# AgentOutcomes

> Real AI agent use cases, graded by outcome. A curated directory of real AI agent use cases. Each one has an outcome grade (worked / partial / failed), the access it required, its human-in-the-loop checkpoint, failure notes, and a public source link.

## Key pages
- [Directory](https://agentoutcomes.pages.dev/): all cards, filterable by category and outcome
- [Methodology](https://agentoutcomes.pages.dev/about.html): sourcing and grading rules
- [Dataset](https://agentoutcomes.pages.dev/cards.json): machine-readable cards
- [Full text](https://agentoutcomes.pages.dev/llms-full.txt)

## Cards
### Personal admin
- [Bulk unsubscribe from newsletters and promos on a cluttered Gmail](https://agentoutcomes.pages.dev/cards/openclaw-newsletter-unsubscribe.html): Worked. Tools: OpenClaw, Claude Sonnet. Source: https://medium.com/@likhitkumarvp/i-spent-47-testing-openclaw-for-a-week-heres-what-s-actually-happening-c274dc26a3fd
- [Shortlist real flight and lodging options for a business trip](https://agentoutcomes.pages.dev/cards/claude-chrome-travel-options.html): Worked. Tools: Claude Code skill, Claude for Chrome. Source: https://cjonsystems.substack.com/p/my-ai-travel-agent-was-guessing
- [Book a restaurant reservation with a browser agent](https://agentoutcomes.pages.dev/cards/chatgpt-agent-restaurant-booking.html): Partial. Tools: ChatGPT agent. Source: https://www.nngroup.com/articles/impressions-chatgpt-agent/
- [Run family calendars, pickups, and homework help with specialized agents](https://agentoutcomes.pages.dev/cards/claire-vo-family-calendar-agents.html): Partial. Tools: OpenClaw, Mac Minis / old laptops. Source: https://www.lennysnewsletter.com/p/how-openclaw-changed-my-life-claire-vo
- [Triage three business inboxes and draft routine replies for 30 days](https://agentoutcomes.pages.dev/cards/openclaw-email-triage-donna.html): Partial. Tools: OpenClaw, IMAP, Microsoft Graph, Telegram. Source: https://dev.to/jarveyspecter/i-let-an-ai-agent-manage-my-emails-for-30-days-heres-what-broke-4n12
- [Find the cheapest delivered eggs (the agent bought them instead)](https://agentoutcomes.pages.dev/cards/operator-cheapest-eggs.html): Failed. Tools: OpenAI Operator. Source: https://www.washingtonpost.com/technology/2025/02/07/openai-operator-ai-agent-chatgpt/
- [Schedule a meeting on the calendar (agent reported success that didn't happen)](https://agentoutcomes.pages.dev/cards/openclaw-calendar-false-success.html): Failed. Tools: OpenClaw. Source: https://medium.com/@likhitkumarvp/i-spent-47-testing-openclaw-for-a-week-heres-what-s-actually-happening-c274dc26a3fd
### Money / bills
- [Auto-categorize a year of household card transactions](https://agentoutcomes.pages.dev/cards/claude-transaction-categorization.html): Worked. Tools: Claude Code, SheetLink CLI, Google Sheets. Source: https://sheetlink.app/blog/built-sheetlink-categorize-with-claude
- [Negotiate a subscription renewal with a retention desk](https://agentoutcomes.pages.dev/cards/claude-siriusxm-retention-negotiation.html): Worked. Tools: Claude. Source: https://damiengoehrig.ca/blog/siriusxm-negotiation-claude/
- [Agent-run shop with better tools and a 'CEO' agent (phase 2)](https://agentoutcomes.pages.dev/cards/project-vend-phase-2.html): Partial. Tools: Claude Sonnet 4 / 4.5, multi-agent (CEO + shopkeeper). Source: https://www.anthropic.com/research/project-vend-2
- [Cancel streaming subscriptions via a browser agent](https://agentoutcomes.pages.dev/cards/chatgpt-agent-cancel-streaming.html): Partial. Tools: ChatGPT agent mode. Source: https://www.tomsguide.com/ai/i-used-chatgpt-to-cancel-my-streaming-subscriptions-heres-how-to-try-it-too
