Most people open Claude, ask one question, read the answer, close the tab.
That is not what this is about.
This is about the other use case. The one where Claude is not something you talk to. It is a system that runs without you, makes decisions inside boundaries you set once, and produces work or income while you sleep.
Opus 4.8 is the first version where this actually holds up. Longer autonomous runs. Better tool use. Fewer broken steps in multi stage tasks. The gap between a cool demo and a system that runs unattended for hours without falling apart finally closed enough that normal people can build these on a laptop.
Here are 60 of them, in 12 categories. Every one is specific enough to build. None of them are Claude being smarter than you. All of them are Claude removing you from work that never needed you in real time.
One rule before you start. Every workflow below is a trigger, an agent, and a verification step. Skip the verification and you get expensive mistakes running unattended. Keep it and you get a system you can actually trust overnight.
Content Machines
1. Ingest and publish pipeline. Drop a transcript, article, or PDF into a folder. Claude extracts the core ideas, drafts a thread, a post, and a newsletter blurb, and saves all three to an output folder. You wake up to drafts, not a blank page.
2. Competitor monitor. A daily scheduled run that checks five competitor accounts or blogs, summarizes anything new, and flags overlap with topics you have not covered yet.
3. Repurposing engine. One long piece in. Ten short variations out. A thread, three standalone posts, a carousel script, a short video script. One input, ten outputs, zero extra thinking.
4. SEO cluster builder. Give Claude one seed keyword. It maps the full topic cluster and drafts ten interlinked articles targeting different long tail variations, ready to publish.
5. Newsletter autopilot. Connect your RSS feeds. Every morning Claude drafts the newsletter from what happened in your niche overnight. You edit for ten minutes and send.
Trading and Market Analysis
6. Wallet pattern scanner. Point Claude at on chain transaction data. It logs wallets with consistently high win rates and their entry and exit timing to a file you review each morning.
7. Overnight chart review. Queue a list of tickers before bed. Claude pulls price data, runs technical analysis across timeframes, and writes a morning briefing.
8. News to thesis pipeline. A loop that watches financial headlines, filters for your watchlist, and drafts a short thesis note on potential impact before the market fully prices it.
9. Earnings call summarizer. Drop in a transcript the moment it drops. Claude extracts guidance changes, tone shifts from the last call, and anything that contradicts prior statements.
10. Portfolio rebalancing alert. A scheduled check comparing current allocation to target, drafting the exact trades needed. It never executes. You stay in control of every dollar.
Code That Ships Itself
11. Overnight bug sweep. Point Claude Code at your repo before bed. It finds and fixes lint errors, failing tests, and obvious bugs, then opens a pull request you review over coffee.
12. Dependency update loop. A weekly run that updates packages one at a time, runs the test suite after each, and only opens a PR for updates that pass clean.
13. Documentation generator. After every merged PR, a loop reads the diff and updates the relevant docs, so they never silently drift out of sync with the code.
14. Landing page variant builder. Give Claude your current page and a list of angles. It generates five complete variants with different copy and layout, ready for your testing tool.
15. Client micro site builder. Find a business with an outdated site. Feed Claude their content and branding. It generates a modern rebuilt version in minutes, ready to send as a finished proof of concept.
Sales and Lead Generation
16. Cold outreach personalizer. Feed Claude a prospect list. It researches each one and drafts a unique opening line, instead of the generic mail merge everyone ignores.
17. Inbound lead qualifier. Connect your form submissions. Claude reads each message, checks it against your ideal customer profile, and tags it hot, warm, or cold before you look.
18. Proposal drafter. After a sales call, feed Claude the transcript. It drafts a proposal on your template, filled with the exact pain points and numbers the prospect mentioned.
19. Follow up sequence writer. Give Claude your stalled deals. It drafts a tailored follow up for each based on how long it stalled and what was last discussed.
20. Prospecting list builder. A scheduled search for businesses matching your criteria, outdated sites, low review counts, missing booking systems, compiled into a daily list.
Research and Knowledge Work
21. Second brain ingestion. Drop any source into a raw folder and say ingest this. Claude reads it, links it to existing notes, and files it into a living wiki that compounds with every source.
22. Literature review assistant. Give Claude a research question and search access. It compiles relevant sources, summarizes each finding, and flags where they agree or contradict.
23. Meeting notes to action items. Connect your meeting transcripts. Claude extracts action items, assigns them to the people mentioned, and drops them into your task manager.
24. Feedback synthesizer. Feed Claude a month of support tickets or reviews. It ranks the top five recurring complaints and the top five praise points by frequency.
25. Due diligence first pass. Claude reads provided documents and flags red flags, inconsistencies, and missing information before you do the deep dive yourself.
Multi Agent Systems
26. Maker and checker pair. One agent writes. A second reviews it against a checklist before it reaches you. This single pattern catches most errors that slip past a solo agent.
27. Debate panel. Run the same question through multiple models and have them critique each other before a final synthesis. Panels consistently beat any single model alone.
28. Research and writer split. One agent gathers and verifies facts. A second, with no search access, writes using only what the first confirmed. Fabricated claims never reach the draft.
29. Triage and escalation. A cheap fast model handles routine classification. Anything it is unsure about escalates to a stronger model. Low cost, high quality where it counts.
30. Self auditing loop. After finishing, the agent reviews its own output against the original instructions before marking the task done, catching errors with no human review.
