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

By the Income Reality Check Editorial Team · Updated 2026-10-20 · 7 min read

Automating your company with Claude: what Nick Saraev’s video skips

Verdict: Half-true — works only if you do the unspoken work. The workflow Saraev describes is real and usable, but it quietly assumes you already run a profitable business and have a week plus many hours of hands-on grading to spare.

Nick Saraev’s video “How To Actually Automate Your Company w Claude (Seriously)” has pulled in nearly 84,000 views with a bold setup: he handed Claude access to an agency that does “around a million dollars a year” and asked it to automate everything, and it “actually did a really good job.” By the end he says 80 to 90 of his roughly 100 daily tasks run without him. Is that real? Mostly — but the parts he races past are exactly the parts that decide whether you get the same result or a very expensive to-do list.

What the video actually claims

The pitch is a process, not a product. Saraev walks through four moves. First, audit your work: he had Claude build a tracker that screenshotted his screen every five seconds (about 12 times a minute) and logged what he was doing, plus prompted him with 10 to 15 questions a day about why he did things. Second, feed that list of deliverables into a fresh Claude instance and ask it to design automations for each. Third, build “evals” — repeated tests that score how well the model does each task — and loop until each one scores roughly 8 out of 10. Fourth, turn the winners into reusable “skills,” then schedule the ones safe to run unattended.

He’s candid that this took effort. The audit ran about a week. He upgraded his plan to run a top-tier model in extended-thinking (“Max thinking”) mode. The eval loop was, in his words, “probably the most time-intensive aspect of this whole process” — active hours of him scoring outputs and telling Claude what was wrong. He also fed it templates: high-performing cold email copy, examples of qualified lead lists, strong proposals.

Two links sit in the description. One is Maker Zero, a free prompt. The other is Maker School, his “90-day accountability program” where he says he’ll “guarantee you get a customer in 90 days, or I’ll give you your money back.” That guarantee, and the implied income behind it, is where a reality check matters most.

What the method actually requires

Start with the tools, because the costs are concrete and public. The workflow leans on Claude Code, Agent Skills, and scheduling — all real Anthropic features. Running a frontier model in extended thinking all day is not a free-tier activity. Anthropic’s own pricing page lists Pro at $20/month, Max 5x at $100/month, and Max 20x at $200/month, with usage metered in rolling five-hour sessions and weekly caps. Saraev recommends the most capable model on Max and warns you may “burn out of usage” and have to wait for a session reset. He also points to a Claude Teams plan so you can share skills with non-technical colleagues; that’s $25/seat/month for standard seats (or $100/seat/month for premium seats with Claude Cowork), billed monthly. For a solo operator, budget $100–$200 a month before you automate a single task.

Now the labor he frames as nearly free. The screenshot audit “took me no additional time,” he says — but it still spanned a week of instrumented workdays plus an hour of voice-dictating context. The eval grind is the real bill. Scoring each task 10 times, eyeballing outputs, and iterating prompts is skilled human judgment, not automation. You cannot delegate “is this cold email good enough to send to a paying client?” to the thing you’re training. That’s the whole point of an eval.

Here’s the piece the title glosses over: there’s a business underneath all of this. Saraev is automating the back office of an agency that already has clients, revenue, proven cold-email templates, and documented processes worth codifying. Strip those away and the workflow has nothing to compress. The AI handles production; it does not hand you a company.

And “automate virtually everything” deserves an asterisk. He still runs every discovery call, sales call, and proposal conversation himself — all the customer-facing work. He explicitly keeps a human in the loop for anything that builds a deliverable or writes a campaign end to end. What got automated was the repetitive internal plumbing. That’s genuinely valuable. It’s also a narrower claim than the headline.

So who actually wins here?

The people who get Saraev’s result look a lot like Saraev. They already own a cash-flowing service business with messy-but-real processes, and they have the technical comfort to run Claude Code, write evals, and debug when a scheduled skill misfires. They also have the judgment to grade outputs, because “8 out of 10” is their standard, built from years of doing the work by hand.

That matters because running a one-person consulting operation is itself a demanding job, not a passive one. The U.S. Bureau of Labor Statistics puts the median wage for management analysts — the closest official category to agency consultants — at $101,860 in May 2025, and notes self-employed analysts are paid by the hour or project and often “work more than 40 hours per week.” Automation can shift where those hours go. It doesn’t conjure the clients who pay for them.

What you’d realistically earn

Saraev isn’t promising viewers a dollar figure, which is to his credit. The implied promise is leverage: automate the busywork, keep the revenue, reclaim your time. For someone already running a profitable agency, that’s plausible and probably worth the $100–$200/month in subscriptions many times over.

For someone starting from zero, the math is different, and the ceiling is the business, not the AI. New firms are fragile: BLS data shows only 34.7 percent of private-sector establishments born in 2013 were still operating in 2023, with about half gone inside five years. Automating a business that doesn’t yet exist returns $0. The Maker School guarantee — a client in 90 days or your money back, around $184/month per independent reviews — is where the beginner’s real spend lives, and a guarantee is a marketing promise, not a floor on earnings.

U.S. readers should know regulators watch this exact genre. The FTC sued the coaching operation Lurn in 2023 for marketing programs with claims like “Fail 98% of the Time & Still Be Able to Make $11,453 Per Month,” securing a $14.08 million judgment and a ban on unsubstantiated earnings claims. In January 2025 the FTC moved to strengthen the Business Opportunity Rule and add an Earnings Claim Rule, targeting “money-making opportunities” that fail to disclose that most buyers earn little or lose money. None of that names Saraev — his video avoids a hard income figure, and the inconsistent revenue numbers across his own sites are a reason to treat any program’s headline results as atypical, not a baseline. (Readers in Canada, the U.K., Australia, and the EU have their own watchdogs — the Competition Bureau, ASA, ACCC, and national consumer authorities — applying similar logic.)

Who this is (and isn’t) for

This method fits you if you already run a service business with repeatable processes, can spend a week auditing and several more hours building evals, are comfortable in a terminal, and can afford $100–$200/month in subscriptions while treating the first month as setup, not payoff. It does not fit you if you’re hoping Claude will supply the business idea, the clients, and the expertise. The workflow compresses work you already do well. It can’t originate work you’ve never done.

What to remember

Saraev is right that the models finally cross the line where this kind of back-office automation holds together, and his four-step process is sound. The gap between his demo and your result isn’t the AI — it’s the profitable business, the week of auditing, the hours of grading, and the ongoing human oversight he mentions but soft-pedals. Judge the method on the unspoken work, not the headline.

Related on Income Reality Check

• 10 Claude AI side hustles that can pay a full-time income
• AI can now run your entire content business — here’s how

Views and figures were accurate at the time of review and may have changed since publication.