{"id":"2089708743101399318","url":"https://x.com/cyrilXBT/status/2089708743101399318","text":"https://x.com/i/article/2087177847479607296","author":{"name":"CyrilXBT","username":"cyrilXBT","avatarUrl":"https://pbs.twimg.com/profile_images/2035229727414534145/aWap3Jbq_200x200.jpg"},"createdAt":"Tue Aug 18 13:39:23 +0000 2026","engagement":{"replies":21,"retweets":50,"likes":345,"views":757456},"article":{"title":"80 Claude Opus 4.8 Workflows That Make Money While You Sleep","previewText":"Most people open Claude, ask one question, read the answer, close the tab.\nThat is not what this is about.\nThis is about the other use case. Claude as a system that runs without you, makes decisions","coverImageUrl":"https://pbs.twimg.com/media/HP4ikkeWIAATHiR.jpg","content":"Most people open Claude, ask one question, read the answer, close the tab.\n\nThat is not what this is about.\n\nThis is about the other use case. Claude as a system that runs without you, makes decisions inside boundaries you set once, and produces work or income while you sleep. Opus 4.8 fits this specifically well: efficient enough for high volume, reliable enough to trust with real, repeated work.\n\nHere is why the model choice matters for a list like this specifically. A workflow that runs once a week barely notices which model powers it. A workflow that runs dozens of times a day, every single day, for months, is a completely different cost equation. Opus 4.8 sits at a price and capability point that makes running genuinely high volume, repeated automation actually sustainable, without needing to reach for a premium reasoning model on tasks that never required that depth in the first place.\n\nHere are 80 of them, across 16 categories. Every one 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 trust overnight. None of these are Claude being smarter than you. All of them are Claude removing you from work that never needed you present in real time.\n\n## Content Machines\n\n1. Ingest and publish pipeline. Drop a source into a folder. Claude drafts a thread, a post, and a newsletter blurb from it automatically.\n\n2. Competitor monitor. Daily scan of competitor accounts, summarized, flagged against topics you have not covered.\n\n3. Repurposing engine. One long piece in, ten short variations out, thread, posts, carousel script, video script.\n\n4. SEO cluster builder. One seed keyword produces a full mapped content cluster with ten interlinked article drafts.\n\n5. Newsletter autopilot. Drafted every morning from your RSS feeds overnight, edited in ten minutes.\n\n## Trading and Market Research\n\n6. Wallet pattern scanner. Logs high win rate wallets and their timing from on-chain data.\n\n7. Overnight chart review. Multi-timeframe technical analysis on a queued ticker list, ready by morning.\n\n8. News to thesis pipeline. Watches headlines, drafts an impact thesis before the market fully prices new information.\n\n9. Earnings call summarizer. Extracts guidance changes and contradictions with prior statements immediately after a call drops.\n\n10. Rebalancing alert. Drafts the exact trades needed to hit target allocation, never executes.\n\n## Code That Ships Itself\n\n11. Overnight bug sweep. Fixes lint errors and failing tests, opens one clean PR by morning.\n\n12. Dependency update loop. Weekly, tests each update individually, only opens a PR for what passes clean.\n\n13. Documentation generator. Updates docs automatically after every merged PR.\n\n14. Landing page variant builder. Five complete design variants from one current page and a list of angles.\n\n15. Client micro site builder. Rebuilds an outdated business site as a finished proof of concept.\n\n## Sales and Lead Generation\n\n16. Cold outreach personalizer. Researches each prospect, drafts a unique opener grounded in something specific to them.\n\n17. Inbound lead qualifier. Tags every form submission hot, warm, or cold against your ideal customer profile.\n\n18. Proposal drafter. Builds a proposal from a call transcript, filled with the actual pain points discussed.\n\n19. Follow up sequence writer. Tailored follow up per stalled deal based on how long it stalled and what was last said.\n\n20. Prospecting list builder. Daily scan for businesses matching specific criteria you define once.\n\n## Research and Knowledge Work\n\n21. Second brain ingestion. Drop a source in, say ingest this, it links into your existing wiki automatically.\n\n22. Deep research agent. Spawns sub agents across independent sub questions, synthesizes only after all report back.\n\n23. Meeting notes to action items. Extracts and assigns tasks directly from a transcript.\n\n24. Feedback synthesizer. Ranks the top complaints and praise points from a month of raw tickets.\n\n25. Due diligence first pass. Flags red flags and inconsistencies in provided documents before your own deep dive.\n\n## Multi Agent Systems\n\n26. Maker and checker pair. A Builder produces work, a separate Judge checks it against real ground truth.\n\n27. Debate panel. Multiple models critique each other before a final synthesis.\n\n28. Research and writer split. One agent verifies facts, a second writes using only what was confirmed.\n\n29. Triage and escalation. Cheap model handles routine classification, escalates only what is genuinely uncertain.\n\n30. Self auditing loop. Reviews its own output against original instructions before marking a task complete.\n\n## Customer Support\n\n31. Support first responder. Drafts a reply to every ticket from your knowledge base.\n\n32. Community onboarding flow. Interactive answers for new members at any hour.\n\n33. FAQ gap finder. Weekly scan for questions your FAQ does not currently answer.