Sell Claude workflows for $2,500 retainers

AI agency people are converging on a pretty simple idea: don’t sell “AI.” Sell one workflow a business already wishes someone owned for them. That’s the pitch in Liam Ottley’s new video, and it lands because the tooling has changed. Claude Code is now cheap enough and capable enough that the hard part is less “can the model do this?” and more “can you wrap it into a repeatable service a business will keep paying for?” ### What’s the actual offer? The offer is not a custom software project in disguise. It’s a managed workflow retainer. Ottley’s framing is: go into a small or midsize business, set up an “AI operating system,” and charge ongoing monthly fees starting at $2,500 for automations, monitoring, optimization, and training. In his example stack, retainers can grow from $2,500 to $10,000 as more workflows get layered in. ### Why Claude specifically? (youtube.com) Because Claude Code collapses a lot of build cost. Anthropic pitches Claude Code as a system that can read a codebase, edit files across projects, run tests, and use developer tools directly. That means a nontraditional builder can stand up internal tools and automations much faster than the old “hire engineers or stitch together contractors” route. Basically, the service is still human-led, but the production layer got dramatically cheaper. ### So what are you really selling? Reliability and outcomes. Not the model. A business owner does not care that a workflow uses Claude, n8n, or a spreadsheet with attitude. They care that leads get qualified, proposals get drafted, reports show up in Slack, or CRM junk gets cleaned without staff babysitting it. One recent Claude-focused agency video makes this explicit — the creator says managed agents are how you turn Claude Code into something you can “charge a retainer for” and points to lead intel, results monitoring, and proposal creation as live agency use cases. (anthropic.com) ### Why does the retainer model fit AI so well? Because AI workflows drift. Prompts break. APIs change. Edge cases pile up. Staff need retraining. A one-time build fee sounds clean, but it leaves the client holding a system that needs tuning. The retainer turns that mess into the product. You’re not promising perfect automation forever — you’re promising someone is actively keeping the machine useful. That’s a much easier sale if the workflow touches revenue or labor costs every week. (youtube.com) ### What changed to make this viable now? Two things. First, Claude’s commercial plans now include Claude Code in the main product lineup, with Pro at $20 monthly and Max tiers starting at $100. Second, Anthropic keeps pushing the “agentic” story — Claude as something that can execute, not just chat. When the base tools are this available, agencies can prototype faster and eat less upfront cost before closing a client. ### Where’s the moat if everyone can use Claude? (youtube.com) It’s not in access to Claude. That moat is gone fast. The moat is vertical knowledge, integration pain tolerance, and distribution. If you know how roofing companies route leads, or how a law firm triages intake, you can package a workflow that feels custom while staying mostly standardized. That’s the trick — one painful recurring problem, one repeatable delivery motion, one monthly fee. The model is replaceable. The workflow and client trust are not. (claude.com) ### What’s the catch? A lot of these videos blur the line between a real service business and “agency opportunity” marketing. The revenue examples are plausible, but they are still sales content. The operational burden is real too — support, QA, security, change management, and client communication do not disappear because Claude writes code. If anything, faster building can tempt people into overselling brittle systems. (youtube.com) ### Bottom line The interesting part here is not “Claude can automate things.” Everyone knows that now. The interesting part is that builders are using Claude to back into product-market fit through services first — sell a $2,500 workflow retainer, learn what breaks, then decide whether the repeatable part deserves to become software. (youtube.com)