Making Money With OpenClaw: From Personal Assistant to Revenue Machine

⬅️ Back to Videos

▶️ Watch the Video

📝 VIDEO INFORMATION

• Content Type: Tutorial / Live Demo / Discussion
• Title: “Making Money With OpenClaw”
• Creator(s): Nick Vasilescu (Founder, Orgo), Greg Eisenberg (Host, Startup Ideas Pod)
• Platform: YouTube
• Publication Date: February 2025
• Link: https://www.youtube.com/watch?v=i13XK-uUOLQ

🎯 HOOK

Everyone sees the viral OpenClaw demos; the grocery ordering, the personal assistant, the cute personal automations. But demos that go viral are, in Nick’s words, “a little bit toyish.” The real money is somewhere else entirely: finding the specific workflow inside a real business that nobody has automated yet, and being the person who does it. Nick Vasilescu, founder of Orgo, sits down with Greg Eisenberg to show exactly how that works; from spinning up client workspaces in under a minute, to using sub-agents to parallelize Upwork job hunting, to building a live TikTok trend hunter from an idea in Greg’s Idea Browser. This is the episode for people who want to go from “I have OpenClaw installed” to “I have a paying client.”

💡 ONE-SENTENCE TAKEAWAY

OpenClaw is not just a personal assistant; it’s a deployable business tool, and the fastest path to revenue is finding a single specific workflow inside a real business, automating it end to end, and using that case study to go deeper into that vertical.

The Core Argument: OpenClaw Is a Business, Not Just a Tool

Every demo that goes viral on Twitter shows OpenClaw doing something clever and personal; ordering groceries, managing a to-do list, summarizing emails. Nick opens the episode by naming this directly:

“All the demos that go viral are a little bit toyish. They’re flashy. But the real power is finding the thing that actually drives business outcomes, saves time for a business, and building the automation around that.”

His live example: he has OpenClaw deployed for a promotional distributorship client, navigating a legacy platform that has no API, downloading product reports, parsing the data, and uploading everything into a Zoho CRM to create a central source of truth. No fancy demo. Real work. Real client. Real money.

The mental model shift this episode is trying to create: stop thinking about what OpenClaw can do for you personally, and start thinking about what specific workflow inside someone else’s business you could automate for them.

The Wedge: Finding the Right Automation to Start With

Nick calls it “the wedge”; the single specific workflow you automate first that gets you in the door with a client.

The wrong approach: showing up to a business and offering to “automate stuff with AI.” Too vague. They won’t know what to say yes to.

The right approach: identify one concrete, repetitive, painful workflow; ideally one that:

• Happens on a regular schedule
• Involves data moving between two systems
• Doesn’t have a clean API connection already
• Costs the business real time from a real person

His example from a real client engagement: product data living in a legacy platform with no API, needing to get into a Zoho CRM. OpenClaw navigates the legacy UI like a human would, downloads every report, parses it, and uploads it. The business owner doesn’t care how it works. They care that it works.

This is also what Andreessen Horowitz is calling out specifically. Their position: properly verticalizing computer use agents; building a specialized OpenClaw deployment for a specific industry (manufacturing, real estate, insurance, distributorships) and helping that industry adopt it; is one of the major startup opportunities of the next few years.

The Upwork Hack: Finding Your First Client for Free

If you have zero clients, zero connections, and zero idea where to start, Nick’s answer is Upwork.

Here’s the insight: people are posting jobs on Upwork right now, today, asking for exactly what OpenClaw can do. The job categories to search:

• Robotic Process Automation (RPA); the legacy, clunky, breaks-if-a-button-moves version of what OpenClaw does better
• Desktop automation
• AI workflow building
• Automation pipeline

Budget ranges Nick shows live: $500, $1,000, $1,500, $3,000, up to $20,000 for more complex workflows.

His exact process:

1. Go to Upwork, search “robotic process automation”
2. Filter to your budget range
3. Read the job description; copy all the context
4. Paste it into OpenClaw or Claude Code: “How much of this can you build out as a demo based on this context alone?”
5. Build the demo
6. Send the proposal with the demo attached

He did this at scale: he spawned sub-agents (covered in the next section) to scan Upwork for jobs, build out mini-demos for each one, compare them, and pick the best one to apply to; all automatically.

