A few weeks ago, AI watchers began to notice something odd: Perplexity CEO Aravind Srinivas, once among the most social-media-savvy executives in the AI world, had gone unusually quiet. The silence stood out at a moment when agent-style tools like Claude Code, Codex, and the viral open-source OpenClaw were dominating the conversation. Perplexity—long positioned as an AI-powered “answer engine” and a Google Search challenger—seemed conspicuously on the sidelines. Some even began to wonder whether the company had lost its way.

But Perplexity wasn’t lost, according to Srinivas—it was just busy building. I spoke with him yesterday, shortly after the company launched Computer, its attempt to turn today’s powerful but intimidating agent tools into something closer to a shared digital workspace that non-experts can actually use. The product is currently available only to Perplexity Max subscribers, with a broader rollout to Pro and Enterprise users planned in the coming weeks.

To my ear, Computer sounds like an OpenClaw for everyone else. Tools like OpenClaw often run on a separate machine, such as a Mac mini, with deep access to files and settings. Perplexity’s approach keeps that work in the cloud instead—letting users hand off tasks like research, writing, or coding to a tool that works for hours or even months, without giving an AI full control over any personal device. 

The defining feature of Computer is also that it isn’t tied to one AI model. Different parts of a task can be routed to whichever model does them best—it currently orchestrates 19 models on the backend, including Claude Opus 4.6 for orchestration and coding tasks, Google Gemini for deep research, Google Nano Banana for images, Google Veo 3.1 for video, xAI’s Grok for speed in lightweight tasks, and ChatGPT 5.2 for long-context recall and wide search.

“When you build a team, you don’t build a homogenous group where everyone has the same skills,” Srinivas told me. “You build a team with diverse strengths. We’re applying that same logic to AI workflows. The orchestration is the product. The model is a tool.”

That model-agnostic stance is not new for Perplexity. Srinivas said more than half of the company’s enterprise users already select multiple models within a single workday. But with Computer, that philosophy becomes the core strategy—and a source of leverage. Srinivas said he isn’t worried about OpenAI, Anthropic, or any other model provider limiting access. “In fact, I got congratulations messages from Anthropic and Google,” he said, adding that model makers benefit when their systems are part of broader workflows. If access changes, he said, Perplexity will adapt. “The model layer is the most competitive it’s ever been.”

Srinivas also drew a sharp contrast with tools like OpenClaw, which typically run on a local machine with broad access to files, passwords, and settings—an approach he compared to malware because of how easily it can damage data or expose sensitive information. Instead, Perplexity’s system runs remotely in the cloud, inside a locked-down environment, and carries out tasks in the background, more like assigning work to a coworker on Slack than watching an AI take over your screen. That creates a system that is safer and more dependable, he emphasized. 

While he said more ambitious goals are coming for Computer, for now the goal is accessibility. Srinivas insisted that with Computer, “Even your mom can text on the app and delegate tasks,” whereas with OpenClaw it “took our own engineers a long time to set up,” he said, ticking through terminals, API keys, and permissions. 

Perplexity, by contrast, wants to make agent-style work feel more like using a Macintosh or an iPhone than configuring a server. Internally, he said, the company is already using Computer to debug code, analyze metrics, and generate marketing assets—often directly from Slack or a phone. “It finally feels like I have a swarm of agents working for me,” he said. “I know that’s a buzzword that everyone uses, but this is the first moment I’ve actually felt it.”

With that, here’s more AI news.

Sharon Goldman
sharon.goldman@fortune.com
@sharongoldman

FORTUNE ON AI

AI capex and the ‘wealth effect’ from tech stocks (like Nvidia) now drive one-third of U.S. GDP growth, top analysts say – by Jim Edwards

What AI bubble? Nvidia posts record $68 billion quarterly revenue and $78 billion forecast, as Jensen Huang cites ‘skyrocketing’ adoption of agents – by Amanda Gerut

Tech companies are spending an unprecedented $700 billion this year on AI data centers. Nvidia’s Jensen Huang says we’re not anywhere near the peak – by Alexei Oreskovic

