved SMB instructions, custom skills and a dozen+ scheduled tasks from local Cowork sessions to Claude Code cloud sessions.
- Outcome: Recurring business workflows on a schedule.
- Source: Agent Use Cases / Reddit (https://agentusecases.ai)
- Prompt: Turn these SOP docs into scheduled agent tasks: [list]. Each run should log what changed and ping me only on failures.

### Negotiate retention discounts on bills by phone

- URL: http://localhost:43123/use-case/bill-negotiation-voice
- Topic: Money, Bills & Benefits
- Agent: Custom voice agent
- Summary: A voice agent calls providers, navigates IVR, waits on hold, and negotiates retention discounts, pausing for OTP.
- Outcome: Hands-free bill negotiation with human OTP handoff.
- Source: Agent Use Cases / HN (https://agentusecases.ai)
- Prompt: Call [provider], navigate to retention, negotiate a discount on account [last4]. Pause for OTP. Log offer and confirmation number.

### Prepare tax returns from a folder of documents

- URL: http://localhost:43123/use-case/claude-tax-folder
- Topic: Money, Bills & Benefits
- Agent: Claude
- Summary: A Redditor pointed Claude at a tax folder; returns for multiple years came back quickly and reportedly matched TurboTax, still needing mailing.
- Outcome: Draft returns from documents; human files/mails.
- Source: Agent Use Cases / Reddit (https://agentusecases.ai)
- Prompt: From this tax folder, prepare federal/state return drafts for [year]. List every assumption and forms still needing human signatures.

### Draft a tax filing from bank statements with Claude Code

- URL: http://localhost:43123/use-case/claude-code-tax-statements
- Topic: Money, Bills & Benefits
- Agent: Claude Code
- Summary: Users had Claude Code draft filings from bank statements/invoices and organize large screenshot dumps.
- Outcome: Tax draft + organized evidence from messy inputs.
- Source: Agent Use Cases (https://agentusecases.ai)
- Prompt: Parse bank statements and invoices in [folder]. Build Schedule C categories and a filing checklist. Do not invent deductions.

### Negotiate a lower Comcast bill via chatbot

- URL: http://localhost:43123/use-case/comcast-chatbot-discount
- Topic: Money, Bills & Benefits
- Agent: ChatGPT / DoNotPay
- Summary: DoNotPay's ChatGPT-based bot chatted with Comcast support and secured a monthly discount for an engineer.
- Outcome: Reported $10/month discount via support chat.
- Source: Agent Use Cases / Vice (https://agentusecases.ai)
- Prompt: Chat with [ISP] support as retention. Goal: lower monthly bill without changing service. Stop if they require identity steps I must do.

### Find unclaimed money and start the claim

- URL: http://localhost:43123/use-case/unclaimed-money-muse
- Topic: Money, Bills & Benefits
- Agent: Meta Muse
- Summary: Meta's Muse agent found over $1,500 in unclaimed funds in minutes and helped fill paperwork; claim filed but unpaid in the report.
- Outcome: ~$1,500 found; claim process started.
- Source: Agent Use Cases (https://agentusecases.ai)
- Prompt: Search unclaimed property for [name/aliases] in [states]. List matches, start the claim forms, and show me anything requiring notarization.

### Dispute an airline EU261 delay compensation rejection

- URL: http://localhost:43123/use-case/eu261-delay-claim
- Topic: Money, Bills & Benefits
- Agent: Grok Bot
- Summary: A Grok-based agent checked flight data, argued the issue wasn't extraordinary, emailed, and won €400 before landing.
- Outcome: €400 compensation won after rejection.
- Source: Agent Use Cases / X (https://agentusecases.ai)
- Prompt: Flight [number] on [date] was delayed/cancelled. Draft EU261 claim citing regs and evidence. If rejected, write the appeal with case law-style arguments for my approval.

### Cancel streaming subscriptions with ChatGPT Agent

- URL: http://localhost:43123/use-case/cancel-streaming-agent
- Topic: Money, Bills & Benefits
- Agent: ChatGPT Agent
- Summary: A writer had the agent cancel Peacock, Disney and Netflix after login; needed nudges and may hit site blocks.
- Outcome: Multiple streaming cancels with some friction.
- Source: Agent Use Cases / Tom's Guide (https://agentusecases.ai)
- Prompt: Log into [services], cancel subscriptions, capture confirmation screenshots, and list renewal dates that remain.