Once you land the job and deliver: get the case study. Go deeper into that industry. Build specialized workflows for every vertical use case you can find in that space.

Sub-Agents: What They Are and Why They Change Everything

A lot of people are confused about what a sub-agent actually is. Nick gives the clearest explanation in this episode.

The problem they solve:

When you ask your main OpenClaw instance to do a long task, it’s occupied. You can’t talk to it. It’s holding a hot cup of coffee; Nick’s analogy; and it can’t also move furniture for you at the same time.

What sub-agents do:

Your main OpenClaw instance (the orchestrator) can spawn up to eight sub-agents, each with their own computer. The orchestrator manages them, checks their quality, and stays free to respond to you.

Two parallelization strategies:

Nick’s Upwork example used the second strategy: four sub-agents each scanning different Upwork job categories simultaneously, each building a demo proposal, and the orchestrator picking the best one.

For client deployments:

Nick’s recommendation is to think of every automation opportunity for a client as its own sub-agent with its own dedicated skill. Rather than building one giant agent that does everything, you build:

• Sub-agent 1: Twitter bookmark hunter
• Sub-agent 2: CRM updater
• Sub-agent 3: Report downloader and parser
• Sub-agent 4: Email monitor and trigger

The orchestrator coordinates them. You can still talk to the orchestrator. Nothing blocks.

“Think of it this way: if you asked it to move a desk and it’s holding a hot coffee, it can’t do both. Sub-agents give your main agent leverage; it becomes the manager, not the worker.”

Design Thinking for Automation: The Value vs. Effort Matrix

Before writing a single line of code or giving OpenClaw any instructions, Nick runs a design thinking exercise. This is the step most people skip, and it’s why they build the wrong thing first.

Step 1: Map every potential automation on two axes

Draw a simple 2x2 grid:

• X-axis: Effort / Cost / Time to build
• Y-axis: Value created for the business

Start with the top-left quadrant high value, low effort. These are your first automations.

Step 2: Map the workflow step by step

Once you’ve identified the right automation to start with, map the exact steps from trigger to outcome. Two tools Nick recommends:

• Figma (with the Figma MCP for Claude): describe the workflow, have Claude create a visual diagram directly in Figma
• Mermaid code + Excalidraw or TLDraw: if you don’t use Figma, output the workflow as Mermaid syntax and paste it into a diagram tool

The goal is a visual, step-by-step map that OpenClaw can follow tip to tail. Vague instructions produce vague results.

Step 3: Extract the workflow from real conversations

If you’re working with a client who doesn’t know exactly what they need automated, record your meeting (Gemini for Google Meet, Granola, or any meeting notes tool) and feed the transcript to Claude:

“What is the step-by-step workflow based on this transcript? Map it out.”

This works especially well when you’re entering an unfamiliar industry and the client uses terminology you don’t understand yet.

Building Automation Pipelines: OpenClaw vs. Claude Code

A question that comes up naturally: when do you just describe a workflow to OpenClaw, and when do you build a proper automation pipeline with Claude Code?

Nick’s framework:

Use OpenClaw’s native capabilities when:

• The task is relatively simple and conversational
• You need it to respond to ad-hoc requests
• The trigger is a Telegram message or something human-initiated

Build a proper pipeline with Claude Code when:

• The workflow is complex, multi-step, and needs to run reliably without human prompting
• You need Python scripts, API calls, or structured data handling
• The trigger is programmatic (a cron job, an email event, a webhook)

His real client example: the Zoho CRM pipeline has a cron job that monitors a specific email inbox. When OpenClaw is CC’d on a relevant email containing a product link, the cron triggers the Python script pipeline; which handles all the web navigation, downloading, parsing, and uploading downstream. OpenClaw is the listener and orchestrator. Claude Code built the underlying automation.

“You’re not relying too much on OpenClaw’s abilities in and of itself. You’re creating specialized AI workers underneath the OpenClaw that it can call individually.”

