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

By the Income Reality Check Editorial Team · Updated 2026-11-08 · 7 min read

The $25k “Grok bot” business: the solo-agent pitch, fact-checked

Verdict: Half-true — works only if you do the unspoken work. The weekend build is real; the money comes from high-ticket sales and babysitting agents, which the video treats as an afterthought.

On Greg Isenberg’s Startup Ideas podcast, a guest named Nick — co-founder of a tool company called Orgo — lays out a plan to build “a $1M+ solo Grok bot business.” The idea: clone a personal-agent platform (he built his, “BOPS,” in a weekend), niche it to one industry like dental or insurance, then sell it to business owners for $25,000 upfront plus a $3,000-a-month retainer. Is it real? Partly. The part he sells hardest — that one person can build the software fast — is true. The part that actually produces $1M is the part he waves past.

What the video actually claims

Nick says he built his agent platform in 36 hours, in a single long chat thread, using an Opus 5.5 model inside a tool called T3, with design done in Paper and the “computers,” email, and phone numbers for each bot wired in through Orgo, Agent Mail, Agent Phone, and Composio. He posted a demo on X, got 75,000 views, and says existing customers — including an insurance agency — asked him to build versions for them.

The business model is the heart of the pitch. Sell the custom bot to a small business for $25,000 up front (collect half on signing), then a $3,000 monthly retainer to maintain it. Nick frames the price against a human hire: “How much do you pay your front desk lady?” If the answer is $50,000–$70,000 a year, then 15 bots for $25,000 sounds cheap. The alternative he floats is productizing one niche — say, a bot built for insurance agencies — and selling it one-to-many at “10 to 20x” the $200/month that a generic platform charges.

He’s refreshingly honest about one thing: he’s a co-founder of Orgo, the company that sells the virtual computers these agents run on. He says so repeatedly. Keep that in mind — the person telling you to build this sells a pick and shovel you’ll need to buy.

What building it actually takes

The build really is faster than it used to be. Vibe-coding a slick dashboard over a weekend is genuinely possible now, and Nick’s demo — texting a bot, calling it on a real phone number, watching it open four screens — is the kind of thing that was science fiction two years ago. Credit where it’s due.

Running those agents isn’t free, though, and the video never mentions the meter. Every message a bot sends, every web page it reads, every phone call it transcribes burns API tokens. At current rates, the Opus-class model Nick uses runs roughly $4 per million input tokens and $20 per million output tokens, and agent workflows that loop, retry, and control a screen chew through tokens fast — his own overnight “loops” are a tell. Add Orgo’s computers, Composio’s connectors, phone and email infrastructure, and hosting, and the real cost of keeping 15 always-on bots running for one client is a recurring bill you’d better price into that $3,000 retainer before you promise it away.

Then there’s the hard part, the part that is the entire job: selling. Nick’s own advice gives it away. He tells you to call 50 people you know, do the first build free as a case study, ask for three warm referrals, and never sell cold. That’s not a growth hack — that’s a description of a full-time B2B sales operation. High-ticket deals move slowly. Sales cycles for contracts in the tens of thousands routinely run six to twelve months, and you’re selling an unproven, self-built tool to owners who’ve never heard of “computer agents.” To clear $1M at $25,000 a deal, you need 40 signed clients. At a free-case-study-then-referral pace, how long is that, realistically?

Can an AI agent really run the front desk?

This is where the pitch and the reporting part ways. Nick demos a bot that answers the phone and runs errands across apps. The promise to a dental or insurance client is that these agents replace staff — handling billing, bookings, outreach, and support out of the box.

The companies with the most money and the best engineers can’t make that work reliably yet. CNBC reported in October 2026 that startups building AI agents are quietly inserting human contract workers behind the software precisely because the bots fail at phone calls and scheduling — the exact tasks in the demo. One agent run by Marissa Mayer’s Dazzle AI confused window glass with tabletop glass and nearly ordered the wrong product; a human caught it in seconds. Meta’s own agent used human call-center contractors during testing. A whole company, Duckbill, exists to hand an agent’s stuck tasks to real people.

The failure mode that should worry anyone selling this into a dental or insurance office isn’t the loud crash. It’s the quiet one. CNBC calls it “silent failure at scale”: an agent that keeps following instructions while drifting slightly wrong. In one reported case, a system misread new holiday packaging as an error and ordered several hundred thousand excess cans before anyone noticed. Now picture that agent mishandling patient scheduling or an insurance quote. You sold it. You own the cleanup. That’s what the $3,000 retainer actually buys the client — and what it actually costs you.

Who actually wins this game

The person who wins here is not a beginner who watched one podcast. It’s someone who already has three things: the technical fluency to debug an agent that silently breaks, an existing network of 50 business owners warm enough to take the call, and the stomach for months of unpaid case-study work before the first real check clears.

Nick has all three, plus a fourth — he sells the infrastructure. His incentive to make “build your own Grok bot” look easy is not hidden; it’s his business model. The creators who make money teaching a method and the people who make money doing it are rarely the same group, and Nick is candidly in the first group while demonstrating the second.

What you’d realistically earn

The $1M headline is arithmetic, not a track record. Nobody in the video has made $1M doing this; the number is 40 deals multiplied together. Nick’s actual evidence is a weekend build, a viral post, and one insurance client “in the process” of being built. That’s a promising start, not a proven income.

For a realistic beginner, year one looks like this: weeks of building and debugging, one or two free case studies, and a serious chance of $0 in signed revenue for the first several months. Most new businesses don’t survive long enough to compound — U.S. Bureau of Labor Statistics data shows only about half of new establishments make it to year five. AI service businesses face an extra headwind: CNBC notes roughly 80–85% of AI projects have failed since 2018. A disciplined, well-networked operator might land two to four clients in year one and build toward a real six-figure agency over two or three years. Reaching $1M solo, in one year, from a standing start, is not the base case. It’s the lottery ticket.

Who this is (and isn’t) for

This makes sense for you if you’ve already got deep roots in one industry, 10–20 hours a week for sales and firefighting, a few thousand dollars a month for API and infrastructure bills, and the patience to eat free work while you prove it. It does not make sense if you’re expecting passive income, if you can’t personally fix a broken agent, or if you’d be selling automated billing and scheduling into regulated fields — dental, insurance, health — where a silent failure isn’t an embarrassment, it’s a liability with your name on the contract.

What to remember

The software is genuinely easy to build now, and the demand for someone who can wire AI into a small business is real. But “solo” and “$1M” are doing a lot of work in that title. Strip them away and you’re left with a high-ticket sales-and-maintenance agency, in a field where the best-funded agents in the world still need humans standing behind them. U.S. readers should also note the FTC has already sued multiple “AI-powered, done-for-you” sellers that charged up to $35,000 and delivered stores that made “little or no money” — so if you build this, the earnings claims you make to clients are your legal risk, not the podcast’s.

If you want the broader version of this trade-off, see our looks at AI businesses worth starting before 2027 and whether AI can actually run a whole business for you.