\n\n34. Refund and dispute drafter. Resolution message ready to approve in seconds for cases meeting policy.\n\n35. Sentiment monitor. Daily scan flagging any spike in negative sentiment early.\n\n## Personal Productivity\n\n36. Daily briefing generator. Calendar, unread email, and relevant news compiled before you wake.\n\n37. Email triage assistant. Drafts quick replies overnight, sorts the rest by urgency.\n\n38. Weekly review compiler. Pulled from notes, calendar, and completed tasks every Sunday.\n\n39. Reading list processor. Weekly digest of everything you saved but have not read yet.\n\n40. Goal tracker review. Weekly check against your real metrics, one sharp question instead of a generic pep talk.\n\n## Infrastructure\n\n41. CI sweeper. Investigates failed builds, opens a fix PR for low risk issues automatically.\n\n42. Changelog drafter. Generated from commit history on every tagged release.\n\n43. Post merge cleanup. Removes dead code and stale flags after a branch merges.\n\n44. PR babysitter. Runs tests, answers simple review comments, pings a human when judgment is needed.\n\n45. Vulnerability sweep. Scans dependencies, drafts a patch PR for anything with a passing fix.\n\n## E-commerce\n\n46. Product description generator. On brand descriptions for a full catalog from spec sheets.\n\n47. Review response drafter. Replies in your brand voice, flags anything needing a human first.\n\n48. Inventory reorder alert. Compares stock against sales velocity, drafts reorder recommendations.\n\n49. Pricing monitor. Tracks competitor pricing, summarizes the moment something shifts.\n\n50. Abandoned cart sequence. Tailored recovery message based on what was in the cart.\n\n## Finance and Admin\n\n51. Invoice drafter. Built from the original project scope the moment a project closes.\n\n52. Expense categorizer. A month of transactions categorized, anomalies flagged.\n\n53. Contract first read. Flags unusual clauses before your lawyer bills you to find them.\n\n54. Subscription auditor. Flags recurring charges you are not actually using.\n\n55. Cash flow briefing. Weekly summary of what came in, went out, and is due.\n\n## Marketing Systems\n\n56. Ad copy generator. Ten distinct angle variations from one offer, ranked by which to test first.\n\n57. Landing page A/B tester. Full alternate page structures generated from one control.\n\n58. Social calendar builder. A month of posts distributed across content pillars automatically.\n\n59. Brand voice guide extractor. Built once from your past writing, reused across every future prompt.\n\n60. Campaign performance digest. Weekly synthesis of what worked and what to kill.\n\n## Recruiting and Hiring\n\n61. Resume screening first pass. Ranks candidates against the actual role requirements, not keyword matching.\n\n62. Interview question generator. Tailored per candidate based on their specific background gaps.\n\n63. Reference check summarizer. Structured synthesis from raw call notes.\n\n64. Offer letter drafter. Built from the agreed terms the moment a decision is made.\n\n65. Onboarding checklist personalizer. Adjusted per new hire's role and prior experience.\n\n## Legal and Compliance\n\n66. Contract redline first pass. Flags deviations from your standard template before a lawyer reviews.\n\n67. Policy update tracker. Weekly scan for regulatory changes relevant to your specific industry.\n\n68. NDA generator. Built from your standard template with deal specific terms filled in.\n\n69. Compliance checklist auditor. Flags gaps against your actual current documented process.\n\n70. Data request responder. Drafts a first response to a subject access request from your data map.\n\n## Analytics and Reporting\n\n71. Weekly metrics digest. Pulled and synthesized from your actual dashboards, not a generic template.\n\n72. Anomaly detector. Flags any metric that moved outside its normal range overnight.\n\n73. Cohort analysis drafter. Structured breakdown from raw user data on a recurring schedule.\n\n74. Board deck first draft. Built from the month's actual numbers and prior deck structure.\n\n75. Churn reason synthesizer. Ranks the actual stated reasons from cancellation surveys.\n\n## Niche and Specialized\n\n76. Job posting to outreach matcher. Finds companies hiring for roles your product solves, drafts why they need you.\n\n77. Grant scanner. Pre filtered opportunities against your eligibility, application draft started for strong matches.\n\n78. Podcast guest research packet. Built before any interview from their recent work and public statements.\n\n79. Price monitoring across competitors. Summary the moment any tracked competitor shifts pricing.\n\n80. Personal brand consistency checker. Weekly scan for contradictions or drift from your stated positioning.\n\n## The Pattern Behind All Eighty\n\nEvery one of these runs on the same three parts. A trigger, a schedule, a new file, an incoming message. A capable agent with the right tools and context. A verification step checking the output before it reaches you or goes live.\n\nMiss the verification and you get expensive mistakes running unattended. Miss the trigger and you are starting everything by hand again. Build all three and you have a system that genuinely works overnight, not a demo that only works while you watch it.\n\n## Why These Categories Specifically, And What Connects Them\n\nLooking across all sixteen categories, a pattern shows up worth naming directly, since it's the actual insight behind why this list works as a whole rather than as eighty disconnected ideas.