The Live Build: TikTok Trend Hunter

The most compelling part of the episode is what happens when Greg shows Nick an idea from his Idea Browser: a “TikTok trend tool that catches viral waves before they peak.”

Nick’s response: “Let’s build it. Right now. On camera.”

What they built, step by step:

Phase 1: Test the core capability

Before writing any code, Nick asks a playground agent (running in an isolated Orgo computer) to simply:

“Open Firefox. Go to TikTok. Scroll. Tell me what the most common videos are on the For You page. Give me a summary.”

The agent opens Firefox, navigates to TikTok, scrolls, takes screenshots, infers video topics from hashtags and thumbnails, and returns a summary. This proves the core action is possible.

Phase 2: Design the skill architecture

Nick pastes the full idea description into OpenClaw and asks it to break down a realistic build plan. OpenClaw identifies what it needs:

• Data access method (browser-based computer use via TikTok’s frontend)
• Niche focus parameters
• Output format

Phase 3: Build the programmatic version

Rather than relying on OpenClaw to manually scroll TikTok every time, Nick uses the Orgo API docs to build a proper computer use agent script. He pastes the llms-full.txt from Orgo’s docs into the playground and asks it to build a Python script that:

• Spins up a browser via the Orgo API
• Navigates to TikTok
• Scrolls the For You page
• Extracts: username, video description, category, like count
• Returns structured data

Phase 4: Deploy as a skill

The output becomes a skill attached to Greg’s OpenClaw instance; something it can call whenever it needs fresh TikTok trend data, without needing to be prompted step by step.

The result: A working (and still being debugged) TikTok trend hunter, built from zero to demo in roughly 10 minutes of conversation, while two people were also having a broader discussion about the state of computer use agents.

Nick’s honest summary: “It needs a little debugging as to be expected. We spent 10 minutes on it, but you get the gist.”

That’s the point. The speed-to-prototype is the story.

The MVP Principle: Start With a Skateboard

Nick uses a classic product design analogy to explain the right way to build automations:

“If I want to build a car, the first thing I do isn’t build the frame or the body. The first thing is to ask: why do I want a car? To go from point A to point B. So maybe I should start by building a skateboard.”

Applied to OpenClaw automation:

• Don’t start by designing the entire multi-agent pipeline
• Start by asking: what is the single simplest thing that demonstrates this automation is possible?
• Build that. Test it. Debug it.
• Then layer complexity on top of something you know works

In the TikTok build: the MVP was “can an agent open TikTok and scroll it?”; not the full trend detection system.

The Business Model: What Clients Pay For

Nick is explicit about the revenue opportunity. Three main ways he sees people making money right now:

1. Setup and management fees Executives (law firms, insurance companies, manufacturing) are reaching out willing to pay just to have someone install and configure OpenClaw for them. If you can get someone set up and running, that’s a service. Some are paying thousands of dollars for configuration alone.

2. Upwork automation jobs $500–$20,000 per project. The jobs exist. They’re posted. They’re looking for what OpenClaw does. The barrier is knowing how to build a demo quickly enough to make the proposal competitive.

3. Vertical specialization The long game: pick one industry, build deep expertise in what workflows that industry needs automated, create a library of pre-built skills and sub-agents for that vertical, and sell it as a managed automation service. This is what Andreessen Horowitz is pointing at when they talk about “verticalizing computer use agents.”

The Bigger Picture: Agents Are the New SaaS

Nick and Greg spend time in the second half of the episode discussing what’s coming:

The arbitrage opportunity is time-sensitive

The vast majority of businesses on the planet would benefit from what OpenClaw can do. Almost none of them know how to use it. The people who learn now, while the tooling is still rough and confusing, are the ones who build the case studies, the vertical expertise, and the client relationships before this becomes commodity knowledge.

OpenClaw’s creator was acqui-hired by OpenAI

During the episode, Greg mentions that Peter Steinberger, the creator of OpenClaw, has officially announced he’s joining OpenAI. The tool is in serious hands and getting serious investment from the industry.

