Skip to main content

50+ Agent Use Cases

info

This content was developed with AI assistance and is regularly reviewed for accuracy.

Snapshot: October 2026. Model names, prices, and product features change monthly; What's New in AI is the maintained chronology.

This page is a catalog of 52 agent recipes, grouped by function. Each one gives the goal, the workflow the agent follows, and a 2026 build path, from the agent mode in an app you already pay for to a Claude Code routine or an automation tool with an agent step. Use it to find a first project, or to show a colleague what is realistic now.

Learning focus​

  • Spot recurring tasks in your own work that fit an agent
  • Match each recipe to a build path: app agent mode, scheduled task, automation tool, or coding agent
  • See where approval gates belong before an agent acts

How to read the catalog​

Every recipe uses the same format. Workflow lists the steps in order. Build with names tools that can do it today, cheapest and simplest first. Where a recipe sends, spends, deletes, or decides about people, a Guardrail line says where a human approval belongs.

Most recipes fit one of four build paths. A scheduled task or routine in Claude, ChatGPT, or Gemini runs a brief on a timer with your connected apps; this is the easiest start for reports, monitoring, and inbox work. An automation tool with an agent step (Zapier Agents, n8n's AI Agent node, Make, Relay.app) suits flows that must be repeatable and auditable. A coding agent such as Claude Code, Codex, or Copilot handles software work, often triggered from GitHub. And a team-chat agent such as Claude Tag in Slack answers questions where your team already works. If the steps never change, a plain workflow may beat an agent; Workflows vs Agents helps you decide.

Software development​

Coding is where agents are most mature, so these recipes are the most reliable to run unattended, provided tests and branch protection are in place.

1. Bug triage agent​

Goal: Categorize and prioritize new bug reports. Workflow: Watch the issue tracker → read logs and stack traces → search the codebase for related code → label severity and component → assign a team → add reproduction steps. Build with: A Claude Code Routine on a GitHub trigger; GitHub Copilot or Codex via GitHub's agent dispatch.

2. Code review agent​

Goal: Review every pull request for bugs, security issues, and style. Workflow: Fetch the diff → check for security and correctness issues → run static analysis → suggest fixes with examples → post review comments. Build with: Claude Code's Code Review (generally available October 15, 2026 for Team and Enterprise), GitHub Copilot code review, or the bundled /code-review skill in Claude Code. Guardrail: The agent reviews; a human approves the merge.

3. Documentation updater​

Goal: Keep docs in sync with code changes. Workflow: Detect changed code in a PR → find outdated docs → regenerate reference sections → draft changelog entries → open a docs PR. Build with: Claude Code in GitHub Actions or a Routine on pull-request events; Cursor.

4. Dependency update agent​

Goal: Keep dependencies secure and current. Workflow: Monitor security advisories → check new versions → read breaking changes → update on a branch → run the test suite → open a PR with notes. Build with: Dependabot plus a Claude Code Routine that fixes breaking changes.

5. Test generation agent​

Goal: Raise test coverage on untested code. Workflow: Find uncovered functions → identify edge cases → write tests in the project's style → run them → fix failures → commit. Build with: Claude Code with /goal set to a coverage target, Codex, Cursor.

Marketing and sales​

Marketing recipes are mostly research and drafting, which agents do well. Keep a human between the draft and anything published or sent.

6. Competitor monitor​

Goal: Track competitor changes weekly. Workflow: Visit competitor sites → note pricing and feature changes → scan social mentions → compile a weekly report → alert on major changes. Build with: A Claude or ChatGPT scheduled task with web search; Claude in Chrome for pages behind navigation; Apify for large-scale scraping.

7. Content calendar agent​

Goal: Plan and draft a month of social content. Workflow: Research trending topics → generate post ideas → write platform-specific variations → draft graphics → queue for scheduling → review performance. Build with: A Claude project with a brand-voice skill plus Buffer; Zapier Agents. Guardrail: Approve posts before they publish.

8. Lead enrichment agent​

Goal: Research and qualify new leads. Workflow: New lead enters the CRM → research the company site → identify decision makers → note likely needs → score the lead → draft personalized outreach. Build with: Clay; Zapier Agents with your CRM; a CRM connector in Claude.

9. Email campaign optimizer​

Goal: Improve email performance over time. Workflow: Analyze past campaigns → write subject-line variants → set up A/B tests → monitor open and click rates → recommend the winner. Build with: Mailchimp or Klaviyo plus Claude; an n8n workflow with an AI Agent node.

10. Customer interview synthesizer​

Goal: Turn sales and support calls into product insights. Workflow: Collect call transcripts → extract pain points and feature requests → group by theme → update the roadmap doc → notify the product team. Build with: Gong or Granola transcripts plus Claude with a Notion connector.