\n\nEvery single category is built around work that has three properties simultaneously. It's repeated, meaning you do some version of it regularly, not as a one-off. It has a checkable definition of done, meaning you or a verification step can actually tell whether it was completed correctly, not just plausibly. And it doesn't require judgment that only you have access to, meaning the actual decision-making, not just the execution, can be handed off within defined boundaries.\n\nTasks that fail any one of these three tests don't belong on a list like this, and it's worth understanding why, since it tells you how to evaluate whether a task in your own work is a good candidate for automation beyond the eighty specific examples above. A one-off task, no matter how tedious, doesn't justify building a system around it, the setup cost isn't recovered. A task with no checkable definition of done can't be safely automated regardless of how repetitive it is, because you have no way to verify it's actually working correctly without watching every single instance. And a task requiring judgment genuinely unique to you, reading a specific person's tone in a sensitive negotiation, making a call that depends on relationship history no system has access to, shouldn't be automated even if it's repeated and checkable, because the verification standard itself would have to be you, defeating the purpose.\n\n## Routing Workflows To The Right Model, Not Just Opus 4.8 By Default\n\nThis list is framed around Opus 4.8 specifically, and for the large majority of the eighty workflows above, that's the right default. But it's worth being honest about where that default should actually shift, since blindly running everything through one model regardless of task difficulty is its own kind of waste.\n\nFor the categories involving routine extraction and formatting, invoice drafting, expense categorization, product description generation, changelog drafting, Opus 4.8 is not just adequate, it's likely more cost-efficient than reaching for a more expensive reasoning model that offers no real benefit on tasks this well-defined.\n\nFor the categories involving genuinely difficult judgment calls under ambiguity, due diligence first passes on complex documents, contract redlines with real legal nuance, deep research synthesis across contradictory sources, it's worth testing whether a more capable model produces a meaningfully better result on your specific real cases before assuming Opus 4.8's default performance is sufficient. The cost difference on an occasional, high-stakes task is rarely the deciding factor, the quality difference is.\n\nFor the categories involving pure volume with minimal per-task complexity, competitor monitoring, pricing checks, sentiment scanning, an even lighter, cheaper model than Opus 4.8 may perform just as reliably, and running dozens of these checks daily at a lower per-call cost compounds meaningfully over a month in a way a single expensive call never would.\n\nThe actual discipline worth adopting: don't pick one model for your entire automation stack. Match the model to what each specific category of task genuinely requires, and revisit that match periodically as pricing and capability shift, rather than locking in a decision made once and never revisited.\n\n## Troubleshooting The Failure Modes That Actually Show Up\n\nA handful of specific problems account for most of the friction people report when building out several of these workflows at once, worth knowing the fix for in advance.\n\nA workflow that worked in testing starts producing worse output after a few weeks of real use. This is almost always a signal that the input data has drifted from what the workflow was originally built and tested against, a new competitor entered your monitoring list with a different site structure, your product catalog added a category the description generator wasn't tuned for. Periodically re-testing a workflow against fresh, real examples, not just trusting that a system tested once stays correct forever, catches this before it becomes a real problem.\n\nVerification steps start feeling like they're just rubber-stamping everything. If a Judge or checking step has approved one hundred outputs in a row without ever flagging an issue, that's worth investigating directly rather than treating as a sign of a flawless system. Deliberately feed it a known-bad example periodically and confirm it still actually catches the problem, the same stress-testing discipline recommended for any verification system generally.\n\nRunning many of these workflows simultaneously makes it hard to know which one is actually costing the most. Track cost per workflow separately, not just an aggregate monthly bill. A workflow that seemed cheap individually can become the largest cost driver once it's running dozens of times a day, and you won't notice unless you're actually tracking per-workflow cost, not just a total number that obscures which specific system is driving it.\n\n## A Worked Example: Combining Several Workflows Into One System\n\nThe eighty items above are listed individually, but the real leverage shows up when several of them chain together into one coherent system rather than running as isolated, disconnected automations. Here's a realistic example of how that looks for a small e-commerce operation.\n\nWorkflow 6, the prospecting list builder, runs weekly, surfacing new potential wholesale accounts matching specific criteria. Workflow 16, the cold outreach personalizer, picks up that list and drafts a genuinely tailored opener for each one, referencing something specific about their actual business rather than a generic template. Workflow 17, the inbound lead qualifier, handles the replies that come back, tagging each one hot, warm, or cold before a human looks at any of it.