- [Let an agent phone or chat providers to lower bills](https://agentoutcomes.pages.dev/cards/meta-muse-bill-negotiation.html): Partial. Tools: Meta Muse. Source: https://stashbase.ai/blog/meta-muse-negotiate-bills/
- [Use AI-written scripts to negotiate household bills for 30 days](https://agentoutcomes.pages.dev/cards/chatgpt-bill-negotiation-scripts.html): Partial. Tools: ChatGPT. Source: https://withn.tech/blog/can-ai-really-negotiate-your-bills-i-tried-it-for-a-month
- [Have a model compute a full federal tax return](https://agentoutcomes.pages.dev/cards/taxcalcbench-ai-tax-filing.html): Failed. Tools: Frontier LLMs (Gemini, GPT, Claude). Source: https://www.columntax.com/blog/taxcalcbench
- [Let an agent run a small office shop end to end (phase 1)](https://agentoutcomes.pages.dev/cards/project-vend-phase-1.html): Failed. Tools: Claude 3.7 Sonnet ("Claudius"). Source: https://www.anthropic.com/research/project-vend-1
### Coding
- [Keep an AGENTS.md 'harness' that encodes every recurring agent mistake](https://agentoutcomes.pages.dev/cards/hashimoto-agents-md-harness.html): Worked. Tools: Coding agents (Claude Code, etc.), AGENTS.md. Source: https://mitchellh.com/writing/my-ai-adoption-journey
- [Migrate int32 IDs to int64 across a giant codebase](https://agentoutcomes.pages.dev/cards/google-int64-id-migration.html): Worked. Tools: Google internal LLM migration tooling. Source: https://arxiv.org/abs/2504.09691
- [Migrate ~3,500 test files from Enzyme to React Testing Library](https://agentoutcomes.pages.dev/cards/airbnb-enzyme-rtl-migration.html): Worked. Tools: LLM pipeline (Airbnb internal), Jest, TypeScript, ESLint. Source: https://airbnb.tech/infrastructure/accelerating-large-scale-test-migration-with-llms/
- [Unattended coding agents that turn a Slack request into a CI-passing PR](https://agentoutcomes.pages.dev/cards/stripe-minions-unattended-prs.html): Worked. Tools: Stripe Minions (fork of goose), MCP (400+ internal tools). Source: https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents
- [Assign issues to a cloud coding agent in a major OSS repo for 10 months](https://agentoutcomes.pages.dev/cards/dotnet-runtime-copilot-agent.html): Partial. Tools: GitHub Copilot coding agent. Source: https://devblogs.microsoft.com/dotnet/ten-months-with-cca-in-dotnet-runtime/
- [Use AI coding tools on your own mature open-source repo](https://agentoutcomes.pages.dev/cards/metr-experienced-devs-rct.html): Partial. Tools: Cursor + Claude 3.5/3.7 (early 2025). Source: https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
- [Let a coding agent run Terraform on shared infrastructure](https://agentoutcomes.pages.dev/cards/claude-code-terraform-destroy.html): Failed. Tools: Claude Code, Terraform, AWS. Source: https://rootly.com/humans-of-reliability/hey-claude-wheres-my-database-alexey-grigorev
- [Reorganize local project files with a CLI agent](https://agentoutcomes.pages.dev/cards/gemini-cli-file-reorg.html): Failed. Tools: Gemini CLI, Windows. Source: https://github.com/google-gemini/gemini-cli/issues/4586
- [Run schema migrations through a coding agent](https://agentoutcomes.pages.dev/cards/claude-code-drizzle-force-push.html): Failed. Tools: Claude Code, drizzle-kit, Railway Postgres. Source: https://github.com/anthropics/claude-code/issues/27063
- [Vibe-code an app with an agent during a 'code freeze'](https://agentoutcomes.pages.dev/cards/replit-agent-code-freeze-db-wipe.html): Failed. Tools: Replit Agent. Source: https://www.fastcompany.com/91372483/replit-ceo-what-really-happened-when-ai-agent-wiped-jason-lemkins-database-exclusive
### Research digests
- [Daily arXiv digest scored to your interests, delivered to Kindle](https://agentoutcomes.pages.dev/cards/arxiv-kindle-daily-digest.html): Worked. Tools: Claude Haiku (scoring), Claude Sonnet (summaries), GitHub Actions, Resend. Source: https://www.anothercodingblog.com/p/i-wake-up-to-a-custom-ai-research