In its fight with the Pentagon, Anthropic confronts one of the biggest crises of its five-year existence – by Jeremy Kahn

Exclusive: Bilt’s new AI ‘Neighborhood Concierge’ takes on Amazon and other e-commerce giants – by Diane Brady

Exclusive: Startup aiming to break Nvidia’s stranglehold on AI data center workloads raises $10.25 million – by Jeremy Kahn

AI IN THE NEWS

AI politics gets messy as Florida Governor Ron DeSantis leans into AI skepticism, seeking a contrast with Vance. I’ve been covering some of the political issues surrounding AI over the past year, and I find the widening gap between Florida Governor Ron DeSantis and his Republican colleagues – particularly Vice President JD Vance, to be a fascinating development. According to coverage by NBC News, he is staking out a rare position in U.S. politics as an AI skeptic, warning that the rapid expansion of AI—especially energy-hungry data centers backed by public incentives—could drive up power costs, displace jobs, and harm the environment. Casting himself as a consumer protector, DeSantis argues that large data centers can consume as much electricity as a mid-size city and says Floridians shouldn’t be left “with the bill.” His stance also reflects a political calculation as AI becomes a sharper fault line: recent polls show 63% of Americans believe AI will reduce jobs, a 33% plurality expects a net negative economic impact, and 41% of voters favor banning data centers near their homes—signals that public unease over AI’s real-world footprint is growing, even as national leaders and industry groups push aggressively in the opposite direction.

Mistral AI lands Accenture as its latest partner. European model maker Mistral AI, which has faded somewhat from the day-to-day conversation about AI’s front-runners, just scored a meaningful vote of confidence by landing Accenture as a major new 'systems integrator' partner. According to the Wall Street Journal, under a multi-year deal, Accenture will make Mistral’s models and tools available to enterprise customers as it helps them move from AI pilots to full-scale deployment. The partnership adds Accenture to a customer list that already includes IBM, Cisco, SAP, Stellantis, and ASML, which invested more than $1.5 billion for an ~11% stake last year, valuing the Paris-based startup at nearly $14 billion, and highlights how consultancies are increasingly shaping which AI models actually get used in the real world. Of course, Accenture is not offering its clients Mistral's models exclusively. The consulting firm also struck a major deal with OpenAI earlier this week to help clients integrate its Frontier enterprise AI agent platform, and it has a multi-year partnership with Anthropic too. 

AI complicates old internet privacy risks. This interesting New York Times piece hones in on a string of recent incidents that is sharpening questions about privacy and accountability in the age of chatbots and AI assistants. There was the federal judge who ruled that conversations with Anthropic’s Claude are not protected by attorney-client privilege. Smart doorbell Ring faced backlash over a Super Bowl ad that highlighted AI-powered neighborhood surveillance. And OpenAI disclosed it had reviewed a user’s ChatGPT messages months before a deadly shooting—raising fresh debate over when AI firms should share private chats with authorities. Privacy experts say the underlying risks aren’t new—data sent to companies has long been accessible under certain conditions—but chatbots change the stakes by encouraging people to share far more intimate, explicit thoughts than traditional tools, a dynamic likely to intensify as AI “agents” gain broad access to users’ emails, messages, and devices.

EYE ON AI NUMBERS

57%

That’s how many U.S. teens say they’ve used AI chatbots to search for information, according to a new Pew Research survey of 1,458 teens ages 13–17, released this week. The study shows tools are already deeply embedded in teen life: 54% say they’ve used chatbots to help with schoolwork, 47% for fun or entertainment, and 10% say they do all or most of their schoolwork with AI. At the same time, 59% of teens say AI-related cheating happens at their school at least somewhat often—but views of the technology skew optimistic, with 36% expecting AI to have a positive impact on their lives over the next 20 years, compared with 15% who expect a negative one.

AI CALENDAR

Feb. 24-26: International Association for Safe & Ethical AI (IASEAI), UNESCO, Paris, France.

March 2-5: Mobile World Congress, Barcelona, Spain.

March 16-19: Nvidia GTC, San Jose, Calif.

April 6-9: HumanX, San Francisco.