Follow : Add us as a preferred source  on Google.

Key takeaways

• Perplexity debuted Computer on Wednesday morning.
• It's positioned essentially as a safer alternative to OpenClaw.
• It harnesses over a dozen leading AI models.

There's been a lot of excitement (and nervousness) lately about AI agents that can work autonomously in the background of a user's computer, accessing sensitive files, API keys and the like to perform various tasks. Some say they're a monumental productivity unlock, others say they're a security nightmare. Perplexity is betting they're the future of AI.

Also: 7 AI coding techniques that quietly make you elite

On Wednesday, the company introduced Computer, a multiagent orchestration system that harnesses the strongest capabilities from more than a dozen frontier AI models. Currently available only to Perplexity Max users -- and expected to roll out to Enterprise and Pro subscribers in the coming weeks -- "Computer is a general-purpose digital worker," the company wrote in a press release, that "reasons, delegates, searches, builds, remembers, codes, and delivers."

Multiagent orchestration

The logic behind Computer is basically that, rather than becoming general-purpose tools, as they're commonly described, AI models have instead branched off into different specialties: Anthropic's Claude, for example, is famously popular among software engineers. Relying on a single model to complete a complex task -- building a website, say -- is therefore a bit like trying to assemble an Ikea dining table using a butter knife; it could be possible, but the finished product is going to be a little wonky. Wouldn't you rather have a multi-bit screwdriver?

Also: From Clawdbot to OpenClaw: This viral AI agent is evolving fast - and it's nightmare fuel for security pros

To use another analogy, think of Computer like the CEO of a company, delegating tasks across a hierarchy of teams and employees. A user can describe their vision for a final outcome ("Build an app that provides up-to-date snow conditions at different ski resorts"), and Computer will automatically break the task down into different tasks and subtasks, according to Perplexity, all of which will be handled by whatever model is called for. 

Its "core reasoning engine" is Claude Opus 4.6. Google's Nano Banana and Veo 3.1 handle imagery and video, respectively, while Grok handles "lightweight tasks" and GPT-5.2 is deployed for queries that require long-context recall and an expansive web search. 

The current model arrangement within Computer is subject to change, according to Perplexity: new models could be added if they excel in specific domains, and the existing lineup could shift as the models evolve. Users also have the option of stepping into the orchestrator role and delegating specific subtasks to particular models. And users can execute dozens of tasks in parallel to one another; Computer can operate quietly in the background for months, according to Perplexity, checking in only "if it truly needs you."

Safety considerations

If you're reading this and thinking, "This sounds a lot like OpenClaw," you're not wrong.

The AI agent formerly known as Clawdbot and Moltbot went viral earlier this month as a kind of always-on automated assistant that could essentially work across users' entire digital ecosystem, and interact with them directly via apps like WhatsApp, Slack, and Telegram. Its creator, an Austrian programmer named Peter Steinberger, was promptly hired by OpenAI: In a X post, company CEO Sam Altman called him "a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people," and that "this will quickly become core to our product offerings."

(Disclosure: Ziff Davis, 's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

But the field of fully autonomous agents that can work across apps and files is a very young one, and mistakes happen. Earlier this week, Meta AI security researcher Summer Yue posted screenshots on X of her desperate attempts to instruct OpenClaw to refrain from deleting her entire email inbox, which it was ignoring. "I had to RUN to my Mac Mini" -- the hardware of choice for running OpenClaw in the background -- "like I was diffusing a bomb," she wrote. 

(Yue wrote in a comment beneath that post that OpenClaw had gained her trust after successfully managing her "toy" inbox, but that when she moved it to her much larger, actual inbox, it triggered a process called compaction, in which an agent is faced with an excessively large context window and starts taking shortcuts -- in this case, overlooking her original instruction not to "action until I tell you to.")

Also: OpenClaw is a security nightmare - 5 red flags you shouldn't ignore (before it's too late)

Yue's episode highlights two very real risks: Prompts can be misinterpreted by agents, and they can act in unexpected (sometimes disastrous) ways.

Perplexity appears to be selling Computer as a safer, more controllable multiagent orchestration system than those that are currently available. The system runs in "a safe and secure development sandbox," according to the company, which means that any security glitches can't spread to a user's main network. The company also said it's "run thousands of tasks" internally using Computer, from publishing web copy to building apps, and "been consistently surprised by the quality of the output."

Featured

Is Perplexity's new Computer a safer version of OpenClaw? How it works

Samsung Galaxy S26 vs. S24: After comparing both models, here's my take

7 AI coding techniques that quietly make you elite

21 hidden Netflix codes, tricks, and settings every power user needs to know

• Is Perplexity's new Computer a safer version of OpenClaw? How it works
• Samsung Galaxy S26 vs. S24: After comparing both models, here's my take
• 7 AI coding techniques that quietly make you elite
• 21 hidden Netflix codes, tricks, and settings every power user needs to know