AI Agents Replacing Workers: CNET, Paperclip, and My OpenClaw Team
Prefer to watch? Watch on YouTube · Subscribe for new builds · Read the transcript
In this video I break down how companies have started giving AI agents real jobs with real responsibility, budgets, and decision making power, and I compare that to the agent team I run for my own content business. It's for builders, small business owners, and anyone wondering what enterprise AI adoption means for them. I recorded this in March 2026, and these tools change fast, so check the current docs before you build anything based on it.
What I cover in this video
- 0:00 AI agents getting real jobs, and Block laying off 4,000 employees
- 2:18 The CNET story: an AI agent named Sarah running procurement
- 4:35 Paperclip, an orchestration layer for managing many agents
- 7:35 My agent team on a Mac Mini with OpenClaw
- 10:40 Why persistent memory makes agents different from n8n or Make.com automations
- 12:57 The economics: human employees vs AI agents
- 17:33 What this means for your business
- 18:20 How to start: pick one workflow and build from there
Companies are giving AI agents real roles
A CNET feature covered companies that treat AI agents as employees, with job titles, responsibilities, and performance reviews. One company has an agent named Sarah who handles procurement. She negotiates with vendors, approves purchase orders up to $50,000, and manages supplier relationships. After 8 months, her review showed she saved the company $2.3 million and cut the procurement cycle from 2 weeks to 3 days.
Another company uses an AI agent as a night shift supervisor. It monitors production lines, makes real time adjustments, coordinates with maintenance, and handles scheduling conflicts. It has run night shifts for 6 months without a single major incident. I relate to the scheduling part. I run a resell business with employees, and I've handed scheduling off to Claude completely.
What stood out to me is that these agents have memory, context, and decision making authority. According to the feature, the bottleneck for these companies is no longer what AI can do. It's integration, training, and finding the right systems to manage teams of agents.
Paperclip and the orchestration layer
That's where Paperclip fits. It's an orchestration layer that sits on top of existing AI models so a company can run Claude agents, GPT agents, and custom agents from one dashboard instead of building everything from scratch. At launch they reported over 50 million agent interactions per day across enterprise customers.
One case study was a logistics company that went from 24 human night shift workers to 12 AI agents plus three human supervisors. They saw about a 60% cut in operating expenses, delivery accuracy went up 23%, and they can handle 40% more volume on the same infrastructure. Paperclip handles agent to agent communication, shared context across models, real time monitoring, and fail safe protocols for edge cases.
How my own agent team works
My setup is a Mac Mini running OpenClaw, and I built it over about 30 days. The agents are named after the Wu-Tang Clan:
- Atlas is the chief of staff, running on Opus 4.6. It coordinates the other agents, decides which trends are worth pursuing, and makes strategic calls about content direction.
- The RZA fires off at 6:00 a.m., scrapes social media, analyzes trending AI topics, and writes hooks, talking points, outlines, and calls to action. I wake up to three to five script options that get me about 90% of the way. I still edit them so they sound like me and not AI slop.
- Inspectah Deck turns the main script into Twitter threads, LinkedIn posts, and Instagram stories, schedules them through Post Bridge, and writes reply templates.
- Prodigy is a podcast scout. It searches X through the Grok API for people in my niche to invite on the podcast. That used to take me at least 30 minutes a day.
The difference from a normal n8n or Make.com automation is persistent memory. The agents know my voice, my goals, my audience, and what I've already covered. I gave the RZA OAuth access to my YouTube channel so it can check click through rates, which thumbnails worked, and where viewers dropped off. I set up memory right after locking down security.
I track all of it in a mission control dashboard built in Next.js. It shows what each agent is working on, success rates, cost per operation, and overall system health. The whole thing runs on about $15 a month in API costs plus OpenClaw, compared to around $400 a month I used to spend on separate AI tools. My content output is up at least 300% while my time spent is down 70%. Things still break sometimes.
The economics and how to get started
A human employee in most enterprise roles costs $80,000 to $150,000 a year once you count salary, benefits, training, and overhead. An AI agent with similar capabilities costs $2,000 to $10,000 a year in compute and infrastructure. For data analysis, coordination, and routine decisions, I think agents are already better than people. The companies doing well with this use agents to handle work people can't do efficiently or don't want to do. I still make the strategic decisions, appear on camera, and answer comments myself.
