Perplexity Computer: New Autonomous AI Agent Updates!

Perplexe Computer, new autonomous AI agent updates with Comet Enterprise. So Perplexe just dropped something that most people in the business world have not even heard of yet. I want to be clear about what this actually is. This is not another AI chatbot update.

This is not a feature inside some app you already use. This is a company that built an AI that sits inside your business, runs on your computer 24 hours a day, connects to every tool your team already uses and starts doing the work. Perplexe actually ran an internal study across more than 16,000 queries. And the result was Computer for Enterprise saved internal teams $1.6 million in labor costs and the equivalent of 3.2 years of work in four weeks.

That number stopped me cold because that's not a product feature. That's a replacement event. And if you're running a business right now, if you have a team, if you're a freelancer, if you're trying to compete in any market involving research, data writing, analysis, customer support, financial sales, this changes the maths on everything. And here's where most people's experience with AI sits right now.

You type something into a chatbot and you get something back, right? Most people have used chatGBT, little more than that, right? That is a workflow most of the world is still running on in 2026. What Perplexe just shipped is categorically different.

We're not talking about a faster chatbot. We're talking about an AI that does not wait for you to ask it questions. An AI that connects your actual tools, your files and keeps working whilst you sleep, whilst you're in meetings, whilst you're on holiday. Perplexe CEO Aravind Srinivas said in the Ask 2026 conference, a traditional operating system takes instructions.

An AI operating system takes objectives. That's the whole shift in one sentence. You stop telling the computer exactly what to do step by step. You tell it what you want it to achieve.

It figures out the rest. Perplexe started as an AI search engine, a thing that gives you answers instead of links, and it grew fast. The company is three years old, valued at $20 billion and has raised about $1.5 billion in total funding. They had a real problem.

They do not build their own AI models. They are not open AI or anthropic. So every time a customer thought about paying for Perplexe, they had a reasonable question. Why not just go directly to open AI?

Why pay the middleman? And the answer to that question is what they just launched. In January 2025, 90% of enterprise queries on Perplexe roosted to just two AI models. By December 2025, no single model commanded more than 25% of usage.

The companies using Perplexe stopped relying on one AI and started using all of them. And Perplexe became the thing that figures out which one to use for each task. Not one model, the orchestration layer on top of all of them. Perplexe computer is what they call a general purpose digital worker.

You can start by describing an outcome. It breaks it into tasks, subtasks, creates sub-agents for execution. Those sub-agents do web research, document generation, data processing or API calls to connected servers. Right?

And so you're not talking to one AI, you're talking to a manager AI that spins up specialist AIs to handle each part of the job. One agent drafts the documents whilst another gathers the data that goes into it. And it can orchestrate work across 20 different models in parallel, matching each task to the best model for the job. The consumer version launched a few weeks ago.

And what happened next was fast. So users demonstrated the agent building Bloomberg terminal style financial dashboards, replacing six figure marketing tool stacks in a single weekend and automating workflows that previously required dedicated teams. More than 100 enterprise customers messaged Perplexe over a single weekend demanding access. 100 enterprise buyers in one weekend on a Saturday, flooding the inbox, demanding access.

So when buyers who normally take 18 months to make a software decision start doing that, something real is happening. So Perplexe held their first developer conference called Ask 2026 inside a former church in San Francisco. And they announced computer for enterprise. Comet Enterprise gives organizations an AI equipped browser with admin controls over where and how the assistant operates.

Admins can restrict comment to answering questions without taking actions, allow more active behavior in lower risk environments and review action logs for every session. And then they brought in CrowdStrike, one of the most trusted names in enterprise security. CrowdStrike and Perplexe announced a partnership to integrate the CrowdStrike Falcon platform with Comet Enterprise, providing real time detection, governance and data protection directly into AI driven workflows. Daniel Bernard, CrowdStrike's chief business officer, put it plainly, the browser is no longer just where people access information, it's becoming the interaction layer for enterprise AI.

The browser used to be where you looked up stuff, right? Now it's where all the AI work happens. If your organization is not thinking about that transition, guess what? You're already behind.

Here's what computer for enterprise actually looks like in practice. The enterprise version connects natively to Snowflake, Salesforce, HubSpot and hundreds of other platforms. Teams can query data warehouses, pull CRM context and build financial models without waiting on a data or analytics team. Enterprise teams can interact with computer directly inside Slack.

You type a message to the AI in Slack, it pulls from multiple systems, runs it and the system comes back with a finished output. Example prompts from Perplexe include triage, weekend support ticks by severity, draft customer responses, write escalation briefs and package everything into a Monday stand-up doc or automate due diligence by fact checking pitch claims against live data, flagging inconsistencies and generating annotated reports. Triaging a weekend's worth of support tickets, drafting responses, writing briefies, packaging it all into a stand-up document that is half a person's Monday gone before the starts before. But now the AI does it before you get to the office.

The enterprise version also adds SOC 2 type 2 compliance, SAML, single sign-on, audit logs and isolated sandboxing for each query. SOC 2 type 2 is the gold standard for enterprise security. Perplexe built it in from day one. The other announcement from AST 2026 is something I want to spend time on because it's genuinely different to anything else that exists.

