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

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

Zapier MCP and Claude: what “I let AI run my entire business” leaves out

Verdict: Half-true — works only if you do the unspoken work. The plumbing Siraj Raval demos is real and genuinely useful, but “runs my entire business” quietly skips the building, watching, and paying that keeps it running.

In “I Let AI Run My Entire Business (Zapier MCP),” Siraj Raval says an AI is handling his operations while he never touches his laptop — pulling leads, drafting emails, charging Stripe invoices, posting tweets. He claims a fleet of 85 agents (he calls the system Aurelian) has run sponsor deals, content scheduling, and financial reconciliation for six months. Is that real? Mostly, yes — the demos work as shown. But the phrase “run my entire business” is doing a lot of heavy lifting, and the parts it hides are the parts that cost you money and time.

What the video actually claims

The pitch is built around Zapier MCP — Model Context Protocol — which Raval frames as “the API moment for agentic AI.” His argument is that Claude was always capable, but stuck inside the chat window. MCP is the pipe that lets it call Zapier’s roughly 9,000 app integrations directly. Connect the server once, authorize what you want, and Claude can act on Notion, Gmail, Stripe, Slack, and Twitter as if they were all “the same shape.”

He runs four live jobs. A new lead hits his contact form; Claude enriches it, scores it one to five, and adds a Notion row in about 12 seconds. It drafts a follow-up email that references the lead’s company and queues it in Gmail drafts for approval — no auto-send. When he flips a Notion row to “closed won,” Claude drafts a Stripe invoice, emails it, and pings Slack. When he publishes a stub video, a tweet announcing it appears within 20 seconds.

Raval is careful, to his credit. He says write actions that touch money or reputation should be “drafted and approved,” not fully autonomous, and he names his own failure modes: latency on chained calls, hallucinated tool arguments, state drift between calls. That’s honest. It’s also the tell.

What the method actually requires

Here’s the reframe: this isn’t a business running itself. It’s a business owner who spent months wiring, authorizing, and babysitting a system that now handles defined, repetitive handoffs. Those are different things.

Start with the meter. Zapier bills by “tasks,” and one MCP tool call consumes two tasks. Raval’s own lead demo has the agent doing “six things across three apps” — call it six tool calls, roughly 12 tasks, for a single lead. Zapier’s Free plan gives you 100 tasks a month (about 50 tool calls, or a handful of leads). The Professional plan runs $29.99/month billed monthly and includes 750 tasks — around 375 tool calls, so maybe 60 leads before you hit the wall. Cross it and Zapier switches you to pay-per-task at 1.25× the base rate. A “fleet of 85 agents” firing all day is not living on the free tier.

Then the model. Claude Opus 4.8 costs $5 per million input tokens and $25 per million output tokens (Anthropic’s pricing). Agentic loops are token-hungry — every tool call round-trips the full context, and chained steps multiply it. This is manageable, but it’s a real monthly bill that scales with activity, not a rounding error.

And the money it moves isn’t free to move. Every Stripe invoice the agent sends costs 2.9% + 30¢ on U.S. cards (Stripe pricing) — the same as if you’d clicked the button yourself. The agent saves the clicking, not the fee.

The bigger unspoken cost is supervision. CNBC has reported on what it calls “silent failure at scale” — agents that don’t crash loudly but quietly take wrong actions that compound. Raval’s own “human in the loop on consequential actions” rule exists precisely because of this. That rule has a price: someone reviews the drafts. That someone is you.

Who actually wins this game

Look closely at who’s demoing this, because it’s the pattern behind almost every “AI runs my business” video. Raval is a well-known AI educator with an existing audience, a working CRM full of real leads, live sponsor deals, and years of Zapier fluency. The agents automate a business that already exists and already generates inbound. MCP made his last mile cheaper. It did not create the leads, the offer, or the reputation that makes a follow-up email worth sending.

The people who get real leverage here are operators who already have a repeatable process and enough volume to justify the wiring — freelancers drowning in intake, small agencies with predictable onboarding steps, solo founders whose bottleneck is genuinely admin. If your “business” is a single sale a week, automating the plumbing around it saves you minutes and adds a monthly bill.

What you’d realistically earn

The video never promises a dollar figure, and that’s worth respecting — it’s a capability demo, not an income claim. But the implied payoff is time: Raval says a lead went from “20 minutes” of his work to “15 seconds” of review.

Take that at face value and do the arithmetic. If you handle 40 leads a month, saving ~19 minutes each is about 12 hours reclaimed. Against maybe $30–$80/month in Zapier and model costs plus several weekends of setup, that’s a genuinely good trade for the right operator. For someone without steady lead flow, the same setup returns a few saved minutes and a subscription. There’s no version of this where switching on 85 agents generates revenue that wasn’t already coming — it redistributes your hours, it doesn’t manufacture customers.

Who this is (and isn’t) for

This makes sense if you already run something with real, repetitive operations, you’re comfortable authorizing an AI to touch your Stripe and email, and you can spend a few weekends building and then a few minutes a day reviewing. It’s a poor fit if you’re looking for a business — not a way to speed one up — or if the idea of an unattended agent issuing invoices makes you nervous (a reasonable instinct). The technology assumes you already have the thing it’s automating.

One caution for U.S. readers on the marketing side: the FTC’s Operation AI Comply has already fined operators for exaggerated “AI will make you money” claims, and CNBC has documented companies quietly rehiring after over-trusting automation. Raval isn’t making an earnings claim — but if you build on top of this and sell it as one, that’s the regulator you’ll answer to.

What to remember

The tech in this video is real, the demos are legitimate, and the human-in-the-loop design is the right way to build. What the title oversells is autonomy. An AI isn’t running Raval’s business; a system he built and monitors is handling the repetitive edges of a business he already had. Copy the architecture, not the fantasy of the empty chair.

For more on the “AI does the work for you” framing, see our reviews of building AI employees on a schedule and a $10M business with zero employees.