You don't need an enterprise orchestration layer to start. You can build a system like mine for under $50 a month. Don't do what I did and try to automate everything at once. Pick one repetitive task that eats hours of your week and build an AI agent for it:
- Content creators: research and script generation
- Ecommerce: customer service or inventory management
- Service businesses: lead qualification or scheduling
Build it, test it for a few weeks, measure the results, then move to the next workflow. Use what's already out there, like OpenClaw, Claude, or ChatGPT with custom prompts and some basic automation.
Key takeaways
- Companies are giving AI agents real roles, like procurement and night shift supervision, with measurable results.
- The hard part at scale is managing many agents, which is the problem Paperclip's orchestration layer is built for.
- Persistent memory and context are what separate an agent team from a basic automation.
- My OpenClaw content team runs on about $15 a month in API costs and took around 30 days to build.
- Start with one workflow, test it for a few weeks, and expand only after it works.
FAQ
Are companies actually hiring AI agents as employees?
Yes. A CNET feature covered companies giving agents job titles and performance reviews, including a procurement agent that approves purchase orders up to $50,000 and saved its company $2.3 million over 8 months.
What is Paperclip for AI agents?
Paperclip is an orchestration layer that sits on top of existing AI models so companies can manage Claude, GPT, and custom agents from one dashboard. It handles agent to agent communication, shared context, monitoring, and fail safe protocols.
How much does an AI agent cost compared to an employee?
In the video I cite $80,000 to $150,000 a year for a human in most enterprise roles, versus $2,000 to $10,000 a year in compute and infrastructure for an agent with similar capabilities. My own content agent team runs on about $15 a month in API costs plus OpenClaw.
How should a small business start using AI agents?
Pick one repetitive task that takes hours each week and build a single agent for it, like scripts for creators, customer service for ecommerce, or lead qualification for service businesses. Test it for a few weeks, measure results, then expand.
Chapters
Transcript
Read the full transcript
0:00AI agents are actually getting hired for real jobs, and that's what I'm going to be breaking down in this video. No more automating spreadsheets, not answering basic customer service questions, actual jobs with real responsibility, real budgets, and real decision-making power. While everyone's been arguing about whether AI will take jobs and everyone's in fear about, "Am I going to lose my job to AI?" Companies are actually doing it and saving millions in the process by hiring these agents to outperform humans in specific roles. If you have heard the news lately, Block, uh the founder of X company Jack Dorsey, they laid off 4,000 employees last week all because they found a cheaper way to outsource their job production. So, let me break it down and describe on what is actually
0:45happening. Right now, these stories are coming out left and right, and the real numbers behind enterprise AI adoption are real. So, that is why it matters way more than just another ChatGPT update. I want to focus on what is actually happening in the real world with AI in actual big business. Because here's the thing, I've been running my entire content business with AI agents for the last 30 days and realizing, "Wow, I don't need to hire a thumbnail designer. I don't need to write hire a marketing chief strategist. I could just do all of this on my own through a an LLM tailored for my business needs." So, every morning at 6:00 a.m. these agents fire off and do different types of tasks for the business that I'm trying to run. And
1:32it wakes up before I do. I wake up at 7:00, agents are up at 6:00, and by the time I'm up, I already have the full game plan for what to be done throughout the day. So, basically they scan trends, they write scripts, they generate thumbnails, social media posts, and have everything ready in the mission control behind me. So, now we're seeing the same type of thing at a much larger scale in Fortune 500 companies. So, the CNET story that broke just last week, they published a feature about AI companies actually hiring AI agents as employees, not just tools, not assistants, actual employees with job titles, responsibilities, and performance reviews. One company they featured has an AI agent named Sarah who handles their procurement process. This is
2:18crazy. Sarah negotiates with vendors, approves purchase orders in up to $50,000, and manages the supply relationships. She's been doing this for 8 months. She. Her performance review, she saved the company $2.3 million while reducing procurement cycle from 2 weeks to 3 days. Another company hired an AI agent as their night shift supervisor. Don't know what exactly that means, but this agent monitors production lines, makes real-time adjustments, coordinates with maintenance teams, and even handles the scheduling conflicts. So, I run a resell business that's doing over seven figures, and similar to kind of how I hate making schedules for the employees,