Dario Amodei’s framing

Nick cites a recent Dario Amodei interview: the constraint to getting to AGI is computer use agents; AI that can operate a computer interface the way a human can, including visual interaction with UIs, not just text. OpenClaw, in Nick’s view, is the ChatGPT moment for this category of technology.

The asset-building angle

Every automation you build is an asset. Every skill you create for a vertical is reusable. Every case study compounds into the next client. Greg frames it as: this is what abundance from AI actually looks like in practice; not just replacing jobs, but enabling one person to build and own things that previously required teams.

“It comes down to taste now. If you have a good idea, you can just build it.”

What the Video Covers: Timestamps

(00:00) Intro Greg introduces Nick and sets the premise: by the end, you’ll know how to make actual money from OpenClaw.

(02:50) Getting Set Up with OpenClaw Nick live-demos spinning up an OpenClaw workspace using Orgo. Shows it’s possible with a single curl command in under a minute. Notes alternatives: Manus, Kimi, Mac Mini.

(05:02) Finding the Wedge The core business framing. Viral demos are toyish. Real money is in automating specific, end-to-end business workflows. Live demo of the Zoho CRM pipeline for a real client.

(07:39) The Upwork Hack How to use Upwork as a client acquisition engine. Search RPA, desktop automation, AI workflow jobs. Build a demo. Send a proposal. Live search of real jobs.

(09:41) Andreessen Horowitz on Computer Use Agents The VC thesis: verticalizing computer use agents for specific industries is a major startup opportunity.

(11:01) Setting Up a Client Workspace in Minutes Nick live-creates a new computer for Greg in Orgo, runs the OpenClaw install curl command, and shows how you’d hand this off to a client.

(12:41) Design Thinking: Value vs. Effort Matrix How to identify which automation to build first. Map on two axes. Start with high-value, low-effort. Use Figma MCP or Mermaid + Excalidraw to visualize the workflow.

(15:23) Using OpenClaw to Prioritize Automations Pass client call transcripts to Claude and ask it to identify the automation opportunities and rank them by value vs. effort.

(17:57) Building Automation Pipelines with Claude Code The distinction between what OpenClaw handles conversationally and what needs a proper Python pipeline built with Claude Code.

(19:33) Sub-Agents vs. Tasks vs. Skills Clear explanation of the three concepts. Sub-agents as orchestrated workers. Two parallelization strategies. The “hot coffee” analogy.

(23:22) The Scale of What’s Automatable When you show clients what’s possible, they light up. Once one automation is working, they have a hundred more ideas. This is where the relationship deepens.

(24:54) Live Build: TikTok Trend Hunter Phase 1: test the core capability (can an agent scroll TikTok?). Phase 2: plan the skill architecture. Phase 3: build a programmatic computer use agent with Orgo API docs.

(32:09) MVP Thinking Start with a skateboard, not a car. Build the simplest thing that proves the automation is possible. Debug. Then scale.

(32:41) Architecture of the TikTok Agent Detailed walkthrough of the Python script: Orgo API key, Anthropic key, browser spin-up, TikTok navigation, data extraction (username, description, category, like count).

(36:59) The Arbitrage Opportunity Most businesses need this and don’t know how to get it. The people who learn now build the moat.

(40:30) Agents Are the New SaaS The macro thesis. From productivity tool to business infrastructure. The comparison to SaaS as a category.

(42:42) Demoing the TikTok Trend Hunter The agent runs. It spins up a computer. It opens Firefox. It navigates TikTok. It starts working. Live, on camera, real-time.

(44:11) Building Assets and the Abundance AI Will Bring Every automation is an asset. The golden age of one-person businesses. What abundance actually means in practice.

(47:58) Closing Advice Get your hands dirty. Find a job on Upwork. Build a demo. Send a proposal. That’s how you start.

Practical Applications

If You Want Your First Client This Week

1. Go to Upwork, search “robotic process automation” and “desktop automation”
2. Filter to $500–$5,000 budget range
3. Read 5 job descriptions
4. Pick the most concrete one (specific workflow, clear input and output)
5. Give the job description to OpenClaw or Claude Code: “How much of this can you build as a demo?”
6. Build the demo; even if it’s rough
7. Send the proposal with the demo attached

One proposal with a working demo beats ten proposals without one.