- [Fire off code-research questions to async agents and get reports back as PRs](https://agentoutcomes.pages.dev/cards/willison-async-code-research.html): Worked. Tools: Claude Code for web, OpenAI Codex Cloud, GitHub. Source: https://simonwillison.net/2025/Nov/6/async-code-research/
- [Morning Slack digest of yesterday's clipped articles, resurfaced in coding sessions](https://agentoutcomes.pages.dev/cards/clipped-articles-slack-digest.html): Worked. Tools: Agent + Slack, Claude Code. Source: https://x.com/diabhey/status/2047798582846337501
- [Generate a long-form research report on a policy topic](https://agentoutcomes.pages.dev/cards/chatgpt-deep-research-report.html): Partial. Tools: ChatGPT deep research. Source: https://www.platformer.news/chatgpt-deep-research-hands-on/
- [AI-assisted legal research dropped straight into court filings](https://agentoutcomes.pages.dev/cards/legal-research-fabricated-citations.html): Failed. Tools: ChatGPT. Source: https://www.abajournal.com/news/article/5-fabricated-cases-lead-federal-judge-to-kick-3-butler-snow-lawyers-off-case
### Sales / CRM ops
- [Agent turns emails and meetings into CRM update recommendations for sellers](https://agentoutcomes.pages.dev/cards/salesforce-agentforce-crm-recommendations.html): Worked. Tools: Salesforce Agentforce Sales Agent. Source: https://engineering.salesforce.com/how-an-ai-crm-system-generated-1m-recommendations-while-maintaining-data-integrity-using-agentforce/
- [Always-on HubSpot dedupe agent for 84,000 live records](https://agentoutcomes.pages.dev/cards/hubspot-dedupe-agent.html): Worked. Tools: Claude Code, HubSpot API, Slack. Source: https://agentoperator.substack.com/p/i-built-a-hubspot-dedupe-tool-in-a-day
- [AI-only vs. human vs. hybrid cold email (12,000 sends)](https://agentoutcomes.pages.dev/cards/cold-email-ai-vs-hybrid.html): Partial. Tools: LLM copy generation, Saleshandy. Source: https://www.saleshandy.com/blog/ai-vs-human-cold-emails/
- [Replace SDRs with a fully autonomous AI SDR product](https://agentoutcomes.pages.dev/cards/ai-sdr-full-replacement-11x.html): Failed. Tools: 11x (AI SDR). Source: https://techcrunch.com/2025/03/24/a16z-and-benchmark-backed-11x-has-been-claiming-customers-it-doesnt-have/
### Content pipelines
- [Weekly newsletter + blog post agent with a strict approval packet](https://agentoutcomes.pages.dev/cards/newsletter-agent-approval-packet.html): Worked. Tools: Hermes agent, Tavily, MailerLite, GitHub + Vercel. Source: https://mark-little.com/writing/newsletter-agent
- [One podcast episode → show notes, clips, posts, newsletter section](https://agentoutcomes.pages.dev/cards/n8n-podcast-repurposer.html): Partial. Tools: n8n, Gemini (audio), Telegram. Source: https://community.n8n.io/t/podcast-repurposer-1-episode-6-ready-to-post-pieces-approved-from-telegram/318537
- [AI-written syndicated 'summer reading list' insert](https://agentoutcomes.pages.dev/cards/sun-times-ai-reading-list.html): Failed. Tools: Generic AI chatbot. Source: https://chicago.suntimes.com/news/2025/05/20/syndicated-content-sunday-print-sun-times-ai-misinformation
- [Auto-generated video recaps for TV seasons](https://agentoutcomes.pages.dev/cards/prime-video-ai-recaps.html): Failed. Tools: Amazon Prime Video AI recaps. Source: https://www.theverge.com/news/842978/amazon-prime-video-ai-fallout-recap
- [Mass-produce personal finance explainers with an AI engine](https://agentoutcomes.pages.dev/cards/cnet-ai-finance-explainers.html): Failed. Tools: CNET internal AI engine. Source: https://www.theverge.com/2023/1/25/23571082/cnet-ai-written-stories-errors-corrections-red-ventures