3:04Claude does all the scheduling for me, and it has made scheduling so much easier. So, I feel the pain on that. I think when you're running a small business and it has to do with having employees, running a schedule and writing a schedule can be a real pain sometimes. I have completely handled that task off to Claude. So, kind of going back to running sort of more of the grunt work in some of the business, this agent has been running night shifts for 6 months without a single major incident. You're seeing these big companies start to adopt these at such a mass scale, and I can kind of relate that back to kind of the smaller business stuff that I'm doing on the side. So, here's what caught my attention. These aren't just simple automation scripts. These agents have
3:49memory, they have context, they have decision-making authority, they learn from past interactions, and then they improve on each conversation. They build relationships with human colleagues, they get promoted based on performance. This is crazy. CNET interviewed the head of operations at one manufacturing company who said, "Our supervisor doesn't call in sick because he's AI. He doesn't need vacation time and makes better decisions than most human supervisors because it is has access to real-time data from every system simultaneously." That's not automating tasks, that's replacing entire roles. Insane. But, here's even the crazier part. These companies are struggling to hire enough AI agents, not humans. The
4:35bottleneck isn't AI capability anymore, it's integration, training, and finding the right orchestration systems to manage agent teams, which brings me to the second big story that I saw within this last week is Paperclip. Paperclip is just launched, it's an orchestration layer. You can check out the GitHub repo, but if you haven't heard of Paperclip, they're building infrastructure layer for enterprise AI AI agents. Think of it as an operating system that lets companies manage hundreds or thousands of AI agents across different departments. Their new role in their new orchestration layer does something really interesting. Instead of companies building custom AI solutions from scratch, Paperclip provides the management layer that sits
5:20on top of the existing AI models. So, a company can deploy cloud agents for customer service, for GPT agents, for content creation like what I'm doing, and custom agents for specialized tasks, all managed through one dashboard. The numbers from their launch are pretty crazy. They are processing over 50 million agent interactions per day across the enterprise customers. That's not 50 million simple queries, that's 50 million complex multi-step workflows handled by AI agents. One of the case studies shows an AI logistics company that replaced their entire night operations team with 12 AI agents. These agents coordinate deliveries, handle route optimization, manage driver communications,
6:05and resolve delivery issues in real time. Hello UPS, still waiting on my package 3 weeks later. So, the company that went from 24 humans night shift workers went to 12 AI agents plus three human supervisors. The cost savings? About 60% reduction in operational expenses. When it comes to big business, it all goes back to the bottom line. And like I said before, these AI agents don't call in sick, they don't care about vacation days, they are working non-stop. That's the beauty. So, the performance improvement is even more dramatic. Delivery accuracy went up 23%. Like I said, hello UPS, where's my packages? Customer satisfac- Customer satisfaction scores improved and they can now handle 40% more volume with the
6:50same type of infrastructure. Paperclip's orchestration layer is what makes this possible. It handles agent-to-agent communication. It manages shared context across different AI models. It provides real-time monitoring and performance analytics. It includes fail-safe protocols when AI agents encounter edge cases. But what really got my attention is how similar this is to what I've built for my own content business. Just kind of reflecting over these last 30 days building out this content pipeline. Reading a lot of these stories have a lot of similarity to what is being built for some of these big Fortune 500 companies, just at an enterprise scale. So, let me show you exactly how my agent team works and talk about it a little bit more. Every single day my business
7:35runs entirely autonomously. Things do break from time to time, but it is nice waking up with trends already analyzed, uh emails sent, newsletters sent, you know, a checklist for my calendar. I mean, fully autonomous workflows that were built over these last 30 days, and it helps me focus on more money-making activities and buying back my time. So, my actual setup is a Mac Mini right behind me. I have OpenClaw. Atlas is the chief of staff running on Opus 4.6, manages the entire operation, and coordinates with other AI agents on the team, handles the strategic decisions. We have a war room, we have daily meetings, making sure the agents got done, what they learned, and any errors that came about. The RZA, we have all
