Perplexity Launched an AI Agent Platform Called
Perplexity unveiled Computer, a general-purpose AI agent platform that delegates tasks across specialized sub-agents. It searches, codes, builds, and remembers context across sessions.
Perplexity has been the quiet overachiever of the AI search world. While OpenAI and Google grabbed headlines, Perplexity kept shipping. Now it's swinging at something bigger.
The company launched "Computer" this week, and despite the generic name, the ambition is anything but. Computer is a full AI agent platform that goes well beyond search. It reasons, delegates work to specialized sub-agents, searches the web, writes code, builds things, and remembers context across sessions. Perplexity is calling it a "general-purpose digital worker."
That's a bold claim. Let me tell you what it actually does and whether the pitch matches the product.
What Computer Actually Does
At its core, Computer works like a dispatcher. You give it a task, and it breaks that task into pieces. Each piece gets routed to a specialized sub-agent. One handles web research. Another writes code. A third manages document creation. A fourth handles data analysis. They all report back to a central orchestration layer that stitches everything together.
Say you need a competitive analysis of the AI chip market. Computer would spin up a research agent to pull recent filings and news, a data agent to compile financial metrics, and a writing agent to produce the final report. The agents pass context between themselves, so the writing agent knows what the research agent found without you having to copy and paste anything.
It's not entirely new territory. OpenAI's been building toward this with ChatGPT's tool use. Anthropic's Claude Cowork does something similar for office tasks. But Perplexity's approach feels more modular. Instead of one model trying to do everything, you've got specialized agents that are good at specific things.
The Memory Piece
The feature that caught my attention is persistent memory. Computer remembers things across sessions. Not just your conversation history, but the actual context of what it learned while doing work for you.
If Computer built you a market analysis last Tuesday, it knows that. When you ask for an update on Friday, it starts from what it already found instead of starting from scratch. That sounds obvious, but most AI tools today treat every conversation like a first date. No memory, no context, no awareness that you've been here before.
The catch is that memory is only as good as the retrieval system behind it. If Computer remembers 500 things about your work but can't surface the right one at the right time, it's just a cluttered filing cabinet. Perplexity hasn't published details on how the retrieval works, which is either because it's proprietary or because it's still rough.
Where This Fits in the Market
Perplexity is positioning Computer somewhere between a search engine and an operating system for knowledge work. That's a crowded space. Microsoft has Copilot embedded in every Office product. Google has Gemini doing the same in Workspace. Anthropic is expanding Cowork into enterprise workflows.
But Perplexity has something the big players don't: trust from users who are sick of ads and tracking. Perplexity built its reputation on giving straight answers without SEO spam. If Computer maintains that philosophy, there's an audience for it.
The startup has also been growing fast. It reportedly hit $100 million in annual recurring revenue faster than almost any SaaS company in history. That gives it runway and credibility, though it's still tiny compared to the cloud giants.
The Agent Wars Are Real Now
Computer's launch is another signal that the AI industry has moved past chatbots. Every major company is now building AI agents that can do multi-step work. OpenAI has operator agents. Anthropic has Cowork and Claude Code. Google has Gemini agents across its products. And now Perplexity is playing too.
The question isn't whether AI agents will become standard software tools. They will. The question is which approach wins. Do you want a single model that tries to be good at everything? Or do you want a network of specialized agents coordinated by an orchestration layer?
Perplexity's bet is on specialization. Build agents that are great at specific things and connect them with smart routing. It's the microservices architecture philosophy applied to AI. That approach has advantages. Specialized agents can be optimized independently. If your code agent is weak, you upgrade just that agent without touching the rest.
The downside is complexity. More agents means more points of failure. Context passing between agents can lose information. And the orchestration layer has to be smart enough to know which agent to call and when. Get that wrong and you've just built a fancy way to produce wrong answers from multiple directions.
What I'm Watching
Three things will determine whether Computer matters.
First, reliability. Agents are only useful if they complete tasks correctly. One wrong data point in a research report and the whole thing is garbage. Perplexity needs to nail accuracy before it nails anything else.
Second, pricing. Agent platforms burn compute at a much higher rate than simple chatbots because you're running multiple models simultaneously. If Computer costs $200 a month, only power users will bother. If it's included in the existing subscription, Perplexity's burn rate becomes the story.
Third, integrations. Computer needs to connect with the tools people already use. Email, Slack, spreadsheets, databases. An agent that can only work inside Perplexity's own interface is a demo, not a product.
Perplexity has earned the right to be taken seriously. But claiming to build a "general-purpose digital worker" is about as ambitious as it gets in AI right now. The good news is they're shipping actual software instead of publishing research papers about what they might build someday. In an industry drowning in vaporware, that counts for something.
Get AI news in your inbox
Daily digest of what matters in AI.
Key Terms Explained
A mechanism that lets neural networks focus on the most relevant parts of their input when producing output.
A measurement of how well a language model predicts text.
An autonomous AI system that can perceive its environment, make decisions, and take actions to achieve goals.
An AI safety company founded in 2021 by former OpenAI researchers, including Dario and Daniela Amodei.