Perplexity Computer: New Autonomous AI Agent Updates!

Perplexe Computer New Autonomous AI Agent Update. So 3.25 years of work done in four weeks, $1.6 million saved. That is not a startup pitching investors. That is Perplexe's own internal data from 16,000 queries tested against benchmarks from McKinsey, Harvard, MIT, and Boston Consulting Group.

And they just made the tool that did it available to everyone. Now, this week, Perplexe dropped four massive updates at once. Personal Computer, Computer for Enterprise, Mobile Access for Perplexe Computer, and they added NVIDIA's brand new Nemetron 3 supermodel to the whole system along with a bunch of other stuff. And all of it landed in the same wake.

All of it changes the game for anyone paying attention. So let me hit each one. First, Personal Computer. This is the headline update.

Perplexe took their cloud-based computer that launched in February and brought it to your actual physical machine. You take a Mac mini, you leave it running, and Perplexe's AI agents get full access to your local files, your apps, your browser sessions, everything on that device. It works for you 24 hours a day, seven days a week whilst you sleep and whilst you're on vacation, whilst you're literally doing anything else. The CEO Aravind Srinivas said it best, a traditional operating system takes instructions, an AI operating system takes objectives.

You do not tell it to click here, open this, type that. You tell it what you want done, and it figures out every step to get there on its own. Now, why does this matter? Because up until this week, the Perplexe computer only worked in the cloud.

It could browse the web, use online tools, generate code, do research, but it could not touch your local files. It could not open the spreadsheet sitting on your desktop. It couldn't access that folder of client documents you've been organizing for months. Personal computer changes that.

Cloud plus local, that is the unlock. And here's the part that matters for the open-claw comparison. Personal computer requires user approval for sensitive actions. Every session creates a full audit trail.

There's a kill switch you can hit to stop everything immediately. You can control it remotely from any device. And why am I emphasizing the safety features? Because what's been happening with open-claw.

Open-claw is the fastest growing open-source project in the history of GitHub. 200,000 stars in 90 days, pass React, pass Linux. Nothing is ever growing that fast. I mean, it does something similar.

It gives an AI agent access to your computer so it can work on your behalf, just like the Perplexe computer. However, there's a big difference here. And that's the problem. Research has found over 40,000 vulnerabilities in open-claw, 40,000.

There was a critical flaw called Clawjacked that let hackers take over your AI agent just by getting you to visit a malicious website. That flaw got patched, but then more showed up. Cisco's security team tested a third-party open-claw score and found it was stealing data and running prompt injection attacks without the user even knowing about it. So that's what we're talking about here.

Now, one of open-claw's own maintainers posted a warning on Discord saying, if you can't understand how to run a command line, this is far too dangerous of a project for you to use safely. And on March 10th as well, China's cybersecurity regulator issued an official government warning about open-claw's security risks. The Chinese government told state agencies and banks to stop using it on office computers. That is not a blog post from some random security researcher.

That is a government telling its biggest institutions to back away. Now, don't get me wrong. Open-claw is incredible technology. It's open-source, it's free, it's wildly powerful.

Peter Steinberg has built something historic there. The creator got hired by OpenAI because of it. The project is moving to an open-source foundation. The community around it is massive and passionate and building amazing things.

But perplexity looked at everything open-claw built and said, okay, we're gonna take that same idea and we're gonna wrap it in enterprise-grade security, sandbox environments, audit logs, SOC 2, type 2, compliance, permission controls, a kill switch, and we're gonna run it on 20 different AR models instead of whichever single model the user configures. And when asked about the difference between perplexity personal computer and open-claw at the developer conference, Srinivas said perplexity computer is meant for serious people, which is a big statement in itself. Now, that is a bold statement. Some people are gonna disagree with it, but when you look at the security track of both products right now, you can see where it's coming from, right?

That's what it's about. And here's the second major update. Perplexity computer is now available on mobile. This might sound small, but think about what it means.

Before this week, you'd only access the computer from a desktop browser. Now you can control your AI agent from your phone. You are standing in line at the grocery store and you can tell your AI to go research, build a report, draft emails, and create a website from your phone. And the AI goes and does it in your cloud or your Mac mini back at home.

That's wild. That is a full digital worker you can manage from your pocket. The third update, computer for enterprise. This is perplexity taking the system and plugging it directly into business tools.

Snowflake, Salesforce, HubSpot, hundreds of other platforms through app connectors. A financial analyst can ask the system to pull revenue by vertical from Snowflake while simultaneously pulling competitive contacts from the web. A sales team can extract CRM data and market research at the same time. That writes the queries, runs them, and delivers structured results.

