How an OpenClaw Agent Applied to 1,200 Jobs a Day and Built a $428K Salary Pipeline

By Beau Johnson··8 min read

Quick answer

The agent found LinkedIn job posts, filled out applications, submitted them at scale, and logged outcomes so the operator could measure conversion and pipeline value.

Most people still think AI agents are fancy assistants.

That framing is way too small.

An OpenClaw workflow built by Narek Gevorgyan applied to roughly 1,200 LinkedIn jobs per day and built a $428,000 salary pipeline in 30 days. Not because the model was magic. Not because the prompts were cute. Because the workflow turned a human process into software.

That is the real lesson here. Job hunting is just the wrapper. The deeper pattern is outreach automation at a scale a human cannot touch.

What happened in this OpenClaw case study

Narek built an OpenClaw agent that found LinkedIn job posts, filled out applications, submitted them, and kept running through the queue. His target was 8,000 applications per day. Real world friction cut that down hard.

LinkedIn captchas showed up. Rate limits kicked in. Filters slowed the system down. Final throughput landed around 1,200 applications per day.

Most people would look at that and say the system missed the goal.

Wrong read.

At 1,200 applications per day, the workflow still pushed through roughly 36,000 applications in 30 days. About 4 percent of companies moved forward in some meaningful hiring capacity. Only 12 percent actively rejected or fired the agent. The rest stayed open, warm, or in progress. Total pipeline value reached $428,000 in potential salary.

That is what matters. Not whether the first target number was perfect. Whether the system produced a strong aggregate result.

Why 1,200 applications per day changes the math

A motivated human might submit 200 job applications over a few months before burning out. This agent ran 180 times more volume in a fraction of the time.

That is the unlock with AI agents.

You are not asking software to think like a genius. You are asking software to execute a repeatable workflow with relentless consistency. Humans are still needed for judgment, interviews, offers, and final decisions. But the repetitive cold outreach layer can be delegated.

Once you see it that way, this stops being a job search story. It becomes a business operations story.

The real takeaway for founders, creators, and operators

Most of you are not trying to automate LinkedIn job applications.

But a lot of you are trying to do one of these:

  • Pitch prospects
  • Apply for brand deals
  • Reach out to podcast hosts
  • Submit to newsletters and directories
  • Follow up with leads who went quiet
  • Contact suppliers, partners, or distributors

Those are the same problem wearing different clothes.

You have a list of targets. You have a standard action. You have an outcome you want tracked. That is an agent workflow.

The bottleneck usually is not the model. It is not even the code. It is the founder failing to define the workflow clearly enough for the agent to run it without getting messy.

The 4 parts of an outreach automation workflow that actually works

1. A clean source of targets

Your target list needs structure. In this case it was LinkedIn job postings. In your business it could be leads from a directory, podcast hosts, creators, brand contacts, or ecommerce stores in a niche.

If the source data is sloppy, the workflow will be sloppy too.

2. One clearly defined action

The agent needs a repeatable action for each target. Find the record. fill the form. send the message. log the outcome. move to the next one.

The more standardized the action, the more reliable the system becomes. If an API is available, use it. Browser automation works, but UI friction like captchas and layout shifts will always lower throughput.

3. A logging layer

This is where a lot of builders get lazy, then wonder why they cannot improve the system.

You need to log what happened, when it happened, and what result came back. That is how Narek knew the workflow had a 4 percent forward rate and a 12 percent rejection rate. Without logs, you are guessing. With logs, you can optimize.

4. An autonomy boundary

This is the part that separates useful automation from reckless automation.

Just because an agent can keep going does not mean it should. The best pattern is simple. Let the agent do the repetitive volume work. Then surface wins, replies, or edge cases for human review. The machine handles scale. The human handles judgment.

That is the design pattern I trust most.

Why captchas and rate limits do not kill the opportunity

The system aimed for 8,000 applications per day and landed at 1,200. That sounds like a miss if you obsess over the original target. It looks very different if you care about results.

Even after platform friction, the workflow still created a six figure pipeline. That tells you something important. You do not need perfect execution to get asymmetric upside. You need enough throughput, enough consistency, and enough logging to find the real bottlenecks.

Friction is data. That is the move.

Instead of quitting when the platform pushes back, treat resistance like a map. It shows you where to switch from browser automation to APIs, where to slow down, where to add retries, and where to redesign the flow.