8:21the members named after the Wu-Tang Clan. We got the RZA, checks the trends, scanning the script generation, making sure that everything that is prepped for the day, you know, as far as scripts, things like that. And that fires off every morning at 6:00 a.m., scraping social media, analyze the topics for the trending AI space, identifies content opportunities as well, generates complete video hooks and talking points, outlines, call to actions. And by the time I wake up, a lot of this is done for me. I get to you about 90% of the way there. You do need to humanize it and make it sure that it's just in your voice and not a bunch of AI slop. So, I have three to five script options that I could choose from, that I could scan through to kind of make it a little bit
9:08more personable. So, that's the option that's sitting in the mission control dashboard behind me here. And Inspectah Deck, the other agent, handles a lot of the social media content, like Twitter threads and LinkedIn posts and Instagram stories. It takes the main video script and adapts it to a social media format, schedules the posts through Post Bridge, and even generates reply templates for engagement. So, I added another agent, which is a podcast scout. It's just called Prodigy. Prodigy basically scrapes X through the Grok API and looks for other people in our niche to get on our podcast, to share ideas, to share what they're building, and kind of keep you up-to-date on what's happening in the AI space. So, this used to take me
9:54at least 30 minutes a day, you know, not a ton of time in the grand scheme of things, but over a course of 7 days, yeah, I do have now vetted people that I could reach out to and send a personal DM to see if they want to jump on the podcast. So, that's the cool little loop that I have sneak peek. We have more agents and ideas being built out every day. Um but, that did take me at least 30 days to build out the system that I actually wanted to build that serves my needs. Your system's going to look different for what you're trying to build and what you're trying to achieve, but that's what I have going on. The entire system runs on literally $15 a month in API costs and then open clock. So, compare that to around $400 a month that I was spending on separate AI tools before I built the system. Here's the
10:40key difference between my setup and like a traditional automation, like something you would have like in an N8N or a make.com. These aren't just like separate tools that I manually coordinate. They're all agents with persistent memory and context about my business, my voice, my goals, my audience. This is the first thing you really need to set up after you get security locked down. So, when the Reza writes the script, he remembers the topics that I've covered in recent episodes and in recent videos and checks what performed well and what didn't and what my current business priorities are. So, for example, I just gave it our YouTube channel with the OAuth access so it could check, you know, click-through rates and what thumbnails performed well and where people dropped off and things
11:26like that and really double down on the videos that perform well and what you guys want to see. So, I have a little bit more of a accurate uh vision for the videos to produce on this channel. So, when Inspect Deck creates these social media posts, he knows my posting schedule, my engagement strategy, and how to adapt content for different platforms while maintaining my voice, humanized voice, not AI slop. Atlas coordinates everything, and this has been the main thing. Atlas is like the brain that fires off all the instructions and the rules to other agents. So, he decides which trends are worth pursuing, prioritizes the content based on business goals, makes strategic decisions about when to pivot topics and try new formats. The result has been pretty astonishing. We have 300 new
12:12subscribers in the last 3 days on this channel generating 34,000 views in the last 24 hours. Thank you so much. If you're new here, consider subscribing if you like anything AI news. I'm going to continue to document my journey, keep you up-to-date in AI news world, but also show you like what I'm building and what I'm shipping. So, my content production has increased at least 300% while my time investment has been decreased 70%. I'm not spending a ton of time building this out anymore now that the infrastructure is relatively finished and built out. It kind of just runs on its own. So, I'm publishing more videos, more social media content, more newsletter, more blogs, and seeing better performance across all metrics. Here's the real thing that kicks it all off. His revenue from the content
12:57business is starting to now actually grow since implementing this six-agent system months ago. That's just an extra bonus, not why I do it, but it's pretty cool to see. So, seeing companies like the ones in the CNET feature doing similar things at just a much larger scale makes perfect sense. If I can run a content business with four AI agents, imagine what manufacturing company can do with 50 agents or a logistic company can accomplish with 200 agents. The economics behind the shift is pretty straightforward. A human employee at most enterprise rules cost between 80 to 150,000 dollars per year when you include salary, benefits, training, and overhead. Not to mention they'll call in sick every now and then. So an AI agent
