What would it take to sell an AI agent to a business that does not want to become an AI lab? In this conversation with Greg Isenberg, Orgo co-founder Nick Vasilescu demonstrates Bops, the business-operations agent app he built, then proposes a $25,000 setup and an optional $3,000 monthly service. The useful lesson is how to package a working, owner-reviewed outcome for one niche. The "$1M+" headline is a business possibility, not verified Bops service revenue.
Find one repetitive, valuable workflow; prove a small version with the owner watching; then sell installation and upkeep as distinct services. Nick's Bops demo makes the platform tangible, but the current Bops license restricts competing products and hosted services. A public repo is not blanket permission to fork it into a rival business. Read the license and scope the client engagement before building.
Watch the Conversation
Nick is an Orgo co-founder and Bops creator. The interview discusses tools and a commercial offer involving his company. This article is independent; it does not endorse the offer or verify the demo's audience and revenue claims.
What Nick Actually Built
Nick calls Bops "business ops." In the demo, Boppy receives a call and a text, turns the request into tasks, and works across four screens on a cloud computer. The intended jobs include inbox triage, pipeline updates, invoices and reports. The product is a Mac app with a visible computer, not merely a chatbot with a new name. "Grok Bot for a business" is the episode's comparison, not the documented model provider: the current Bops README describes OpenAI agent and computer-use components.
His account of a 36-hour weekend build is a founder story, not a realistic estimate for a client-ready deployment. A prototype can skip procurement, access review, monitoring, support, training and the awkward edge cases in a real business. The current repository also differs from the simple "each bot has its own computer, email and phone" description: its hosted-plan notes say bots on Free and Pro share one computer, and phone/email availability varies. The product site and repository currently show different pricing presentations, so confirm the active plan before estimating a client's costs.
Research, Design, Build, Then Operate
Nick's sequence starts with research: compare existing assistants, observe how a target business works, and find the repeated task the owner will pay to resolve. He then uses Paper's design MCP to make several interface directions and combine the strongest parts. He describes a long-running build in T3 Code, checking each provider's API documentation and using screenshots to refine the result.
For a client service, add an explicit operational phase. Define who approves messages and payments, who sees logs, how a failed task is surfaced, how accounts are revoked, and what the customer keeps if the retainer ends. A good first workflow may be one agent drafting for review, not 15 agents with broad permissions. That is the difference between an impressive demo and a reliable service.
The Stack Behind the Demo, by Job
The Bops repository documents the current components. Nick's episode describes the build process; the table below maps the pieces to the job they perform. Each integration adds a cost and a permission boundary.
| Component | What it solves | Builder note |
|---|---|---|
| Bops | The business-facing app, agent team and task view. | Read its setup and FSL license before modifying or delivering it commercially. |
| Orgo | Cloud computers with screens the agent can use. | Separate computer-hours, storage and supervision costs from model tokens. |
| AgentMail | An inbox for an agent to receive and send email. | Start with drafts and an allowlist; outbound messages need a named human owner. |
| AgentPhone | Numbers, calls and texts for an agent. | Check availability, consent and disclosure rules for the client's location and use case. |
| Composio | Authenticated connections to business applications. | Grant only the scopes the pilot needs; test disconnect and token rotation. |
| Paper MCP and T3 Code | Design exploration and agent-assisted implementation. | Build tools, not runtime requirements for every customer deployment. |
Nick also mentions model, memory and research services. The current README distinguishes required keys from recommended and optional services, so do not buy every component before validating the first workflow. It notes that self-hosting requires an Apple-silicon Mac, Node, Orgo and OpenAI keys; calls and texts need more setup. Bops' client app on a Mac does not mean the agent's computer or model inference stays on that Mac.
Can You Fork Bops and Sell It?
You can read and adapt the code, but commercial reuse has a boundary. Bops currently uses the Functional Source License 1.1 with an Apache 2.0 future license. The repo's plain-language summary permits internal use and professional services for a compliant licensee, while prohibiting making it available as a competing product or substantially similar service. Each released version becomes Apache 2.0 after two years.
That makes "fork it today and sell your own Grok Bot clone" too broad. Serving a client that uses Bops may be different from selling a hosted Bops competitor, but the exact contract and deployment matter. Get a license review for your intended offer. An independently built workflow that uses vendor APIs is another route, with its own contracts, costs and responsibilities. This is a practical warning, not legal advice.
The $25,000 Offer: What the Numbers Mean
Nick suggests a $25,000 implementation, possibly split into milestones, then an optional $3,000 monthly retainer after a 30-day implementation. He discusses a website redesign, a free first case study and a guarantee as ways to reduce buyer friction. Those are offer ideas in the conversation, not a published price sheet or evidence that a typical buyer will accept them.
| Illustrative contract | Gross billed | Not included |
|---|---|---|
| Setup only | $25,000 | Implementation time, hosting, models, support and tax |
| Setup + 6 paid support months | $43,000 | Churn, usage spikes and ongoing labor |
| Setup + 12 paid support months | $61,000 | All delivery and operating costs |
At the full-year scenario, 17 clients would bill $1.037 million gross. That is arithmetic, not a forecast, profit figure or claim that Nick has sold 17 such contracts. The demanding question is whether one person can deliver, review and support that many client systems at the promised quality. Price the pilot from the actual workload and margin, then earn the right to expand.
Find One Client, Prove One Result
Nick's customer-development advice is refreshingly concrete: call 50 people you already know who own or influence a business. Ask where staff repeat manual work; watch one process with permission; pick a niche only after hearing the same problem several times. He proposes a first free build in exchange for a case study and warm introductions. A free pilot is still a contract: specify time, data, review rights, and what happens if it does not work.
A strong first test is narrow: for example, draft follow-ups for overdue invoices from a synthetic dataset and have the finance owner approve each message. Record minutes saved, correction rate, missed cases, cost per accepted draft and any customer confusion. Stop if the agent needs more supervision than the original process or crosses an access boundary. Only then consider sending, voice, payments or multiple agents.
Copy a One-Workflow Pilot Brief
This original prompt converts the episode's business idea into a buyer interview and a bounded, human-reviewed proposal. It does not authorize an agent to access accounts or contact prospects.
Validate one useful agent
Interview, scope, costs and a stop rule.
Video Chapters
The product and build
| Time | Topic |
|---|---|
| 00:00 | The solo agent-business idea |
| 03:09 | Bops, a custom agent team |
| 05:07 | Generic SaaS versus custom apps |
| 07:02 | Selling agent teams |
| 09:49 | Bops demonstration |
| 13:57 | One agent per platform |
| 15:07 | Research |
| 17:49 | Design |
| 22:12 | Choose the stack |
| 26:43 | Host and ship |
The service offer
Source Note and Limits
Checked 2 November 2026. The interview and supplied transcript are the sources for Nick's 36-hour account, demo, proposed pricing and customer-development advice. We did not inspect his sales contracts, verify the reported social views or reproduce the build. Identity, present features, plan notes and dependencies were checked against Orgo, Bops and its repository. Commercial-use wording comes from the license text. Vendor roles were checked against AgentMail, AgentPhone, Composio and Paper. The revenue scenarios, pilot metrics and copyable prompt are our analysis, not Nick's reported results. Plans and terms can change.