AI Side Hustles Half-true — works only if you do the unspoken work

By the Income Reality Check Editorial Team · Updated 2026-05-16 · 7 min read

Nate Herk’s 5 levels of Claude: the income math the tutorial skips

Verdict: Half-true — works only if you do the unspoken work. The tutorial is good. The implied money-making ladder is much steeper than a 21-minute video suggests.

Nate Herk’s “Every Level of Claude Explained in 21 Minutes” walks viewers through five tiers of using Anthropic’s AI, from the casual chatbot user to the “architect” running autonomous Claude agents on cloud infrastructure. Around the 12-minute mark, the framing shifts: level 3 is where “non-coders can sell automation as a service,” level 4 unlocks “$5,000 to $15,000 projects” for freelancers and agencies, and the closing pitch directs viewers to a paid community offering “a path to making your first dollar with AI.” The tutorial portion is genuinely useful. The income ladder it implies is not so much wrong as it is missing roughly half the staircase.

What the video actually claims

Herk frames Claude in five levels. Level 1 is anyone who treats Claude as a search bar. Level 2 is a paid Pro user with Projects, connectors, memory, file creation, and the Microsoft Office add-ons — useful enough to “save 5+ hours a week.” Level 3 introduces Claude Cowork, a desktop app that takes actions on your computer: organizing your downloads folder, building skills, scheduling tasks, and using Claude Design to prototype apps without writing code. Level 4 is Claude Code, the engineering-focused CLI with subagents, parallel work trees, plan mode, and MCP tool access. Level 5 is “architect” — Cloud Routines, hooks, channels, agent SDKs, and headless mode, where Claude runs while your laptop is closed.

The monetary claims sit on top of this. The video says level 3 is “your minimum bar” if you’re running an AI automation business, level 4 is where “freelance and agency work becomes $5,000 to $15,000 projects,” and level 5 is where Claude becomes “infrastructure.” Herk does not promise a specific dollar figure to the viewer. What he does do — repeatedly — is link levels of tool fluency to tiers of client revenue, then close with a pitch to his paid community for those who want to “make their first dollar with AI.”

That community is real and worth naming. Herk runs AI Automation Society (a free Skool group of roughly 305,000 members) and a paid tier, AI Automation Society Plus, priced at about $99/month or $839/year for premium access to two flagship courses. The video itself is, in part, a funnel into that paid product.

What the method actually requires

Start with the unavoidable subscription bill. According to Anthropic’s own pricing page, Claude Pro costs $20/month (or $17/month billed annually), Claude Max starts at $100/month for 5× the Pro usage and runs to $200/month for the 20× tier, and Team plans run $20–$125 per seat per month. Cowork, Claude Code, and the rate-limit headroom you actually need to run “five parallel sessions a day” — Herk’s example from Anthropic engineer Boris Churnney — sit on the paid tiers. The free plan gets you nothing past level 1 in any practical sense.

If you build production agents (level 5), you also pay per token. Claude Opus 4.7 lists at roughly $5 per million input tokens and $25 per million output tokens; Sonnet 4.6 runs $3 and $15. Prompt caching can knock 60–90% off cached input on long sessions, but autonomous agents running on schedules, with subagents, MCP tools, and verification loops, are not cheap. A handful of busy routines per day on Opus can easily clear $100–$300/month in API spend before you charge a single client.

Then there’s the part the tutorial doesn’t price at all: the hours. Herk opens the video by saying he spent “over 400 hours inside of Claude” to master it. That isn’t a throwaway line — it’s the actual entry cost for everything he describes from level 3 onward. The cowork-to-architect arc he sketches out covers skills authoring, MCP servers, hooks, work trees, agent SDKs, GitHub-event triggers, and trust-building reps before you can responsibly let a cloud routine run unattended. None of that is “no-code” in the casual sense, even if you never open a terminal.

Once you have all that built, the actual income choke point is finding clients. Industry data on AI freelancing on Upwork puts entry-level AI engineer rates at roughly $30–$50/hour, with mid-level (2–5 years of experience) at $80–$120/hour. Time-to-first-client estimates from agency-acquisition blogs cluster at 2–3 weeks via marketplace bidding and 2–4 months via cold outreach. The “$5,000 to $15,000 project” tier exists, but it’s the upper end of agency pricing for delivered systems, not the typical first invoice. And competition is intense: Upwork itself reports AI-related freelance demand more than doubling year over year, which is good for total volume and brutal for new entrants pitching the same automations.