How to apply this OpenClaw pattern to your business

If you are in sales, media, ecommerce, recruiting, or outbound partnerships, this pattern is worth stealing.

  1. Start with 100 targets, not 10,000. Small batches make debugging easier.
  2. Write the workflow in plain language first. If you cannot explain the process simply, the agent will not run it cleanly.
  3. Test the first 10 actions. Watch where the workflow breaks.
  4. Log every result. Success, failure, retry, rejection, all of it.
  5. Scale only after the pattern is stable. Volume multiplies both wins and mistakes.

This is how real operators build agent systems. Not by chasing the flashiest demo. By getting one workflow stable, measurable, and scalable.

What this means for the future of AI outreach automation

The biggest shift is not that agents can do tasks faster. It is that they can run entire workflow layers continuously.

That changes what one person can produce.

One founder with a clean OpenClaw workflow can now execute like a small team, whether the use case is outbound sales, prospecting, market research, customer follow up, or content distribution. The companies that win are going to be the ones that design these systems first and put humans at the highest leverage point, not the busiest point.

That is the real story behind the 1,200 applications a day.

Not volume for the sake of volume. Leverage.

FAQ

What did this OpenClaw agent actually do?

The agent found LinkedIn job posts, filled out applications, submitted them at scale, and logged outcomes so the operator could measure conversion and pipeline value.

Why does this matter if you are not applying for jobs?

Because the same pattern applies to sales outreach, partnership outreach, sponsor pitching, directory submissions, and other repetitive workflows that depend on volume and logging.

What are the core parts of a workflow like this?

You need a clean source of targets, a clearly defined action, a logging layer to track outcomes, and an autonomy boundary that decides when the agent should stop and ask for human review.

Should you let an agent run fully autonomous?

Usually no. The best pattern is to let the agent handle repetitive volume work, then bring the human in for judgment calls, approvals, and high leverage responses.

Want help building workflows like this?

If you want to build real agent systems that create leverage in your business, join Shipping Skool. That is where we break down the workflows, test what actually works, and help builders ship faster without guessing.

Chapters

  1. 0:00The Story: $428K From a Cron Job
  2. 2:30The Math Behind 36,000 Applications
  3. 5:00Why This Pattern Works for Any Outreach
  4. 7:30The 4 Components You Need to Build This
  5. 10:00Autonomy Boundaries and Approval Design
  6. 12:00How to Start Small and Scale What Works

Transcript

Read the full transcript

0:00In this video, I'm going to be talking about an open claw use case that is very, very interesting, especially when it comes to the job market. We're going to be talking about a guy named Nerik. And Nerik applied to 8,000 jobs on LinkedIn with his open claw setup in one day. And after 30 days, he had created $428,000 pipeline in salary offers. And I want to be straight up with you because the reason I'm talking about this is not because of the job hunting is the whole point of the story. The point is is what this story shows you about what agents are actually doing when you treat them like software instead of a slightly better assistant. So, Nerik posted on his X a few days ago. He built an open claw workflow back in January that would

0:47find job postings on LinkedIn, and then it would fill out the applications, submit them, and the craziest part is is the agent was set to auto start. So, if he actually got hired somewhere, no human in the loop after that step. Just the agent going end-to-end on the entire process. His target was 8,000 applications per day. And you know what happened the moment he started running it? LinkedIn captchas, rate limits, posting filters, all of that kicked in and knocked his actual daily throughput down to about 1,200 applications per day. Still crazy when you think about it. Which, you know, in the end is only 15% of what he was aiming for, but still. And I think a lot of people hear that number and immediately think, well, you know, the thing right? It only hit 15% of the goal, but the framing was

1:34completely backwards. Here is what 1,200 applications per day actually means over the course of a month. 36,000 job applications in 30 days after running a full month of 4.0% of the companies had move forward with him. So, 4% of the companies move forward with him in some type of hiring capacity. And all of this happened on autopilot. So, only 12% had actively rejected or fired the agent. And that means the vast majority of contracts were still warm or in the process. And the total value of the roles he was in conversations about added up to $428,000 in potential salary. So, $428,000 essentially for like a cron job that's running in the background. So, if you