Customer support​

Support agents touch customers directly, so they need the tightest scoping. For a step-by-step build of the first recipe, see the inquiry router tutorial.

11. Ticket routing agent​

Goal: Categorize and route incoming tickets. Workflow: New ticket arrives → read content and sentiment → set urgency and category → check for a known solution → route to the right team → suggest a reply. Build with: Zendesk or Intercom AI features; Sierra; Zapier or n8n with a Claude classifier step.

12. Knowledge base maintainer​

Goal: Keep help articles accurate and complete. Workflow: Review unresolved tickets → find knowledge gaps → draft new articles → flag outdated ones → notify the content team. Build with: A weekly Claude scheduled task with your help-center and Notion connectors.

13. Refund processor​

Goal: Handle refund requests within policy. Workflow: Receive request → check purchase history → verify policy → calculate amount → queue for approval → process payment → confirm by email. Build with: Stripe plus Relay.app or n8n with a human approval step. Guardrail: A person approves every payout.

14. Customer sentiment analyzer​

Goal: Track satisfaction trends across channels. Workflow: Collect feedback from all channels → score sentiment and themes → spot rising issues → write a weekly report → alert on negative trends. Build with: A Claude Routine with your support and survey connectors; a BI dashboard for the output.

Content creation​

Content agents save the most time on research and first drafts. The final edit is still yours, along with disclosure where your audience expects it.

15. Blog post drafter​

Goal: Research and draft blog posts. Workflow: Pick a topic → research competing content → outline → draft with examples → write SEO metadata → generate a featured image → stage in the CMS. Build with: Claude or ChatGPT with web search and a style skill; the WordPress API; image generation in the same chat app or an image API (see Development and APIs). Guardrail: Stage as a draft; a person publishes.

16. Newsletter curator​

Goal: Compile a weekly industry newsletter. Workflow: Monitor news sources → find the week's key stories → summarize → add commentary → format for email → schedule. Build with: A Claude Routine or ChatGPT scheduled task plus Beehiiv.

17. Video script writer​

Goal: Turn a topic into a ready-to-film script. Workflow: Research topic and audience → outline → write the script with a strong opening → suggest B-roll → propose thumbnails → draft the description and tags. Build with: A Claude project with a scriptwriting skill; a Gemini Gem.

18. Podcast show notes agent​

Goal: Produce show notes from an episode. Workflow: Transcribe → find topics and timestamps → pull quotes → summarize → suggest social clips → format the show-notes page. Build with: Descript transcripts plus Claude; a CMS integration.

19. Social media repurposer​

Goal: Turn long content into platform-specific posts. Workflow: Read the post or transcript → extract key points → write posts per platform → suggest quote graphics → queue for scheduling. Build with: A repurposing skill in Claude plus Buffer; Zapier Agents.

20. SEO content optimizer​

Goal: Improve rankings on existing pages. Workflow: Audit page performance → compare with ranking competitors → find keyword gaps → suggest edits → rewrite weak sections → update metadata. Build with: Search Console data plus Claude; Surfer SEO.

Data and analytics​

Reporting is the most common first scheduled agent. The weekly report playbook builds recipe 21 step by step.

21. Report generator​

Goal: Produce a recurring business report. Workflow: Query data sources → compute key metrics → chart them → flag anomalies → write an executive summary → send to stakeholders. Build with: A Claude Routine or scheduled task with spreadsheet and database connectors; Microsoft 365 Copilot with Power BI.

22. Data quality monitor​

Goal: Catch data problems before they reach reports. Workflow: Run daily checks → find anomalies, gaps, and duplicates → compare with expected ranges → alert the data team → suggest fixes. Build with: dbt tests plus a Claude Code Routine that investigates failures.

23. Dashboard builder​

Goal: Build a dashboard from a plain-language request. Workflow: Gather requirements → identify data sources → design the layout → build charts and filters → document usage. Build with: Claude artifacts for quick prototypes; Looker or Tableau with their built-in agents for production.

24. Survey analysis agent​

Goal: Turn survey responses into findings. Workflow: Import responses → clean and categorize → find themes in open answers → compute summary statistics → chart → list actions. Build with: Typeform or Google Forms exports plus Claude's code execution.

25. Natural-language data assistant​

Goal: Let non-technical colleagues ask questions of a database. Workflow: Accept a question → read the schema → write SQL → run it read-only → format results → explain the answer. Build with: An MCP database server connected to Claude through a read-only database user. Guardrail: Read-only credentials, always.