And this is where that 3.25 years number comes from. Perplexity testing perplexity computer for enterprise internally. It's 16,000 queries measured against institutional benchmarks from McKinsey, Harvard, MIT, and BCG. The result, 3.25 years of work completed in just four weeks.

$1.6 million in labor costs saved. Now, I wanna be honest here, but Black Sea has not released a full methodology behind that claim. We do not know exactly how they calculated years of equivalent work. The register pointed that out and they are right to question it.

Anytime a company gives you a number that dramatic, you should ask how they got it. But even if you cut that number in half, even if you say, okay, maybe it was only a year and a half of work in four weeks instead of 3.25 years, it's still insane. That is a level of productivity that would have been science fiction two years ago. And enterprise teams can now teach the system custom workflows.

They can collaborate with it directly in Slack through DMs and shared channels. So the AI is not sitting in some separate app you have to go visit. It's right there in the tools your team already uses every single day. The fourth update, NVIDIA's Nemetron 3 Super is now available inside the Black Sea computer.

This one flew under the radar for a lot of people, but it is a big deal. Nemetron 3 Super just launched on March 11th. It's a 120 billion parameter model, open source from NVIDIA with only 12 billion parameters active during any task. That means it's incredibly efficient.

It delivers five times higher throughput than the previous Nemetron model. And it has a 1 million token context window. And that means the model can hold an enormous amount of information in its memory at once without losing track of what it's doing. Why does that matter for agents?

Because AI agents need to remember what they're working on across long, multiple step tasks. If you tell an AI agent to research topic, write a fact, fact check it, format it, and then email it to your team, it's a lot of steps, it's a lot of explaining. Most models start forgetting what they were doing partway through. They drift, they lose a thread.

Nemetron 3 Super's 1 million token context window means it can hold the entire workflow in memory from start to finish. And NVIDIA made it completely open. It's open training data, 10 trillion tokens of training data published. The full recipe for how they built it so that any developer can download it, customize it, and deploy it to their own infrastructure.

That's a massive move from the company that makes the chips powering almost every AI system on the planet. The timing is not an accident either. NVIDIA's GCC conference starts Monday, March 16th, next week. Jensen Huang is about to take the stage in San Jose in front of 30,000 people.

They dropped Nemetron 3 Super right before the conference as a preview. And there are rumors NVIDIA is about to announce something called NemoClaw, an open source platform that would let businesses build and deploy AI agents across their entire organization using NVIDIA's models and infrastructure. If that happens, you're going to have Flexi Computer, OpenClaw, and NemoClaw all compete for the same space, along with many other Chinese AI technologies. Even Manus are getting in on this.

So AI agent wars are just getting started. Now, let me zoom out for a second because I think a lot of people are gonna hear this and think, okay, cool tech stuff. But what does it actually mean for me? Here's what it means.

Right now in China, OpenClaw has become a national obsession. Tencent just launched a full suite of OpenClaw products that work with WeChat. Alibaba launched a mobile app called JVS Claw that helps people with zero coding knowledge deploy AI agents in minutes. Baidu released its own OpenClaw app this week.

ByteDance built a version called ArcClaw that runs in a web browser so you don't even need to install anything. JD.com is selling a service where you pay $58. And Lenovo technicians will remotely set up OpenClaw on your computer for you. Events about OpenClaw are drawing over a thousand people in Shenzhen.

A 27-year-old software engineer in Beijing set up a side business on a second-hand shopping app, offering OpenClaw installation support, and is making serious money from it. MIT Technology Review called it a gold rush. And at the same time, Chinese companies are forcing their employees to learn AI. Bloomberg reported that project manager at one of China's largest finance companies said her managers demanded employees prove they could use OpenClaw in a competitive contest during the Lunar New Year holiday.

She said, since OpenClaw came along, things have gotten insanely competitive. Managers warned staff would be replaced if they did not use. That's happening right now. Not in five years, not in some theoretical future, right now in one of the largest economies on earth.

And here's the thing, whether it's Perplexe Computer or OpenClaw or whatever Nvidia comes up with, and whatever they're about to announce, the underlying trend is the same. AI is moving from something you chat to to something that works for you, from a tool you use to an employee you manage, from answering questions to completing projects. The chatbot era is ending and the AI agent era is starting. Now, I'm gonna be real with you for a second because I know some people watching this are thinking, I do not even know where to start.