2:19just want to do the math out loud because I think when you hear it laid out, it really hits different. The most aggressive human job seeker, someone who is, you know, really grinding, might send out 200 to 300 applications over the next few months before they are completely burned out. Nerik sent out 36,000 job applications in 30 days. I mean, that's a startup idea in itself, just helping people find jobs. That's the way to go. Real quick, before we keep going, if you're watching this and you want to actually build with some of these tools >> [music] >> not just watch videos about them, then you're going to want to check out our community down below, Shipping School. We have a full Claude code course, a full Open Claw course, and four live bootcamps every single week where we actually help you get set up from scratch. Like actually set this thing

3:05up, not just watch a tutorial and figure it out by yourself. And we also provide one-on-one coaching, so you could book a call with me, we could share screens, and I can help you get Claude code or Open Claw running on your machine. That's it, no fluff. I built this community because watching YouTube's only gets you so far. We launched it just 3 days ago and we have over 55 members. You need people around you who are actually building, people who hold you accountable, and coaches who can help you when you get stuck. I'll put the link in the description down below. Get in now before the price goes up. And that is 180 times more applications in a fraction of the time. He did not work harder, he did not have a better resume, he had an agent running at scale that a human body physically can't match. And

3:51that is the whole point, and that is the whole thing. You know, that is the insight that gets missed when people talk about AI agents. Agents give you a software level distribution on human workflow. You're not replacing the human, you are giving one human the throughput of an entire team running 24 hours a day. The job seeker does not disappear from the process. Nerik is still the one who has to show up in the interviews, pass the technical assessments, and decide which offer to take. But the entire cold outreach layer, the part that normally burns people out, that got delegated to a machine. Which is pretty cool when you really think about it. I mean, that's the whole point of running Open Claw or, you know, if you're running something like Perplexity Personal Computer or the new Claude Co-work. That's the whole point of autonomous, redundant tasks

4:38that are handed off to AI agents, so we can work on the higher level stuff. Now, I know some of you are thinking about whether this is the right thing to do, whether if it's fair to other job seekers, whether LinkedIn should even allow it, and I totally get that. That is a real conversation worth having, but from a pure systems thinking perspective, what Nerik built is just a workflow running at scale. Every single action the agent takes is something a human could do manually. Find a posting, fill out a form, click submit, right? And then move on to the next one. The only difference is the volume. And volume is exactly what software does better than most humans. So, the more interesting question for me is what does this pattern unlock for your business? Because most of you watching this are

5:24not trying to automate job search. But I would bet a lot of you have workflows that look exactly like this, like cold outreach, right? And basically, you have not built an agent for them yet. So, let me think about it. You need to reach out to potential clients, we all do. You need to pitch your product to newsletters or directories. You need to apply to brand deals or podcast appearances. I do that. You need to follow up with people who have not responded to your first message. And you need to contact suppliers or partners or distributors. Every single one of those is a job application problem with different clothes on it, right? It is a list of targets, a standard action, and a logging system to track what happened. The category is outreach. The form in the form it takes is different for

6:09everyone. But the underlying structure is the same. And here's the thing. The bottleneck is almost never the technology. Open Claw can run cron jobs. It can hit APIs. It can fill out forms through browser automation. It can even send messages. It can log results to a database. All that's really easy. And all of that has been built in and documented. The bottleneck is basically you sitting down and actually defining the workflow clearly enough that the agent can actually execute it on a reliable basis. That is the real work, not the code at all, and not the setup. It is the thinking that happens before you open a terminal. You got to map out your workflow and then hand it off to the agent. So, I'm going to break down what that setup actually looks like for Nerik and like in terms of the components. Like what you would actually need if you were building something similar for your business. So, like the

6:56first piece is just basically a clean source of targets, like in Nerik's case, it was LinkedIn job postings filtered by whatever criteria he had set up. So, for you, it might be a spreadsheet of podcast hosts or like a database of brands in your niche or it could be anything, right? A list of potential clients from a directory. That's not the point. Whatever it is, you need that source to be structured so the agent can pull from it consistently. If your list is messy, your agent will be crazy and going everywhere. The second piece is the action itself, the specific thing the agent does for each item in the list. This needs to be clean and well-defined. The more standardized the action, the more reliable the execution. Nerik was going through LinkedIn's interface,