The U.S. Federal Trade Commission has been actively warning about a tier below this, where the pitch isn’t a tutorial but a packaged “AI business opportunity.” The FTC’s Operation AI Comply sweep, launched in September 2024 and continued through 2025 and 2026, has hit operators including Ascend Ecom (alleged $25M in consumer losses), FBA Machine (alleged $15M), Ecommerce Empire Builders, and Air AI, which was banned from marketing business opportunities in a March 2026 settlement. The agency’s standing guidance is blunt: in the 2023 piece headlined “AI” stands for “allegedly inaccurate”, the FTC told business-opportunity sellers that AI labels do not exempt anyone from substantiating earnings claims. None of that is Herk — he’s selling a community and courses, not a “done-for-you AI storefront.” But it’s the regulatory weather the income-claim layer of his video lives inside, at least for U.S. readers. U.K. readers should note the ASA’s similar stance on substantiating earnings claims in marketing; Indian readers should know there is no equivalent enforcement specific to AI yet.

Who actually wins this game

The people who clear $5k–$15k client projects with Claude tend to share a profile. They were already engineers, technical PMs, or business analysts before AI got hot — Herk himself was a Business Intelligence analyst at Goldman Sachs before going full-time on AI education. They have an existing audience (Herk has roughly 650,000 YouTube subscribers, which is the actual unfair advantage in his story). Or they have prior consulting relationships and just folded automation into the menu. The viewer with no audience, no technical background, and no client list, watching this in 2026 to start “making their first dollar,” is starting from a meaningfully different place than the people whose income screenshots they’re hearing about.

That is also where the paid community arithmetic gets interesting. At $99/month, AI Automation Society Plus costs around $1,200/year. Plenty of joiners use it and get value; reviews cluster around 4.3/5. But the same content that helps a working freelancer charge $200/hour is roughly inert for a complete beginner who hasn’t yet built a single end-to-end automation for a real organization.

What you’d realistically earn

A grounded range looks like this. Months 1–3, while you’re learning Claude Cowork and Claude Code and building portfolio pieces, you should expect $0. Months 3–6, with a Pro or Max subscription, a couple of finished demos, and consistent outreach, an Upwork beginner with no audience might land one or two small jobs at $25–$50/hour for total monthly income in the $200–$1,500 range. After 12 months, freelancers with delivered systems and a referral pipeline can credibly clear $3,000–$8,000/month in retainer-style work. The $5k–$15k project tier the video name-checks is real but tends to be reserved for senior operators selling full automations to mid-market companies — not someone three months past their first Claude Pro upgrade.

That’s a meaningful income for someone with the time and grit to stick with it. It’s also a long way from “save 5 hours a week and you’re paying for Claude.”

Who this is (and isn’t) for

This video is genuinely useful if you already do some kind of knowledge work — analyst, developer, marketer, consultant, ops manager — and want to compress a few months of trial-and-error into 21 minutes of vocabulary. It’s a sound starting map for someone who has 5–10 hours a week, a Pro subscription budget, and a real workflow to apply Claude to. It is not a great starting point for someone with no existing technical foothold who is looking at AI as a route out of a job they dislike. That person will need to plan for six to twelve months of unpaid learning before the income side starts working, and should be skeptical of any community pitch that compresses that timeline into “your first dollar.”

For more on similar Claude-monetization pitches, see our previous breakdowns of a $20,000/month Claude Code YouTube claim and the “Claude AI side hustle nobody is talking about” framing — they share the same shape.

What to remember

The tutorial is fair. The product Herk is selling — community, templates, courses — is real and rated reasonably well. The half-truth is in the link between tool fluency and client revenue. Knowing what Cloud Routines do is necessary but nowhere near sufficient for landing $10,000 contracts. Plan for subscription costs, API bills, hundreds of hours of skill-building, and a quarter or two of low-pay client work before the numbers in this video start applying to you.

View counts and pricing details may have changed since publication.