### Negotiate a billing credit from AT&T

- URL: http://localhost:43123/use-case/att-billing-credit
- Topic: Money, Bills & Benefits
- Agent: Claude in Chrome
- Summary: Claude in Chrome negotiated a $100 credit from AT&T in a billing dispute with no human involvement in one test write-up.
- Outcome: Reported $100 billing credit.
- Source: Agent Use Cases (https://agentusecases.ai)
- Prompt: Open AT&T billing chat/support. Dispute [charge]. Goal: courtesy credit. Stay polite, escalate once, stop if identity verification needed.

### Cancel an Amazon subscription and get a refund

- URL: http://localhost:43123/use-case/amazon-cancel-refund
- Topic: Money, Bills & Benefits
- Agent: Codex
- Summary: Codex cancelled an Amazon subscription then secured an immediate $15 refund via customer service chat.
- Outcome: Cancel + $15 refund.
- Source: Agent Use Cases / X (https://agentusecases.ai)
- Prompt: Cancel Amazon subscription [name], then request a refund for unused time. Save confirmation emails.

### Cancel and refund subscriptions by searching email and calling

- URL: http://localhost:43123/use-case/grok-cancel-refunds
- Topic: Money, Bills & Benefits
- Agent: Grok Bot
- Summary: Grok Bot read email across accounts, logged into systems, phoned providers and sent emails, cancelling/refunding two long-pending subscriptions in about ten minutes.
- Outcome: Two stuck subscriptions cancelled/refunded ~10 min.
- Source: Agent Use Cases (https://agentusecases.ai)
- Prompt: Search my email for subscriptions I tried to cancel. Finish cancellation and request refunds. Draft any emails for my send approval.

### Write a parking ticket appeal that got dismissed

- URL: http://localhost:43123/use-case/parking-ticket-appeal
- Topic: Money, Bills & Benefits
- Agent: ChatGPT
- Summary: ChatGPT drafted a street-sweeping parking ticket appeal that was dismissed.
- Outcome: Ticket dismissed after appeal.
- Source: Agent Use Cases / Reddit (https://agentusecases.ai)
- Prompt: Draft a polite parking ticket appeal for [city] ticket [details]. Cite relevant code if public. Keep under one page.

### Run a weekly competitive-monitoring briefing

- URL: http://localhost:43123/use-case/competitive-briefing-hermes
- Topic: Research & Knowledge
- Agent: Hermes Agent
- Summary: A PM had Hermes produce weekly competitor briefings; the skill rewrote itself and later surfaced patterns he'd missed.
- Outcome: Briefing time fell across weeks; new patterns surfaced.
- Source: Agent Use Cases (https://agentusecases.ai/topics/research)
- Prompt: Every Monday, brief me on [competitors]: pricing, launches, hiring, complaints. Diff vs last week. Flag non-obvious patterns.

### Research Reddit customer pain points with a skill

- URL: http://localhost:43123/use-case/reddit-pain-points-codex
- Topic: Sales & Growth
- Agent: Codex
- Summary: A Codex skill finds pain points, verifies communities, collects evidence, clusters problems, and analyzes commercial signals.
- Outcome: Structured pain-point research without a heavy custom agent.
- Source: Agent Use Cases (https://agentusecases.ai/topics/research)
- Prompt: Research Reddit pain points for [product idea]. Verify subreddits, collect quotes, cluster problems, score commercial intensity.

### Summarize Reddit threads before writing code

- URL: http://localhost:43123/use-case/reddit-summarize-mcp
- Topic: Coding & Shipping
- Agent: Claude Code + MCP
- Summary: An MCP server lets Claude Code summarize ~30 Reddit threads in ~10 seconds as research before coding.
- Outcome: ~30 threads summarized in seconds; reported large time save.
- Source: Agent Use Cases (https://agentusecases.ai/topics/research)
- Prompt: Summarize the top threads about [library/bug]. Extract solutions, version gotchas, and open issues before I write code.

### Scrape AI news three times a day with a custom bot

- URL: http://localhost:43123/use-case/ai-news-scraper-bots
- Topic: Research & Knowledge
- Agent: Grok Bot
- Summary: A paying Grok Bot user built a simple news scraper collecting AI news three times daily.
- Outcome: Reliable boring news digests on a schedule.
- Source: Agent Use Cases (https://agentusecases.ai/topics/research)
- Prompt: Three times daily, collect top AI news from [sources], dedupe, and send a 10-bullet digest with links.