7:42which is why the captchas hit hard, and they have that in place for a reason. If you can work through an API instead of a browser, your throughput will be much higher and far more consistent because you are not fighting the platform's bot detection. So, when you are designing this for your use case, always ask yourself, is there an API I can use instead of the browser browser automation? The third piece to this puzzle is the logging layer, and this is the one that people skip. You need your agent to record what it did, when it did it, and what the result was. Because when you are running 1,200 actions a day, you can't manually review each one. The log is how you know what is working and what is not. And Nerik knew his 4% conversion rate because the system was tracking outcomes at every step. Without

8:27that tracking, you're just guessing about what is happening inside your pipeline. The fourth piece is the autonomy boundary, how far the agent goes before it checks in with you. If you're struggling to keep up with content, what I'm about to save you about 40 days worth of work. [music] I built something called Content Machine. It's 10 AI agents that run on the Open Claw orchestration, and they handle everything. Scripts, thumbnails, X [music] posts, blogs, outreach, clips, newsletters, all of it. So, I went from 1,000 subscribers to 4,000 subscribers on YouTube in 7 days using this exact system. Every single morning I wake up [music] and the content's already done. I spend maybe 15, 20 minutes reviewing

9:12and approving them, and I move on with my day. [music] It works for any niche, fitness, finance, real estate, marketing, whatever you are building, and it is 100% completely customizable to your [music] use case. So, you get the mission control dashboard, all of the cron jobs, everything I've built over the last 40 days helping me gain more and more people to subscribe and join the community. So, you plug in your own thing, and it molds it to you, it learns how you talk, and it writes so it doesn't sound like AI slop. $97 one time, it's not a subscription. I'll put the link down below, and you'll thank me later. Nerik ran his pretty far out letting it auto start if hired. That is pretty aggressive and not, you know, what I would recommend for most businesses, but what I think what works

9:58better is having the agent handle the volume work and then surface the wins free to review. It processes 1,200 things and then it pings you on Telegram when something positive comes back. You look at the response, you decide what to do next. The ball's in your court. The agent does the scale work basically and you do the judgment calls. I do that for my social media. You know, that is the design pattern I keep coming back to. It is the one where, you know, you get leverage without the risk of the machine doing something you did not want it to do. For anyone inside the community, you know, who is thinking about building their first agent, listen to this story. The story should be the thing that gets you moving. The barrier to entry here is just not high at all. You need to replicate Narek's exact setup. You do not need to hit 1,200 actions per day

10:44out of the gate. Start with a list of 100 targets. Write down in plain language in a spreadsheet exactly what the agent should be doing for each one. Build a small test and then run it. Something that hits 10 like 10 action items. Log the results, see what breaks, then clean up the edge cases and then scale it up from there. That is the whole game. Define the workflow, log everything, learn from those logs, and then scale what works. Narek ended up with a $428,000 pipeline because he was willing to run the experiment and let the numbers tell him what was actually happening. Most people are literally still just trying to decide whether they should start running things on Open Cloud. I know you watching the vi- this video are already far ahead. I'm just doing these use

11:29cases to kind of set an example for people that want to do cold outreach, want to build their sales pipeline. Take the components from what Narek did and apply them to your use case. The one thing that it was you're going to have problems with, and so did Narek, is the capture problem. You know, that cut his throughput from 8,000 to 1,200 per day, but still that's a lot of volume. But he didn't quit, too. Neither should you. He noted it as a constraint, he kept running, and just let the math do its thing. That's the mindset you should have. You're not looking for perfection on every run. You are looking for a positive aggregate result across the volume. And when you have an agent doing the work, you have the volume. You could see Narek's original post, I put a link down below. Go read it. Pay attention to how he frames the capture program because the way he turned a miss on his

12:17daily target into a data point about where real friction lives in the workflow, that thinking is worth starting on its own. And I personally would take that article and give it to your agent and say, "How can we apply what Narek did in his LinkedIn outreach to my sales, to podcast outreach, to customer service?" Start doing that. Take this video, link the the link to the video. Give it to your agent and start applying some of these concepts and principles cuz they're no different. And if you haven't joined the community yet, we have 218 people that have joined in the last 28 days. We go over these exact systems built out with AI agents. I'll put the link to it down below. We do nine live boot camp calls a week. We just added a local LLM call where you can install your first local model and

13:03get it running on your hardware today. So, I'll put the link down below. If you haven't subscribed to the channel already, please do cuz we do videos like this every single day. We'll see you in the next one. Be blessed.

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 Skool

Keep reading