Source: raw/The_1M+_Solo_Grok_Bot_Business_Full_Course.md

Creator: Greg Isenberg, The Startup Ideas Podcast, with Nick, co-founder of Orgo (cloud computers for agents) · URL: https://www.youtube.com/watch?v=67BmkptgTms · Duration: under an hour (stated) · Platform: YouTube (auto-captions, fetched 2026-11-05).

Interest to note: Nick co-founds Orgo, which sells the computers this build uses, and recommends it throughout. Both speakers say they have no ties to Paper, T3 Code or CleanShot. The pricing is a proposed offer; no client revenue from it is shown.

The idea: personal-agent platforms such as Grok Bot are general and confusing to a business owner who “just wants their problems solved”. Build your own version for one industry, with the bots already set up for that industry’s jobs, and sell it either to a few businesses at a high price or as a product to many. Nick built his version, BOPS (“business ops”), in about 36 hours and open-sourced it. Each bot has its own email address, phone number and four-screen cloud computer. The episode covers his build method, the stack, the offer and how to find a first niche without guessing.

Key Takeaways

  • The product. A team of bots for one kind of business (his examples: insurance agencies, marketing agencies, dental clinics), each with its own phone number, email, computer and channel connections such as Slack. Orgo is building a custom-branded insurance-agency version for a customer. In the demo he texts and then phones his chief-of-staff bot, “Boppy”, and asks it to list prospect companies with their websites, pull up the S&P 500 and watch Hacker News for relevant posts.
  • One agent per platform. His tip: give each platform its own logged-in agent (X, email, Skool, LinkedIn) that owns that platform’s routines. His LinkedIn agent finds prospects, sends a connection request, messages after they accept, books a call, and has a brief ready when he joins.
  • How he built it (his principles):
    1. Research first. His first message: build “a version of Grok Bot or Meta’s Muse or OpenAI’s Dots… research them so you understand first”, to make sure the model knows what he means.
    2. Design before code. He designed every screen in Paper (a design tool he calls “Figma for the AI era”) through its MCP server, going back and forth in plain English before any building. Method: diverge then converge. Ask for three or four variations, then combine “this part from choice one and this part from choice three”.
    3. Pick the stack, then make the agent read the docs. For every feature: “read the docs and read the API references” of the service involved.
    4. Fix by screenshot. Screenshot bugs, things you dislike, and UI elements you like in other products (he used Grok Bot’s), and send them to the agent. Isenberg adds: keep a swipe file of designs and copy you admire to feed models.
    5. Give each bot what makes it useful (email, phone, computer), pick hosting (Orgo, Fly or Vercel), and ship fast with a short demo video. His was one 36-hour thread in T3 Code on Opus 5.5 at high effort, with loops running overnight and no GitHub connection; he calls building that way idiotic but it worked. The launch video got about 75,000 views.
  • The offer for small businesses (one to few). About 3,000 a month retainer for maintenance. Nick’s reasoning: owners “don’t want to… feel like they’re held by the neck on a retainer for these AI employees”, and selling them a system they own avoids that. He says an earlier model, renting a Hermes agent to a business for $5,000 a month, met a lot of friction.
  • The value conversation. Isenberg, playing a dental-clinic owner, asks how 36K a year pays back. Nick’s answer: point at the front-desk bot and ask what the person doing email, support, billing and bookings costs (“50 or 60 or $70,000 a year”), then say that is one of the 15 bots in the package.
  • Making it hard to refuse (Isenberg’s additions):
    • Include a redesigned website at no extra charge, shown to them already built.
    • Frame the 12 an hour or something like that”, his rough figure).
    • Add a guarantee tied to their bottleneck: if they are not replying faster or texting customers, they get the $3,000 back.
    • Bundle lead-magnet tools, for example a braces-cost calculator on an orthodontist’s site.
  • The product route (one to many). Grok Bot costs $200 a month, per Nick. Isenberg estimates an industry-specific version commands 10 to 20 times that, with seat-based pricing and expansion revenue. Niching gives pricing power and less competition.
  • Finding the niche: diverge before you converge. Nick’s warning: people pick a niche, build a site and messaging, then find a week later they don’t care about it. Instead:
    1. call the 50 people in your phone who run a business, know someone who does, or are senior in one;
    2. explain what you are building and ask if they would be interested;
    3. do the first one free as a case study;
    4. sit with that client, have them share their screen through a manual process (an invoice, a customer presentation) and build agents for what you see. His first niche was promotional products (company swag), an industry he did not know existed.
  • Isenberg’s case-study terms. Tell the free client you normally charge 5,000 credit for the intros makes it easy for them. Nick: this keeps every sales call warm, “like dating… I want to meet you through a friend.”
  • On “anyone could build this in a weekend”. Nick cites an a16z post that about 2% of the US pays for AI tools, so the skill is rarer than builders assume. Isenberg: the moat is “creating value for customers”, and clients also buy because they like you.

Try It

  1. List your first 50 calls before choosing an industry: owners, friends of owners, senior employees.
  2. Do one free build in exchange for a written case study and three text introductions; sit through a screen-share of one manual process before designing any agent.
  3. Prototype the team design first: research two or three existing agent products with your coding agent, then ask a design tool for three or four variations of each screen and combine the parts you like.
  4. Give each bot one platform to own, logged in, with its routines.
  5. Price the package against a named salary (the front-desk role) rather than per bot, and attach one guarantee tied to the client’s stated bottleneck.
  6. Bundle one cheap-to-build asset (a website refresh or a calculator lead magnet) into the proposal.
  7. Before forking BOPS, test whether a configured Grok Bot, Muse or Dots account already does the client’s jobs; the case for a custom build is the industry setup and the client owning it.

Implementation

Tool/Service: BOPS (open-source, link in the episode’s show notes, not in the transcript); Orgo for per-bot cloud computers (can also host); an email-for-agents service captioned “agent mail”; a phone-number service captioned “agent phone”; Composio for connectors (“one API connection… I get all of these connectors”); Paper with its MCP server for design; T3 Code (he says it is free, works with an existing Claude subscription, and he thinks it is open source) running Opus 5.5; CleanShot for screenshots.

Setup: Fork the repo and point your coding agent at it, or rebuild with the research → design → stack → build → screenshot-fix sequence above.

Cost: Not stated for any service. Nick’s prices to clients: ~3,000 a month optional.

Integration notes: Each bot gets its own computer, which Nick presents as an advantage over Grok Bot “for privacy reasons and just multitasking reasons”. That claim about Grok Bot conflicts with Grok Bot’s product lead; see Open Questions.

Open Questions

  • Does Grok Bot give each bot its own computer? Nick says “each bot gets its own screen, but it doesn’t get its own computer.” Grok Bot’s product lead says the opposite, first-party (“it’s actually really important that these bots have their own computer”). Nick sells per-bot computers through Orgo. The conflict is logged as an unresolved contradiction in Grok Bot.
  • No client results, revenue or retention for the 1M+” is not substantiated in the transcript.
  • The exact names and pricing of the email and phone services are not given (captions: “agent mail”, “agent phone”).
  • Isenberg’s 10-20x niche premium and “$12 an hour” are off-the-cuff estimates.
  • The a16z “2% of the US pays for AI tools” figure is relayed, not checked.