### Save AI Q&A summaries into git repos as persistent notes

- URL: http://localhost:43123/use-case/git-decision-notes
- Topic: Coding & Shipping
- Agent: ChatGPT + GitHub MCP
- Summary: ChatGPT uses GitHub MCP and per-repo AGENTS.md to write discussion/decision summaries into repos.
- Outcome: Persistent decision memory in-repo.
- Source: Agent Use Cases (https://agentusecases.ai/topics/research)
- Prompt: After this design chat, update AGENTS.md / docs/decisions with what we decided, alternatives rejected, and open questions.

### Ask grounded questions over scientific PDFs

- URL: http://localhost:43123/use-case/paperqa2-pdfs
- Topic: Research & Knowledge
- Agent: PaperQA2
- Summary: PaperQA2 answers questions over local scientific PDFs with cited passages.
- Outcome: Grounded Q&A with citations over paper folders.
- Source: GitHub (https://github.com/Future-House/paper-qa)
- Prompt: Index PDFs in [folder]. Answer [question] using only paper text. Quote short passages with citations.

### Run an AI Scientist idea→experiment→write loop

- URL: http://localhost:43123/use-case/ai-scientist-loop
- Topic: Research & Knowledge
- Agent: AI Scientist
- Summary: Sakana AI Scientist-style systems propose ML ideas, run experiments, and draft papers for human review.
- Outcome: End-to-end research loop templates for ML.
- Source: GitHub (https://github.com/SakanaAI/AI-Scientist)
- Prompt: Run an AI Scientist-style loop on [template] with budget [N] experiments. Return best idea, plots, and draft for review.

### Overnight experiment-evaluate loops editing one training script

- URL: http://localhost:43123/use-case/karpathy-autoresearch
- Topic: Coding & Shipping
- Agent: autoresearch
- Summary: Karpathy-style autoresearch agents edit a training script, run trials overnight, and keep only improvements.
- Outcome: Hundreds of overnight trials with rollback discipline.
- Source: GitHub (https://github.com/karpathy/autoresearch)
- Prompt: You may only edit [train.py]. Hypothesize, change one thing, run, keep only if val improves. Stop after [N] trials.

### Reproduce a paper's experiments with a coding agent

- URL: http://localhost:43123/use-case/reproduce-paper-repo
- Topic: Coding & Shipping
- Agent: Claude Code / Cursor
- Summary: Labs ask Claude Code/Cursor to clone a paper repo, install deps, and reproduce metrics, flagging gaps.
- Outcome: Faster reproduction attempts with explicit blockers.
- Source: Awesome Auto Research (https://github.com/EdwardLeeLPZ/Awesome-Auto-Research)
- Prompt: Reproduce paper [cite]. Clone [repo], run main experiment, compare to Table X. List exact blockers if it fails.

### Draft a structured peer review from a PDF

- URL: http://localhost:43123/use-case/peer-review-draft
- Topic: Writing & Content
- Agent: Claude
- Summary: Reviewers use agents to draft summary/major/minor comments from a manuscript, then rewrite in their own voice.
- Outcome: Faster first-draft reviews; human judgment required.
- Source: Awesome Academic Skills (https://github.com/O0000-code/awesome-academic-skills)
- Prompt: Read manuscript.pdf. Write summary, strengths, major issues, minor issues. Every critique must cite the PDF. No fabricated citations.

### Turn reviewer comments into a rebuttal skeleton

- URL: http://localhost:43123/use-case/rebuttal-skeleton
- Topic: Writing & Content
- Agent: Claude / GPT
- Summary: Authors paste reviewer reports into an agent to map each comment to replies and manuscript changes.
- Outcome: Organized response-letter outline quickly.
- Source: Awesome Academic Skills (https://github.com/O0000-code/awesome-academic-skills)
- Prompt: Build a response table: Reviewer | Comment | Reply | Manuscript change. Mark where new experiments are needed.

### Run a de-AI tone pass before publishing

- URL: http://localhost:43123/use-case/de-ai-polish
- Topic: Writing & Content
- Agent: Claude / GPT
- Summary: Writers use agents to rewrite AI-sounding prose into tighter human/academic voice while preserving claims.
- Outcome: Cleaner tone; claims unchanged.
- Source: Awesome Academic Skills (https://github.com/O0000-code/awesome-academic-skills)
- Prompt: Rewrite for natural precise English. Do not add claims. Flag generic or unsupported sentences.

### Batch social posts from one long article

- URL: http://localhost:43123/use-case/social-content-batch
- Topic: Writing & Content
- Agent: Claude / GPT
- Summary: Creators use agents to turn a blog/video transcript into a week of X/LinkedIn posts with hooks and CTAs.
- Outcome: A week of drafts from one source doc.
- Source: Common agent workflow (https://agentusecases.ai)
- Prompt: From this article, write 7 X posts and 3 LinkedIn posts. Vary hooks. No fake metrics. Include a soft CTA to [URL].