I hear about all these tools, my head is spinning, I feel like I'm already behind. I get it, I really do. The speed of change right now is genuinely overwhelming. A new frontier AI model was launching every 17 days throughout 2025.

Perplexe went from a search engine to a full AI operating system in less than a year. OpenClaw went from one guy's side project to the most starred project on GitHub history in 90 days. That pace is kind of hard to keep up with, so I get it. And here's what I want you to understand.

You don't need to master every tool, you don't need to understand every technical detail, you do need to understand the direction. And the direction is crystal clear here. The entire agents that can work independently on your behalf are here. They're getting safer, they're getting easier to use, they're getting more powerful every single week, and the people who learn how to use them are going to have an enormous advantage over the people who wait.

This is exactly why we built the AI Profit Boardroom. Available link in the comments description or the AIProfitBoarding.com. It's a community where we break down exactly how to use these tools to actually make money and save time. Not theory, not hype, practical step-by-step walkthroughs of the tools that are working right now.

If you want in, the link is in the description. The people inside are already using AI agents in their businesses and their careers and they are seeing real results. But I'm not going to keep pitching you. You either see where this is going or you don't.

Let me get back to the updates. So here's what's fascinating about the perplexity versus open core dynamic. They're solving the same problem in completely different ways. So the way that you think about that difference tells you a lot about where AI is heading.

Open core is open source, right? Free, you can see every line of code. You can customize it however you want. You can add skills and plugins from the community.

You own your data because everything runs locally on your machine. The community is massive and passionate and building incredible things on top of it. And the openness is its greatest strength. But that openness also comes with a vulnerability because when anyone can build a plugin, bad actors can build malicious plugins.

When the software runs with full access to your computer, a security flaw means an attacker gets full access to your computer. Cisco found a third-party skill that was stealing data without the user knowing. 40,000 plus vulnerabilities were discovered. The Chinese government banned it from state institutions.

The Black Sea Personal Computer is the opposite. It's closed source. It costs $200 a month for max subscribers. You can actually get it on the pro plan now as well for even cheaper.

You can't see the code. You can't customize it the same way, but it comes with enterprise-grade security out of the box. Sandbox environments, audit trails, permission controls, a kill switch. And it runs on 20 AI models instead of whichever one model you happen to configure.

Think of it like this. OpenCore is like building your own house from scratch. You can make it exactly how you want, but you're responsible for the plumbing, the electrical, the foundation, the security system, everything. If something breaks, it's on you.

But Black Sea Personal Computer is like moving into a luxury apartment building. You do not get to knock down walls, but the security system is already installed, the plumbing works, the electricity is on, and someone else is responsible for maintaining it all. Neither approach is right or wrong. They serve different people with different needs and different risk tolerances.

But here's what I think is going to happen. And it's just my prediction based on watching this space every single day. I think the perplexity approach is gonna win for most people. Not because it's better technology, but because most people do not want to manage their own security.

Most people do not want to configure AI models and API keys. Most people do not want to debug command line errors in Terminal. And most people just want to describe what they need done and have it handled, right? The hardcore developers and tinkerers will stick with OpenCore, and they should.

It's an incredible tool for people who know what they're doing. But the mass market, the small business owners, the freelancers and marketing managers, they're gonna want something that just works straight away. And that's what perplexity is building. Now, let me talk about what all of this means in the bigger picture because something happened in the same week that connects directly to everything I've talked about.

Atlassian, that makes Jira and Confluence and Trello, the tool that millions of teams use every single day to manage their work, fired 1,600 people, 10% of the workforce, gone in one announcement. And their CEO, Mike Cannon-Brooks, wrote a memo to staff and said, our approach is not AI replaces people. It would be disingenuous to pretend AI does not change the mix of skills we need or the number of roles required in certain areas. Read that again, or think about that again.

AI does not replace people, but it changes the number of roles required. What do you think that means? It means fewer people doing the same amount of work. That's a layoff.

They are spending $250 million on severance and restructuring. They replaced a CTO with two new AI-focused CTOs. And Atlassian is not alone. Blockfired 4,000 people last month.

CEO Jack Dorsey said they were shifting to an intelligence-native operating model. WiseTech Global announced 2,000 cuts. Their CEO said the era of manually writing code is over. And tech layoffs in 2026 have already passed 45,000 globally by early March.

Now, here's the connection. When Publixi says that AI completed 3.25 years of work in four weeks, that's not just like a cool demo. That is the math that CEOs use when they decide how many people they need. That is the math that is driving those layoff decisions.