If You’re Building for a Client

1. Get a recording of your discovery call (Gemini for Meet, Granola, any tool)
2. Feed the transcript to Claude: “What’s the step-by-step workflow? Map it out.”
3. Plot automation opportunities on the value vs. effort matrix
4. Pick the top-left quadrant item first
5. Build the MVP version; the skateboard, not the car
6. Test it. Debug it. Then layer on complexity.

If You Want to Build a Vertical Business

1. Pick one industry (manufacturing, real estate, legal, insurance, distributorships; all mentioned)
2. Land one client in that industry via Upwork or direct outreach
3. Build the automation
4. Get the case study
5. Use the case study to approach the next 10 businesses in the same vertical
6. Build a library of pre-built skills and sub-agents for that vertical
7. Sell it as a managed service

Common Mistakes to Avoid

Showing up without a wedge “I can automate things with AI” is not a pitch. Come with a specific workflow in mind; ideally something you already know is painful in that industry. The more specific, the more credible.

Trying to build everything at once Design the full pipeline in your head, then build the MVP first. If you can’t prove the core action works in isolation, you’re not ready to connect it to everything else.

Relying on OpenClaw to do everything natively Complex, reliable automation pipelines need real code; Python scripts, proper API calls, structured data handling. Use Claude Code to build the underlying system. Use OpenClaw as the orchestrator and human interface.

Skipping the workflow visualization If you can’t draw the automation as a step-by-step diagram, OpenClaw can’t execute it reliably. Map it out in Figma, Excalidraw, or even on paper before you start.

Using your personal accounts for client deployments Create dedicated accounts (Google, Gmail, etc.) for every client workspace. Clean separation, clear permissions, easy to hand off or shut down.

Notable Quotes

1. “All the demos that go viral are a little bit toyish. The real power is finding the thing that actually drives business outcomes.” Nick on the gap between viral demos and real deployments

2. “We’ve got jobs on Upwork asking for $500, $1,000, $5,000 for AI workflow automation. You can give it to OpenClaw, build a demo, and send a proposal. You have your first customer right there.” Nick on the Upwork hack

3. “OpenClaw can spawn up to eight sub-agents, each with their own computer. Right now, we’re at the phase of having one OpenClaw. It’s going to happen quickly; you’re going to want 10, 100.” Nick on where sub-agents are headed

4. “Think of your main agent as holding a hot cup of coffee. Sub-agents give it leverage; it becomes the manager, not the worker.” Nick’s analogy for the orchestrator model

5. “The constraint to getting to AGI is computer use agents. The ability to have an AI that can operate a computer like you and I can, but better.” Nick citing Dario Amodei’s framing from a recent interview

6. “Properly verticalizing computer use agents and assisting companies in adopting it will be a major area of exploration for startups.” Andreessen Horowitz, as cited by Nick

7. “It’s as easy as: if I want to build a car, I should start by building a skateboard. What’s the MVP that gets me to the dream outcome?” Nick on the MVP principle for automation

8. “The vast majority of businesses would love to have better automation, but they don’t know how. That’s the arbitrage opportunity.” Greg on the market gap

9. “It comes down to taste now. If you have a good idea, you can just build it.” Greg on what the abundance of AI really means for builders

10. “Get your hands dirty. Find a job on Upwork. Build a demo. Get some case studies. Go deeper into that vertical. That’s where we’re at.” Nick’s closing advice

Resources

• 📺 Video: Making Money With OpenClaw
• 🏢 Orgo: startup-ideas-pod.link/orgo; Nick’s startup for deploying OpenClaw on virtual machines with multi-instance management
• 💼 Upwork: Search “robotic process automation” and “desktop automation” to find paid jobs
• 🔗 Figma MCP: Figma’s Model Context Protocol integration for Claude that lets Claude create diagrams directly in Figma
• 📊 Excalidraw / TLDraw: Free diagramming tools for Mermaid-generated workflow maps
• 🎙️ Show: Startup Ideas Pod with Greg Eisenberg

Crepi il lupo! 🐺