Finance and operations​

Finance agents should prepare and check, not pay. Every recipe here keeps money movement behind a person.

26. Invoice processor​

Goal: Extract and record invoice data. Workflow: Receive invoice → extract vendor, amount, and date → categorize → match to a purchase order → route for approval → schedule payment. Build with: Zapier Agents with your accounting software; Vic.ai. Guardrail: Approval before any payment is scheduled.

27. Expense report reviewer​

Goal: Validate expense submissions. Workflow: Receive the report → match receipts to claims → check policy → calculate reimbursement → flag issues → recommend approve or return. Build with: Expensify or Ramp plus Claude.

28. Budget variance analyzer​

Goal: Explain spending against budget. Workflow: Import actuals → compare to budget → calculate variances → find concerning trends → write the explanation → alert budget owners. Build with: QuickBooks or NetSuite exports plus Claude; Microsoft 365 Copilot in Excel.

29. Contract review agent​

Goal: Flag key terms and risks in a contract. Workflow: Upload the contract → extract dates, amounts, and obligations → compare to your standard template → flag unusual clauses → write a summary memo. Build with: A Claude project holding your template and playbook, plus a contract-review skill. Guardrail: A qualified reviewer signs off; the agent prepares.

30. Vendor onboarding agent​

Goal: Speed up new vendor setup. Workflow: Collect vendor details → verify business credentials → run compliance checks → create system records → draft the contract → schedule a kickoff. Build with: n8n with an AI Agent node; Relay.app for approval steps.

HR and recruitment​

HR recipes involve decisions about people, which brings bias, privacy, and in some places legal obligations. See AI for Human Resources before deploying any of them.

31. Resume screener​

Goal: Summarize applications against the job's requirements. Workflow: Receive application → parse resume → match to requirements → extract key qualifications → write an interviewer brief. Build with: Greenhouse or Lever plus Claude. Guardrail: The agent summarizes; people decide. Audit outcomes for bias.

32. Interview scheduler​

Goal: Coordinate interview logistics. Workflow: Candidate advances → check interviewer calendars → offer times → book the meeting → send invites and prep materials. Build with: Calendar connectors in Claude or Gemini; Microsoft 365 Copilot; Calendly.

33. Onboarding coordinator​

Goal: Run the new-hire checklist. Workflow: Trigger before start date → request accounts and access → send welcome email → schedule orientation → assign training → track completion. Build with: BambooHR plus Zapier or n8n.

34. Employee FAQ agent​

Goal: Answer common policy questions with sources. Workflow: Receive question → search the handbook → answer with the policy citation → escalate complex or sensitive cases → log gaps. Build with: Claude Tag in Slack (beta for Team and Enterprise) with the handbook in a project; Microsoft Copilot agents in Teams.

35. Performance review drafter​

Goal: Produce first drafts for managers. Workflow: Gather goals, metrics, and feedback → compare with objectives → draft each section with examples → suggest development areas → hand to the manager. Build with: Workday or your HRIS exports plus Claude, with sensitive-data settings reviewed.

Personal productivity​

These are the easiest recipes to start with, because the only data involved is yours. The first agent tutorial builds a version of recipe 36.

36. Email triage agent​

Goal: Sort and prioritize your inbox. Workflow: Scan new email → categorize by type and urgency → draft replies to routine requests → extract action items → add them to your task list. Build with: A Claude scheduled task with the Gmail connector; ChatGPT Work; Gemini in Gmail. Guardrail: Drafts only; you send.

37. Meeting prep agent​

Goal: Arrive prepared for every meeting. Workflow: Read today's calendar → find agendas and documents → research attendees and topics → write talking points → send a pre-read. Build with: A morning Claude scheduled task with calendar and Drive connectors; Microsoft 365 Copilot.

38. Task prioritizer​

Goal: Plan each day around real priorities. Workflow: Review open tasks → weigh deadlines and dependencies → produce a prioritized plan → time-block the calendar → send a daily brief. Build with: Todoist or Asana connectors plus Claude; see Organize Your Week with AI.

39. Reading list curator​

Goal: Find and summarize reading worth your time. Workflow: Monitor chosen topics → find quality articles → summarize key points → group by theme → estimate reading time. Build with: Readwise Reader plus a weekly Claude Routine.

40. Personal CRM agent​

Goal: Keep professional relationships warm. Workflow: Track interactions → set follow-up reminders → suggest times to reconnect → draft catch-up messages → note milestones. Build with: Airtable or a spreadsheet plus Claude with email and calendar connectors.

Education and learning​

Education agents work best as tutors and preparers. Keep assessment decisions with the teacher.

