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

By the Income Reality Check Editorial Team · Updated 2026-06-29 · 7 min read

Claude Code didn’t make money in this video — it made a to-do list

Verdict: Half-true — works only if you do the unspoken work. The four upgrades genuinely make Claude Code produce better work faster, but the video proves productivity, not profit.

In “I asked Claude Code to make me as much money as possible,” Nate Herk of the channel Nate Herk | AI Automation opens with a big number: he says four “upgrades” to Claude Code made him “three times more money in the past 30 days.” The video has roughly 98,000 views. The upgrades are real and worth knowing. The income claim, though, never gets tested — because by the end of the video, nothing has actually been sold.

What the video actually claims

The pitch is that Claude is “tuned to make you feel productive,” not to make you money, and four fixes close that gap. First, a custom skill called roast that forces Claude to argue against your idea using a “council” of personas instead of agreeing with you. Second, a verification habit where Claude checks its own work with browser automation (Playwright) before declaring a task done. Third, context management — commands like /context and /clear plus a homemade session handoff skill to keep the model sharp. Fourth, parallel sub-agents and a /goal command that loops until a separate evaluator model agrees the work is finished.

To show the upgrades working, Herk builds a fictional product live: a $9/month tool that turns a YouTube transcript into a week of LinkedIn posts. He runs roast on it, has Claude build and stress-test a landing page, manages the context window, then fires off a /goal run that spins up six sub-agents and produces a full go-to-market kit in about eight minutes — positioning, market research, a 14-day launch plan, outreach templates, 25 post drafts, a content calendar.

Here’s the part worth sitting with. The product in the demo has zero customers and zero revenue. The deliverable isn’t income; it’s a folder of planning documents. The video’s actual call to action is to join Herk’s free Skool community (“over 400,000 people”) for the prompts, with a paid “plus” community upsell. That’s the business model on screen — not the LinkedIn tool.

What “3x more money” leaves out

Start with what’s true, because a lot of it is. Each upgrade is anchored to real research, and Herk cites it accurately.

Sycophancy is documented. The ELEPHANT study he references found that on open-ended advice, leading models avoid challenging a user’s framing about 88% of the time versus roughly 60% for humans — so forcing Claude into devil’s-advocate mode is a legitimately good instinct (arXiv). The verification point holds too: an NYU study generated 1,689 programs with GitHub Copilot and found about 40% contained security weaknesses, which is a fair reason to make AI test its own output (arXiv/NYU). Context rot is real — Chroma tested 18 frontier models, including Claude, and every one degraded as input grew (Chroma). And the sub-agent claim checks out: Anthropic’s engineering team reported a multi-agent setup beat a single agent by 90.2% on an internal research eval (Anthropic).

So the workflows do what Herk says they do. They make Claude Code produce sharper output, faster, with fewer silent failures. That’s a real gain.

But faster, sharper output is not money.

And this is where the video and reality split. Making money requires the one thing a launch kit can’t do for you: get strangers to pay. The demo skips the entire distribution problem and presents a finished planning folder as if it were a finished business.

You don’t have to take my word for it — the video’s own tool said so. When Herk ran roast on the $9 tool, the council scored it 2-3 out of 10 and returned a “reshape/kill” verdict. Its reasoning, quoted in the video: “no moat and a free substitute,” “no distribution with no audience,” and customer acquisition cost (CAC) that “will exceed a $9 LTV on day one.” That’s not a nitpick. That’s the whole game, and the AI nailed it. Independent B2B SaaS benchmarks put outbound-sales CAC around $400 per customer and paid-channel CAC for SMB software in the $200–$500 range. Against a $9–$19 monthly subscription, you’d need a year or more per customer just to break even — before churn, before refunds, before your own time.

Then there’s the cost the “free community” framing glosses over. Running this stack — Opus 4.8, a million-token context window, six parallel sub-agents, /goal loops that run eight-plus minutes — is not free. On the API, Opus 4.8 runs $5 per million input tokens and $25 per million output (Anthropic pricing), and a multi-agent run burns tokens fast. On a subscription, the kind of usage shown here lives on Claude’s Max plan, which starts at $100/month and climbs to around $200 — not the $20 Pro tier. The tools cost money to run before they make any.

So who actually makes money here?

Look at who’s already winning in the frame. Herk is. He has a 400,000-member community and a YouTube channel with an audience that trusts him. For someone in that position, “I built a launch kit in eight minutes” is genuinely valuable, because he already solved distribution years ago — he just points his audience at the next thing. The upgrades multiply the output of someone who already has a market.

The viewer starting from zero is in the position roast described: no audience, no moat, a few hundred dollars of runway. For that person, the eight-minute go-to-market kit is the easy 10% of the work. The hard 90% — finding 20 to 30 people in a narrow niche who’ll actually pay, and convincing them — is exactly what the video fast-forwards past. As side-hustle researcher Nick Loper put it to CNBC, “AI alone is not a business… You still have to figure out your product or service and find people to sell it to” (CNBC).

What you’d realistically earn

The honest range for a brand-new micro-SaaS is sobering, and it’s nowhere near “3x in 30 days.” Surveys of indie-hacker products consistently find that more than half make no revenue at all, and roughly 70% earn under $1,000 in monthly recurring revenue. Among those that do earn, the median sits around $500/month, and crossing $1,000 MRR typically takes 12 to 18 months of steady work — for founders who execute well.

Now layer on the product’s specific weaknesses. A $9 transcript-to-LinkedIn tool is a commodity with free substitutes, which means high churn; a 5%-plus monthly churn rate means you’re refilling a leaking bucket every single month. NerdWallet’s blunt summary of “passive” income applies here: “Most ways to generate passive income require an upfront investment of either money, time or both — the income comes later (in some cases, much later)” (NerdWallet). For year one, a realistic expectation is $0 to a few hundred dollars a month while you grind on marketing — not a tripling of your income in a month.

One more thing U.S. readers should know. The FTC has proposed rules specifically to police “money-making opportunity” sellers, and it requires anyone making earnings claims to be able to substantiate them (FTC). A “3x more money in 30 days” claim with no figures attached is exactly the kind of statement that rule targets. To be clear, this video isn’t selling a paid money-making course — the community is free — so it’s a softer case. But the standard is a useful filter for the viewer: if a number isn’t backed by numbers, treat it as marketing.

Who this is (and isn’t) for

These four upgrades make sense for people who already build with Claude Code and want better results: developers, automation freelancers, agency operators, anyone who already has clients or an audience and needs to ship more, faster. If that’s you, adopt the verification loop and the sub-agent pattern — they’ll pay for themselves in saved rework. This is not for the viewer hoping the /goal command will conjure a customer base out of nothing, who has no distribution, no budget for paid acquisition, and limited time. For that person, the bottleneck was never Claude’s output quality. It was always the market.

What to remember

The methodology is real and the research behind it is solid. What the video quietly equates is “I produced a polished plan” with “I made money,” and those are different things — a gap the creator’s own roast tool spelled out before he built the product anyway. Treat the four upgrades as what they are: a way to work better inside Claude Code. Just don’t mistake an eight-minute launch kit for a launch.

For more on what it actually takes to earn with these tools, see our look at 10 Claude AI side hustles that can pay a full-time income and the closely related I asked Claude to make me as much money as possible.