13:44with similar capabilities cost between two and 10,000 dollars per year in compute and infrastructure cost. Now doesn't that just make sense? Like when you are running a business, you want to be able to minimize as much overhead as possible, which is why I truly believe you're going to start to see million dollar businesses being ran by no humans at all and it will be all AI agents managing and running the business better than a human could. I hate to say it, but that is what's coming, ladies and gentlemen. It's not just about the cost anymore. AI agents don't get tired. They don't make emotional decisions. They don't call in sick, like I said, and they could process information from multiple sources simultaneously. So for specific types of work, especially like
14:29data analysis, coordination, routine decision-making, agents are already better than humans. I just said it. Stone me if you want. They are already better than humans. Paper click orchestration layer is addressing the biggest challenge with this enterprise AI adoption, which is management complexity. When you have dozens or even hundreds of agents, you need systems to coordinate all of this, monitor their performance, handle failures gracefully, and maintain consistent behavior across organization. That's exactly what I built for my own business inside of our mission control. It's just a Next JS dashboard that shows me what every agent is working on, their success rates, cost per operation, and overall system health. I can see it when The Rizzle
15:14generates a script, when Inspect Deck publishes a post, and when Atlas makes a strategic decision about like the content direction. But I'm a one-person business with four agents under me. Enterprise companies need to manage hundreds of agents across multiple departments with different goals, different access levels, and different performance metrics. They're all tracking different KPIs. They all have different SOPs. So, Paperclip solution makes sense. It's basically like mission control for enterprise. Real-time monitoring, agent-to-agent communication protocols, shared knowledge base, and integration with existing enterprise systems. The companies succeeding with AI agents right now are the ones that understand this. It's not about replacing humans entirely, but it's about augmenting human capabilities and
16:01handling the tasks that humans either can't do efficiently or just don't want to do. In my business, I still make the strategic decisions. I still appear on camera. This is me, not a HeyGen avatar. So, I still engage in all the comments that you guys leave. And like I said, I'm so thankful for all the new subscribers. If you guys like the content or want uh, you know, any, you know, get leave just a comment down below on things that you want me to cover, maybe my personal builds. I do have a build-in-public playlist here on when I shipped my first app and grew it to over $300 MRR currently. Um, so just let me know what you want to see. But, I'm showing my process with building with an OpenClaw, and I'm all about buying back my time, and that's what I'm sharing now currently. So, and that's
16:47how I kind of share it in our Shipping School community for people that want to do something similar. We do three boot camps a week, by the way, sharing the journey. Maybe you're having a hard time getting this thing off the ground, but we're here to help you. And uh, that's the goal. That's the goal. So, the agents that I have going, they handle a lot of the research in the writing and the social media management and content distribution. So, it allows me to have more time to be a more one-on-one with people that need help. So, that's the result of it all is like I could focus on the high value activities while my agents handle the operational work. Also, too, I I manage actual people coming to our warehouse every day. We have eight employees that you know, need to be managed and trained in our in our live resell business as well. So, that
17:33takes time. So, these are the agents that we got going on. We use them for data analysis, routine decision-making, and and that's the cool thing. You know, you're able to buy back that time and and get things going for the high value activities that you have going on in your life. So, like what does this mean for your business? Like if you're building a business right now and you're wondering what the heck does this even mean for me? If you're not thinking about AI agents, it's probably time you start, even if it's just for optimizing your life to be able to have a little more time back. And if you're not thinking about it, you're already behind. I'm sorry to say. The companies that I'm seeing succeed in 2026 are the ones that have figured out how to integrate AI into their core operations, not just use it as a tool. The barrier to entry is lower than you think. Like
18:20you don't need paper clips enterprise orchestration layer to get started. You could build an AI agent system like mine for under $50 per month that handles a significant portion of your business. The key is starting with one workflow and building from there. Don't try to automate everything at once like I did, spent 12 hours every single day, and then by the time you have 24 agents doing 24 different tasks, things break. So, pick with one repetitive task that you do multiple hours every day or every week and then build an agent to handle it. For content creators, this is huge. Start with research and script generation. For e-commerce businesses, start with customer service or inventory management. For service businesses, I would probably start with lead qualification or some type of scheduling. Build the agent, test it for