\n\nSeparately, workflow 46, the product description generator, keeps the catalog current as new products get added, while workflow 48, the inventory reorder alert, and workflow 49, the pricing monitor, run in the background checking stock levels and competitor pricing daily without anyone remembering to check manually.\n\nNone of these five workflows individually looks like a dramatic transformation. Together, they represent the entire top-of-funnel sales motion and a meaningful share of ongoing catalog maintenance running with the business owner touching only the qualified, warm leads and the occasional pricing decision that genuinely needs human judgment. That's the actual shape of what \"eighty workflows\" is pointing toward, not eighty separate, disconnected automations, but a smaller number of connected systems built from combining the individual building blocks above into something that matches your actual business.\n\nThe practical lesson from this example: don't evaluate any single workflow in isolation from what it could connect to. When you're deciding which of the eighty to build first, also ask which two or three naturally chain together into a system bigger than the sum of its parts, since that combination is usually where the real time savings compound.\n\n## Deciding Which Of The Eighty To Build First\n\nWith eighty options, the actual bottleneck for most people isn't lacking ideas, it's picking one and actually finishing it rather than starting several and finishing none. Here's a concrete way to decide.\n\nRank your candidates by a simple two-factor comparison: how often you currently do this task manually, and how painful or error-prone doing it manually actually is. A task you do daily that's also genuinely tedious, email triage, competitor monitoring, lead qualification, sits at the top of that ranking regardless of which category it falls under. A task you do monthly and don't particularly mind doing by hand sits at the bottom, even if automating it sounds impressive.\n\nWithin your top few candidates, apply one more filter: which one has the clearest, most checkable definition of done. A workflow where you can write down exactly what correct output looks like, in a sentence or two, is faster to build and safer to trust than one where \"good output\" is a vaguer, more subjective judgment call. Start with the clearest one, not necessarily the most exciting one.\n\nResist the urge to pick a workflow because it sounds impressive to talk about rather than because it actually addresses your real daily friction. The workflows that end up mattering most in practice are rarely the most sophisticated-sounding ones on this list. They're the boring, repeated, checkable tasks that quietly ate an hour of every single day without anyone noticing how much that hour actually cost over a month.\n\n## What This List Doesn't Replace\n\nWorth being direct about the limits here, since a list this long can start to feel like it implies full automation of an entire business, and that overstates what's actually true.\n\nEvery workflow above handles execution of well-defined, repeated work. None of them replace the strategic decisions about what your business should actually be doing, which market to pursue, which product to build next, which relationship is worth investing real personal time in. Those decisions still require you, specifically, not because a model couldn't generate a plausible-sounding answer, but because the actual accountability and judgment behind a strategic call isn't something that should live inside an unattended system regardless of how good the underlying model is.\n\nThe honest way to think about the relationship between these eighty workflows and the strategic side of running something: automation clears the operational noise that competes for your attention, so the time and mental space you get back is available for the decisions that actually need a human making them. It's not a replacement for that judgment. It's what makes room for it.\n\n## What This Actually Buys You Over Time\n\nThe value of building even a handful of these compounds in a way that's easy to underestimate looking at any single workflow in isolation. A single automated task saving twenty minutes a day sounds modest. Ten of them running simultaneously, each saving a similar amount, adds up to hours reclaimed every single week, hours that used to disappear into repeated, checkable work that never actually needed your personal attention in the first place.\n\nThe real compounding effect shows up specifically in month two or three, once several workflows are running reliably at the same time and you notice the shape of your actual workday has changed, less time spent on the mechanical, repeated parts of running things, more time available for the parts that genuinely benefit from your specific judgment and attention. That shift, not any single workflow's individual output, is the actual thing this entire list is pointing toward.\n\n## How To Actually Start\n\nPick 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.\n\nThe 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.\n\nThat compounding is the real unlock. Not the first workflow. The tenth, built in a fraction of the time the first one took.\n\nPick one. Build it properly. The other seventy nine get easier once the first one actually runs, and the connections between them, the way workflows chain into real systems rather than staying isolated, are where the actual time savings compound past what any individual item on this list can deliver alone.\n\nFollow @cyrilXBT for the full builds behind each of these."}}