41. Study guide generator​

Goal: Turn course material into study aids. Workflow: Read course content → extract key concepts → write practice questions → make flashcards → plan a study schedule. Build with: NotebookLM; a Claude project with the course files.

42. Research assistant​

Goal: Survey the literature on a question. Workflow: Clarify the question → search academic sources → select relevant papers → extract findings → summarize with citations → note gaps. Build with: Deep-research modes in Claude, ChatGPT, or Gemini; Semantic Scholar; Zotero. Guardrail: Check every citation exists and says what the summary claims.

43. Lesson plan creator​

Goal: Design lessons aligned to objectives. Workflow: Set objectives → research approaches → outline the lesson → create activities and assessments → align to standards. Build with: A lesson-planning skill in Claude; a Gemini Gem.

44. Assignment feedback agent​

Goal: Draft detailed feedback against a rubric. Workflow: Receive submission → evaluate against the rubric → note strengths and gaps with examples → suggest resources → draft a grade rationale. Build with: Canvas exports plus Claude with a rubric skill. Guardrail: The teacher reviews and owns the grade.

45. Language learning coach​

Goal: Practice conversation with corrections. Workflow: Hold a conversation in the target language → spot errors → correct with explanations → suggest alternatives → adjust difficulty. Build with: Voice mode in Claude, ChatGPT, or Gemini with a standing instruction in a project.

E-commerce and retail​

Retail recipes mix content work, which is low risk, with pricing and inventory, which needs approval.

46. Product description writer​

Goal: Write consistent, search-friendly product copy. Workflow: Read product data → check competitor listings → identify key benefits → write the description and bullet points → suggest cross-sells. Build with: Shopify plus a dynamic workflow or batch run in Claude Code for large catalogs.

47. Inventory alert agent​

Goal: Prevent stockouts and overstock. Workflow: Monitor stock → analyze sales velocity → predict stockout dates → factor lead times → recommend purchase orders → alert procurement. Build with: ERP exports plus a daily Claude Routine.

48. Review response agent​

Goal: Reply to customer reviews promptly. Workflow: Monitor new reviews → read sentiment and issues → draft replies → queue for approval. Build with: Trustpilot or Google Business reviews plus Claude, with a moderation queue.

49. Price optimization agent​

Goal: Recommend price changes from market signals. Workflow: Monitor competitor prices → track demand → model margin impact → recommend changes within rules → test. Build with: Prisync plus Claude. Guardrail: A person approves price changes.

50. Abandoned cart recovery​

Goal: Win back abandoned purchases. Workflow: Detect abandonment → read cart contents → choose an offer within rules → write a personalized email → schedule reminders → track conversion. Build with: Klaviyo plus Claude.

Healthcare and wellness​

Healthcare recipes handle protected health information. Use only tools covered by the appropriate agreements, and read AI for Healthcare first.

51. Appointment reminder agent​

Goal: Reduce no-shows. Workflow: Check upcoming appointments → send personalized reminders → include directions and prep → offer rescheduling → update the schedule. Build with: Practice-management software plus Twilio and Claude.

52. Clinical documentation assistant​

Goal: Draft visit notes from a recorded conversation. Workflow: Record the visit with consent → extract clinical details → structure as a SOAP note → suggest billing codes → flag items for review → file to the EHR. Build with: Specialized ambient-scribe products on compliant infrastructure. Guardrail: The clinician reviews and signs every note.

Building your own​

Pick the recipe closest to a task you already do every week, then follow five steps. Write out the steps you take today, which become the agent's brief. List the data and tools it needs, which become its connectors, scoped to the minimum. Mark every step that sends, spends, deletes, or decides about a person, which become approval gates. Run it attended three times before you schedule it. Then keep a short list of what went wrong, which becomes your eval set.

If you want more ready-made know-how, community skill directories such as skills.sh list installable skills for many of these recipes. Read any skill before installing it, as Skills, AGENTS.md, and CLAUDE.md explains.

Key takeaways​

  • Most agent recipes fall into four build paths: scheduled tasks in your assistant, automation tools with agent steps, coding agents, and team-chat agents.
  • Reports, monitoring, triage, and drafting are the safest first projects.
  • Anything that sends, spends, deletes, or judges people needs a human approval step.
  • Start from a task you already do, run it attended first, and keep a list of failures as your eval set.

From the author​

Want to try a browsing agent like the competitor monitor (recipe 6) without building anything? FreshVisit, built by the author of this site, is a working version. Point it at a competitor's pricing page with a task like "find where they explain their tiers" and watch it click through, screen by screen. Put the run on a weekly repeat and it reports which links, buttons, and prices changed.

Next steps​