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

By the Income Reality Check Editorial Team · Updated 2026-11-07 · 7 min read

Nick Saraev’s 31 Claude demos are real — but the skill is the part he skips

Verdict: Half-true — works only if you do the unspoken work. The demos genuinely happened; almost every one was pulled off by a working engineer, and not one of the 31 actually earns a dollar.

Nick Saraev’s “31 Life-Changing Ways to Use Claude in 2027” has pulled 890,728 views by doing something most income videos don’t: it barely mentions income. Instead it rattles through 31 jaw-dropping things people have done with Claude — stripping ads out of a smart TV, jailbreaking a Kindle, finding a planet candidate in NASA data, building a research team of 30 AI agents to attack a cancer diagnosis. His pitch is that we’re in a “capabilities overhang,” and that you can reproduce these outcomes “right now” for about “$5 in tokens.” The demos are real. The catch is who’s holding the keyboard.

What the video actually claims

The framing is that Claude — specifically Claude Code, Anthropic’s command-line agent — can now do far more than people realize, and that the barrier is imagination, not skill. Saraev narrates each example as something a person “just asked Claude” to do: one person “just said to Claude, ‘hey, can we build this?’” and got a dictation microphone with custom firmware and a 3D-printed case in a couple of days. Another “just plugged a Kindle into their laptop” and had the Amazon software replaced. A developer wrote “something like 1,500 scripts to search 7 years” of NASA telescope data and surfaced a new exoplanet candidate.

Several of these check out. Anthropic really did run Project Deal, a one-week experiment in December 2025 where Claude agents bought and sold real items for 69 employees and closed 186 transactions — and the agents running the stronger Opus model did negotiate better prices than the ones on Haiku. The exoplanet story traces to developer Pavel Rabtsevich, whose candidate in TESS data is now queued for NASA follow-up. Pieter Levels really is a well-known indie developer. The anecdotes aren’t fabricated.

Here’s the part that’s easy to miss. The money never appears in the 31 examples. It appears at the end, when Saraev points you to “Maker School, first link in the description” — his paid community — “specifically ways to use Claude to make money.” That’s the product. The 31 demos are the trailer.

What the method actually requires

Watch what the successful people in the video have in common. Miriam, who built the 30-agent cancer-research team, is a software engineer. Rabtsevich wrote 1,500 scripts — that’s not a prompt, that’s a project. “Cody,” the solo founder whose biotech Anthropic reportedly acquired, ran a company. Pieter Levels ships software for a living. Every headline result in this video belongs to someone who could already code, debug, and reason about hardware before Claude entered the picture.

Claude Code is a developer tool. It lives in a terminal, it edits files, it runs commands, and it assumes you can read an error message and know whether its output is sane. “Jailbreak this Kindle” sounds like a sentence you say out loud; in practice it’s a loop of flashing firmware, bricking the device, reading logs, and trying again. The video shows you the finished screensaver, not the three evenings of failed boots. That compression — collapsing “Claude did it” into “you can do it” — is the single biggest thing the video leaves out.

The tool is extraordinary; the operator still matters.

Then there’s the cost claim. “$5 in tokens” and a “$20 Claude subscription” describe a light afternoon, not the workloads on screen. As of October 2026, Claude Opus 5.5 runs $4 per million input tokens and $20 per million output; Haiku 5.5 is cheaper at $0.10/$0.50. A thousand-agent sweep of a DNA database (another example in the video) is not a $5 errand — Anthropic can run that because, as its own IPO filing shows, it has effectively unlimited compute and booked $16.29 billion in revenue in the first half of 2026. You and I are paying retail.

So who’s actually pulling these off?

The winners here are technical people who finally got a force multiplier. If you can already design a circuit, Claude writing the firmware is a genuine accelerant — maybe the difference between a weekend and a month. That’s the real, defensible claim buried in the hype: for a competent builder, Claude Code collapses the time between idea and working prototype. Saraev isn’t wrong that this is remarkable.

But notice the survivorship. We hear about the Kindle that became a home dashboard, not the dozen that got bricked. We hear about the planet that got found, not the thousands of hours of null results in archival data. A reality check isn’t about whether the best outcomes happened — they did — it’s about whether they’re what a typical viewer will get. And a typical viewer is not a software engineer with a 3D printer and a spare DJI drone.

What you’d realistically earn

Nothing, from the 31 examples themselves — and that’s not a knock, because none of them is a business. Removing TV ads saves you from an annoyance. Rebuilding a baby-tracking app saves a subscription fee. Fixing your Wi-Fi saves a service call. These are money-saving and hobby projects dressed up as “life-changing,” and the value is real but personal.

The earning starts only if you turn the underlying skill into client work, which is exactly what Maker School sells: $184 per month, 218 videos, and a stated guarantee of your “first paying AI client in 90 days or your money back.” That’s a legitimate business model — AI automation consulting is a real market. Just be clear about the sequence. The video’s magic trick is free; the path to income is a subscription; and the income itself depends on you learning to build and sell systems, which takes months, not a weekend. If you’re comparing paths, our breakdown of Claude AI side hustles that can pay a full-time income lays out what that ramp actually looks like.

One guardrail worth knowing, especially for U.S. readers. The FTC has proposed expanding its Business Opportunity Rule to cover “money-making opportunities,” including business coaching, and it wants sellers to be “ready to provide information about what [earnings] claims are based on – in writing – upon request” (FTC, 2025). A “first client in 90 days” guarantee is a refund promise, not an earnings claim — but any program that tells you what you’ll make is now squarely in the regulator’s sights.

Who this is (and isn’t) for

This path fits you if you already write code or genuinely enjoy debugging hardware, you have 10-15 hours a week to spend learning Claude Code, and you treat the free examples as skill-building rather than a shortcut. Under those conditions, the video is an honest tour of what’s possible, and a paid community can shave months off the learning curve.

It isn’t for you if you took “no skills needed, $5 in tokens” literally and expected passive income. Could you follow along and remove the ads from your own TV? Maybe, with patience. Could you turn that into a thousand-dollar month by next quarter without learning to build and sell anything? Almost certainly not — that part is work, and the video’s charm is that it makes the work invisible. (If “no experience” is the hook that drew you in, our look at proven ways to make money with AI with no experience is a sober companion to this one.)

What to remember

The capabilities are real, the people in the video are real, and the stories mostly happened as told. What’s missing is the prerequisite: nearly every result came from someone who could already engineer, and the income Saraev gestures at lives in a $184-a-month course, not in the token bill. Treat the 31 demos as a map of what’s now possible for builders — and treat the “anyone can do this for $5” framing as the trailer it is. The broader AI gold rush is loud right now (Anthropic is chasing a valuation CNBC’s sources call a “stretch” at $1.5 trillion); loud markets reward people who separate the demo from the day job.