They call it personal computer. Personal computer is not hardware. Perplexe team manufacturers, it is software. It runs continuously on a user provided Mac mini, giving the AI direct access to local files, apps and sessions.

Now, this depends. If you're using Perplexe personal computer, then it runs on a Mac mini, right? And that Mac mini is always on. If you're using Perplexe computer directly, then it's running in the cloud already.

So with personal computer, you leave a Mac mini on, always on. Perplexe's AI has a permanent home on that machine. It can see files, work inside your apps, monitor triggers and take actions without you sitting there watching. You can check in from your phone anywhere in the world.

Instead of typing precise commands, users describe broad goals. Things like an interactive educational guide or create a podcast about Wales. And the system figures out which local apps to open, which files to use and how to get it done. Personal computer is available to Perplexe's Mac subscribers at $200 per month, including 10,000 monthly compute credits.

I want to talk about that number. $200 a month sounds like real money until you ask the right question. What does a junior researcher cost per month? A data analyst, a virtual assistant.

We're talking $5,000 to $10,000 a month in human labor for the kind of tasks personal computer automates. And the AI does not call it sick, doesn't need managing, doesn't need to leave for a better offer. Right? So for a solo founder, a small team, a freelancer trying to punch above their weight, $200 a month for a tool that works 24 seven and connects to all your apps isn't expensive.

It might be the best $200 in your business. Now, let me steel man the skeptic. I know what some of you are thinking. AI agents make mistakes.

They hallucinate. They go in circles. I tried one of these things and it just broke and made everything worse. How is this different?

And that's a fair concern if you're saying that. The first wave of agent tools had a real reliability problem. But here's what changed. The models powering these AI agents have crossed the reliability threshold in the last few months.

Cloud Sonic 4.6 and Opus 4.6, which now power Comet, provide significant gains over predecessors, especially for agentic tasks. That phrase, especially for agentic tasks is a key one. And model builders know that agents are the use case. They have been optimizing specifically for it.

And Publixy adds another layer on top by routing tasks to whichever model handles them the best. The whole system is more reliable than any single model alone. Is it perfect? No.

Is it dramatically better than anything from 18 months ago? Absolutely. But here's the bigger point. The debate about whether AI agents are perfect misses the actual question.

The question is whether the people who figure out how to deploy these tools effectively gain a compounding advantage over the ones who wait until they feel comfortable. And the data is not subtle. According to a February survey by Crew AI, 100% of surveyed enterprises plan to expand agentic AI use systems. 65% already using AI agents in production.

Organizations report automating an average of 31% of their workflows. Gartner forecasts that 40% of enterprise applications will feature task-specific AI agents by 2026, up from less than 5% in 2025. And 40% in one year. It's not gradual.

That is a category going vertical. Here's what this actually means for people running businesses without Fortune 500 budgets. The real unlocks from personal computer are business-shaped. Teams drowning in repetitive workflows.

Founders who can't yet afford to hire. Operations running on duct tape and spreadsheets. If you have five people who each get back half a day through AI automation, you suddenly have the output of seven and a half people without hiring anyone new. The solo founder who understands agent tools effectively competes with teams of 10.

The small agency that has built AI automation into its workflow competes with the big agency that hasn't. The freelancer who delivers faster, higher quality commands, better rates. This is not a future scenario. It's happening in real businesses with real clients producing real revenue right now.

And this is exactly where I want to talk to you directly. If you're sitting here thinking, I get it. Agents are becoming a thing, but I don't know which tools to use. I don't know how to connect any of this to my business.

That's exactly what the AIProfitboarding.com. It's a community of business owners, freelancers and creators and agencies and marketers who are actually implementing this stuff. Not theorizing, not posting takes, actually building workflows, actually automating, actually seeing it in their numbers. We break down tools like Plexi computer, show you how to connect them in your workflow and find the shortcuts to save you hours every single week.

In plain English, that doesn't require a computer science degree. You want to stop watching AI happen and start being someone who works for. The link's in the comments description or come join us at the AIProfitboarding.com. Now, let me close with the thing that ties this all together.

We're watching a moment where the company's winning with AI and not necessarily the ones with the best models. They are the ones who figure out how to put multiple models to work on real tasks inside real workflows. OpenAI built GPT, Anthropic built Claude, Google built Gemini. These are the most powerful models in the world.

And Perplexi, a three year old company with no model of its own, just built a product on top of all of them. Enterprises are floating their inbox to access. The insight is not build the best model. The insight is build the best system for deploying models against real work.

That applies to you too. Your competitive advantage is not going to come from building your own AI model. It's going to come from being better than your competitors and putting AI to work inside your specific domain. The game being played right now is not human versus AI.

It is human plus AI versus human without AI. And that gap widens every month. The people who thrive through this period are the ones who stay curious, stay in motion, and commit to learning the tools rather than hoping the tools go away. Perplexi just proved that again in a former church in San Francisco with an AI that never clocks out.

See you in the next one. Thanks so much for watching. And if you haven't already, check out the AI Profit Boardroom. There's a link in the comments description or go to the AIprofitboardroom.com to get access.