Investors are now looking at a metric called revenue per employee, AKA how much money does each person at your company generate? Nvidia is right at the top of that list. And the companies figuring out how to use AI to increase that number are the companies Wall Street is rewarding. So here's the reality.

AI agents are gonna replace everyone tomorrow. That's not what's happening here. What's happening is that companies are realizing one person with AI agents can do the work of five people without them. And they are restructuring accordingly.

So the question is not whether this is happening. It is happening. Which side of that equation do you wanna be on? Do you wanna be the person who gets replaced because they never learn to use these tools?

Or do you wanna be the person who uses these tools to become so productive that you are literally irreplaceable? Because that's a real choice right now. So let me bring it back to the specific tools. Here's what you should actually do with everything that I've talked about today.

If you're a business owner or a freelancer and you can afford the pro subscription or the max subscription of Publixi, get on the Publixi max wait list for personal computer. That's a no brainer for anyone who does knowledge work. If those are part of your job and if you're doing a lot of knowledge work yourself and AI that doesn't fool you around the clock is worth way more than $200 a month. If you're technical and you like to tinker and customize, OpenClaw is still the most powerful open source agent out there.

Just make sure that you keep it updated. That Clawjacked vulnerability was patched, but new issues pop up all the time. So don't give it access to sensitive accounts until you understand the security implications. If you're running a team or a company, pay very close attention to what Publixi announced with Computer for Enterprise.

The ability to connect AI agents to Snowflake, Salesforce, HubSpot and hundreds of other business tools through Slack is exactly what enterprise teams have been waiting for. This isn't like a chatbot. You've got to visit this and AI team member that works inside your existing workflows. And everyone, regardless of your technical level should be watching what happens at NVIDIA GTC this Monday.

Jensen Huang is about to take the stage at the SAP Center in San Jose in front of 30,000 people. There are strong rumors about Nemiclaw an enterprise AI agent platform. There will be announcements about new chips designed specifically for running AI agents faster and cheaper. And this could be one of the most consequential tech keynotes of the entire year.

Now, let me address the skeptics directly because I know some of you are thinking, Julian, every week there's some new AI tool that's supposed to change everything. And six months later, nobody's using it. And that's a fair point, I hear you. And you're right that there's a lot of hype in AI, right?

A lot of announcements that don't lead anywhere. But here's what's different about agent AI. This is not a new chatbot. This is not a better way to generate images.

This is a fundamental shift in how computers work. For 50 years, we've told computers what to do step by step. Click here, type this, open that. For the first time ever, we can tell a computer what we want accomplished and it figures out how to do it.

That's not hype, that's an entirely new computing paradigm. The numbers back it up. OpenClaw became the most starred project in GitHub history in 90 days. Not because of marketing, but because people used it and their jaws literally dropped.

Perplexity saved $1.6 million in labor costs in four weeks, not in theory, in their own company with their own employees. So the question is not whether AI agents work. The question is how fast does it scale to the rest of us? And based on what dropped this week, I would say pretty fast.

Now, let me leave you with this because I've been covering AI every single day for a long time now. And I keep coming back to the same core truth. The gap between people who use AI and people who do not is getting wider every single day. Not every month, every day.

A year ago, the gap was about who could write better prompts to chat cheaper tea. Six months ago was about who could use AI coding tools to build apps. Today, the gap is about who has AI agents working for them around the clock. It's a new playing field.

You're either on it or you're watching from the stands and falling behind. And right now, today, this week, you have more free and affordable tools than at any point in human history to start using AI. OpenClaw is free. Plexi have a free tier.

Chat Cheaper Tea is a free tier. Claude has a free tier. The excuses are gone. The barriers are gone.

The only thing standing between you and the future is whether you decide to start. And the people who started six months ago, they're already running circles around the people who are still thinking about it. I'm not trying to scare you. I'm trying to wake you up because the opportunity right now is enormous.

The ability to have AI agents work for you 24 hours a day, doing research, writing, analysis, coding, design, whatever your work requires, that used to be something only billion dollar companies could afford. Now it's available to you. If you have a laptop and a Wi-Fi connection or a phone. And that's not the future.

That is literally this week. So if you got value from this, make sure you subscribe, hit that notification bell. If you're serious about learning how to actually make money and save time with AI and you want a community of people who are all figuring this out together, check out the AI Profit Boarding. Link in the comments description or go to the AIProfitBoarding.com.

Thanks for watching.