### Run a use-case directory monetized with rising-price sponsor slots

- URL: http://localhost:43123/use-case/stripe-sponsor-directory
- Topic: Building Agents
- Agent: Grok Bot
- Summary: Alex Finn's Grok Bot built agentusecases.ai and sold weekly site+newsletter sponsor slots starting around $500 and rising after each sale.
- Outcome: Publicly claimed early sponsorship revenue (~$1.1k–$1.8k) with large X-driven traffic.
- Source: agentusecases.ai (https://agentusecases.ai)
- Prompt: Act as CEO of an AI-agent use-case directory: add verified cases daily, ship Daily 5 email, improve SEO, and prepare sponsor outreach for my approval.

### Monetize a Cursor ecosystem directory with ads and listings

- URL: http://localhost:43123/use-case/cursor-directory-ads
- Topic: Building Agents
- Agent: N/A (directory business)
- Summary: Cursor Directory is a widely cited directory of Cursor rules/resources that publicly reports material MRR from ads and listings.
- Outcome: Self-reported $35K+ MRR from ads/listings in indie roundups.
- Source: Cursor Directory (https://cursor.directory)
- Prompt: Maintain a curated directory of [Cursor rules/MCP/skills]. Add daily finds, feature sponsor slots, and keep submission + SEO pages fresh.

### Spin up a background research agent while computers are busy

- URL: http://localhost:43123/use-case/background-research-dot
- Topic: Building Agents
- Agent: Dot
- Summary: A Dot user created a native background research agent with read-only web tools while other computer environments were occupied.
- Outcome: Concurrent research worker without computer control.
- Source: Agent Use Cases (https://agentusecases.ai)
- Prompt: Create a background research agent on [topic] with read-only web tools. No computer control. Summarize findings with source links.

### Turn vet charges into ready-to-send pet insurance claims

- URL: http://localhost:43123/use-case/pet-insurance-claims
- Topic: Ops & Admin
- Agent: OpenClaw
- Summary: An OpenClaw agent spots vet charges in the bank feed, finds invoices in Gmail, fills insurer PDFs, drafts email, and tracks replies.
- Outcome: Claim packets ready for owner send approval.
- Source: Agent Use Cases / GitHub (https://agentusecases.ai)
- Prompt: When a vet charge appears, find the invoice, fill the insurer form, draft the claim email, and wait for my send approval.

### Build expense reports from forwarded receipts

- URL: http://localhost:43123/use-case/expense-report-receipts
- Topic: Ops & Admin
- Agent: Claude Cowork
- Summary: Users report Cowork building Monday expense reports from forwarded receipts and tax-prep summaries from a year of transactions.
- Outcome: Recurring expense packs from email receipts.
- Source: Agent Use Cases (https://agentusecases.ai)
- Prompt: From this week's receipt emails, build an expense report by category with totals and missing-receipt flags.

### Find cheaper car insurance with an agent

- URL: http://localhost:43123/use-case/cheaper-car-insurance
- Topic: Money, Bills & Benefits
- Agent: Meta Muse
- Summary: A founder said Muse found car insurance costing thousands less per year than the original plan.
- Outcome: Reported ~$3,500/year cheaper plan found.
- Source: Agent Use Cases (https://agentusecases.ai)
- Prompt: Compare car insurance quotes for [car/drivers/zip]. Rank by annual cost and coverage gaps. Do not purchase without approval.

### Appeal a denied travel insurance claim

- URL: http://localhost:43123/use-case/travel-insurance-appeal
- Topic: Money, Bills & Benefits
- Agent: Gemini
- Summary: A poster used Gemini during an appeal of a rejected travel insurance claim and said the appeal succeeded.
- Outcome: Appeal reportedly succeeded.
- Source: Agent Use Cases / Reddit (https://agentusecases.ai)
- Prompt: Draft an appeal for denied travel claim [id] using policy text and my timeline. Quote policy clauses. Keep factual.

### Produce investment memos with Claude Cowork

- URL: http://localhost:43123/use-case/investment-memo-cowork
- Topic: Research & Knowledge
- Agent: Claude Cowork
- Summary: Guides walk non-developers through using Cowork to produce a full investment memo in about 15 minutes.
- Outcome: Fast first-pass investment memo for human edit.
- Source: Agent Use Cases (https://agentusecases.ai)
- Prompt: Write an investment memo on [company] covering business, market, risks, and questions. Cite sources. Mark speculation.