Customer Support and Community
31. Support first responder. Claude drafts a reply to every ticket using your knowledge base. A human sends, but nobody starts from a blank box, so time per ticket drops hard.
32. Community onboarding flow. New members get an interactive conversation answering their questions and pointing them to the right resources, at any hour, with no one awake.
33. FAQ gap finder. A weekly review of tickets that surfaces questions your FAQ does not answer, so your docs improve on real demand instead of guesswork.
34. Refund and dispute drafter. For cases that meet your policy, Claude drafts the resolution message and internal note, ready to approve in seconds instead of writing from scratch.
35. Sentiment monitor. A daily scan of mentions across channels that flags any spike in negative sentiment before it becomes a visible problem.
Personal Productivity Systems
36. Daily briefing generator. Before you wake, a loop compiles your calendar, important unread email, and relevant news into one briefing document.
37. Email triage assistant. Claude reads your inbox overnight, drafts replies to quick items, and sorts the rest by urgency. You start with decisions, not scrolling.
38. Weekly review compiler. Every Sunday, Claude pulls from your notes, calendar, and completed tasks to draft a review, surfacing what got done versus what was planned.
39. Reading list processor. Everything you saved this week gets summarized into one digest, so you keep the value without revisiting every source.
40. Goal tracker review. A weekly check against your metrics, with Claude flagging trends and asking the one question that matters most from the data, not a generic pep talk.
Infrastructure That Runs Itself
41. CI sweeper. Failed builds get investigated automatically. Low risk fixes open a PR immediately. Ambiguous ones escalate to a human with full context attached.
42. Changelog drafter. Every tagged release gets a draft changelog generated from commit history, formatted and ready for a light edit instead of a blank page.
43. Post merge cleanup. After a branch merges, a scheduled pass removes dead code, unused imports, and stale feature flags nobody got around to clearing.
44. PR babysitter. A loop that watches open pull requests, runs the tests, answers simple review comments, and pings the right person when human judgment is needed.
45. Vulnerability sweep. A scheduled scan of your dependency tree for known vulnerabilities, with Claude drafting the patch PR for anything with a fix and a passing test suite.
E-commerce and Product
46. Product description generator. Feed Claude a spec sheet or supplier data. It writes on brand descriptions for the full catalog, formatted for your platform, no placeholder text.
47. Review response drafter. Claude drafts replies to new product reviews using your brand voice, flagging anything that needs a human before it posts.
48. Inventory reorder alert. A scheduled check comparing stock levels against sales velocity, drafting reorder recommendations before anything runs out.
49. Pricing monitor. Track competitor pricing across their public pages and get a summary the moment something shifts, instead of finding out a quarter late.
50. Abandoned cart sequence. Claude drafts a tailored recovery message for each abandoned cart based on what was in it and how far the buyer got, ready to schedule.
Finance and Admin
51. Invoice drafter. When a project closes, Claude drafts the invoice from the scope you set at the start, itemized and formatted, ready to review and send.
52. Expense categorizer. Feed Claude a month of transactions. It categorizes each one, flags anything unusual, and drafts the summary your bookkeeper actually wants.
53. Contract first read. Claude reads an incoming contract and flags unusual clauses, missing terms, and anything that deviates from your standard, before your lawyer bills you for it.
54. Subscription auditor. A scheduled review of recurring charges that flags subscriptions you are not using, so money stops leaking on tools you forgot about.
55. Cash flow briefing. A weekly summary pulling from your accounting data, showing what came in, what went out, and what is due, drafted before you start the week.
Niche and Specialized
56. Job posting to outreach. Scan job boards for companies hiring for roles your product solves, and draft outreach explaining exactly why they need you right now.
57. Grant and application scanner. A scheduled scan of relevant funding opportunities, pre filtered against your eligibility, with a draft application started for strong matches.
58. Podcast guest research packet. Before any interview, Claude compiles a packet on the guest, their recent work, and three questions nobody else has asked them.
59. Brand consistency checker. A weekly scan of everything you published that flags contradictions, repeated claims, or drift from your positioning, so your public voice stays coherent.
60. Localization pipeline. Feed Claude your best content. It adapts each piece for a different market, not just translated but localized for tone and reference, ready to publish.
The Pattern Behind All Sixty
None of these are about Claude outsmarting you. They are about removing you from work that does not need your judgment in real time.
Every one runs on the same three part structure.
A trigger. A schedule, a new file, an incoming message.
A capable agent. Claude with the right tools and the right context to do the work well.
A verification step. Something checking the output before it reaches you or goes live. A test suite, a second agent, a rule based filter.
Miss the verification and you get expensive mistakes running while you sleep. Miss the trigger and you are back to starting everything by hand. Build all three and you have a system that genuinely works overnight, not a demo that only works while you watch it.
How to Actually Start
Pick one workflow. The one that maps onto something you already do every single day.
Get it running reliably for a week before adding a second. Reliable means it ran without you checking constantly, and the output was good enough to use without heavy editing. If you are still rewriting most of what it produces, it is not done. It just looks done.
The people who build five fragile systems babysit all five. The people who build one solid system build the next nine faster, because they already know exactly what breaks and why.
That compounding is the real unlock. Not the first workflow. The tenth one, built in a fraction of the time the first one took.
Pick one. Build it properly. The other fifty nine get easier once the first one actually runs.
Follow @cyrilXBT for the full builds behind each of these.
