Perplexity has made its Personal Computer app broadly available on Mac, expanding its agentic desktop layer from limited beta to general release, with the product able to search the web, summarize documents, take actions across apps and files, answer questions about on-screen content, and operate as a persistent desktop assistant rather than a tab in a browser.
The product works through a system-level overlay that can see what is on screen, read files and documents, execute searches, and take actions through AppleScript and accessibility APIs. A user can highlight text in any app and ask Perplexity to summarize, fact-check, or act on it. The assistant handles multi-step tasks like researching a topic across the web, opening related files, and drafting a document without the user switching contexts manually. Screen awareness means the assistant has context that a browser tab cannot: what you are reading, writing, or editing becomes the prompt without additional input. That ambient context is what separates a desktop agent from a chatbot.
Perplexity's strategic position makes the Mac launch more consequential than a feature rollout. The company built its user base as an answer engine: people who wanted search results without ads, with sources cited and follow-up built in. Perplexity has reportedly reached 30 million monthly active users and processes hundreds of millions of queries weekly. That search loyalty is an asset. Users who already trust Perplexity for research are more likely to grant it desktop permissions than they are to install an unknown agent from a startup they have never used. Converting search users into workflow users is a lower-friction acquisition than competing for agent installs from scratch.
The competition is converging simultaneously from every direction. OpenAI's Operator can browse the web and take actions in a sandboxed browser. Anthropic's Claude can use computers through its computer use API. Google's Gemini integration into Chrome and Android gives it ambient context on hundreds of millions of devices. Apple Intelligence on Mac gives the operating system itself search, summarization, writing, and app integration capabilities that Apple ships at zero marginal cost to every Mac user. Perplexity is running a distribution race against players who either have the OS advantage or the model quality advantage, often both.
For SF founders, the desktop agent category illustrates the trust and permission bottleneck that limits how fast any agent can grow. Users grant screen reading permissions reluctantly. File system access raises privacy concerns. Cross-app actions require accessibility permissions that security-conscious users often disable. Mac's permission model, while more permissive than iOS, still surfaces system dialogs for each new capability a desktop app requests. Perplexity's early beta users self-selected for comfort with these permissions. Broad availability tests whether mainstream Mac users will grant them. That conversion rate matters more than feature parity.
The workflow ownership question is distinct from search loyalty. Perplexity users return for queries. Agent users need to return for tasks, meaning the assistant must integrate into existing workflows around email, calendars, documents, and communications that users already own. The best wedge is the one that sits inside the workflow rather than adjacent to it. Notion's AI integration is valuable because it lives inside the documents users already maintain. Linear's AI understands the project management workflow natively. Perplexity's challenge is that its natural habitat is research and fact-finding, which is a distinct workflow from the document editing, calendar management, and email triage where most desktop time is spent.
Whether desktop agents become startup wedge products or get absorbed by OS incumbents depends on execution speed and distribution. Apple Intelligence is improving but remains behind ChatGPT and Claude on complex reasoning. Microsoft Copilot is deeply integrated into Windows and Office but struggles with trust among users who associate it with corporate surveillance. The window for Perplexity is the gap between where OS AI assistants are today and where they will be in 18 months. If Perplexity builds enough habit and workflow integration in that window, it creates switching costs that survive Apple Intelligence improvements. If it does not, the OS layer wins by default, and Perplexity reverts to a search product with agent features rather than an agent product that happens to search.
Also read: AI data center backlash is turning into bipartisan political reality • Apple's camera AirPods are the ambient AI hardware layer that startups cannot build alone • OpenAI's new audio models unlock voice-native agents with realtime reasoning and translation