19:06a few weeks, measure the results, then expand to the next workflow. Don't overcomplicate this. The mistake I see most people make is trying to build the perfect AI system from day one. Start simple, and it's going to take time, and things will break. Use existing tools like Open Claw, Claude, ChatGPT if you want, with some custom prompts and some basic automation. Focus on solving one real-world problem at a time, and worry about the enterprise-grade orchestration layer when you get there, if that's what you're trying to build. Here's like the critical part. The companies that wait too long to start experimenting with AI agents are going to get disrupted by the companies that embrace this technology. The cost savings and efficiency gains are too significant to ignore. If your competitor can deliver the same service at 60% of your cost because they're
19:51using AI agents, you're in trouble. If your competitor can respond to customer inquiries in 30 seconds instead of 30 minutes because they have AI agents handling the the first-tier support, you're going to lose customers. This isn't theoretical anymore. It's happening right now across every industry. The enterprise adoption we're seeing this week, especially on CNET being featured in all these articles, Paperclips launch is just the beginning of it all. Within two years, every mid-size to large company will have AI agents handling significant portions of their operations. If you're on the fence about this, this can seem very scary scary and fearful, but I think it's a beautiful time to be alive, guys. To build skills, to get skills in AI, how to use AI, how to use AI to be a
20:37better person, to learn skills faster. This is going to be a huge opportunity for people that want to run their own business, be an entrepreneur, or maybe you're already one of those things and you just want to become more efficient. Now's the time. So, the question isn't whether AI agents will become mainstream in enterprise. They already are. The question is whether you'll be ready to transition when this accelerates. If you want to start building with AI agents yourself, that's exactly what we focus on the Shipping School community. We don't just talk about AI and hang out and go, "Wow, great. We actually do stuff. We build. We ship. We learn new technologies that come out." That is why we do three classes a week and we're going to be implementing five per week working with you with the project that you want to start and getting you from zero to one as fast as possible. So, I
21:22walk you through setting up these agent workflows, connecting them with your business processes, and then measuring the impact of your bottom line. So, for now, like I said, just three calls a week with me and literally working with you on building these systems. I created a Next.js starter kit mainly for web applications to help you ship your app within a day. So, you can use Claude code, ChatGPT, whatever. We talk about about them all inside of Shipping School and you'll go from basically an idea that no one knows about to a shipped product within a week. So, the goal isn't to build cool demos, even though that is amazing. It's to build systems that actually make your business more efficient, more profitable, and more competitive because based on what I'm seeing in like the enterprise space, AI agents aren't the future anymore. They
22:08are now. And if you're not building with them, you're going to fall behind. That company's already doing this. You don't want to be left behind. So, the link to Shipping School is down below if you want to check it out. And if you haven't subscribed already, what are you doing? Come on now. We got more videos coming to you tomorrow. We'll see you in the next one.
Ready to start building with AI?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.
Join Shipping SkoolKeep reading
Agency-Agents GitHub Repo: 51 AI Agent Personas for Your Business
Agency-Agents is an open-source GitHub repo by msitarzewski with 51 specialized AI agent personas, like a marketing director, sales manager, and technical debt specialist. You load a persona into Claude, ChatGPT, Cursor, or another AI tool and talk to it like a specialist on your team.
Paperclip AI Agent Orchestration: Run Your Agents Like a Company
Paperclip is an open-source, self-hosted Node.js platform that runs your AI agents as a company, with org charts, per-agent budgets, goal tracking, governance, and a dashboard. It works with OpenClaw, Claude Code, Codex, and Cursor, and you install it with npx paperclipai onboard --yes.
OpenClaw Content Machine: 13 AI Agents Behind $5,400 in 10 Days
I used OpenClaw running a 13-agent content machine on cron jobs to make $5,400 in 10 days. The agents handle research, scripts, thumbnails, clips, blogs, newsletters, and outreach, and that content fed a YouTube channel and a paid community. You can apply the same funnel to any niche.
