On March 9, a federal judge handed Amazon a preliminary injunction blocking Perplexity's Comet browser from accessing its site. On March 12, at its inaugural Ask 2026 developer conference - held, with apparent deliberateness, inside a former church in San Francisco's North Beach - Perplexity CEO Aravind Srinivas announced Personal Computer, a persistent AI agent that runs 24 hours a day on a user-owned Mac mini, with full access to local files, apps, and active sessions. The two events, separated by 48 hours, define the strategic bind that every agentic AI company now faces: the open web is closing, and the local machine is the last frontier.
The Lawsuit That Set the Stage
Amazon filed suit against Perplexity in November 2025, alleging that its Comet browser disguised automated agent sessions as regular Google Chrome traffic to evade detection while its agents made purchases and accessed user accounts on Amazon on behalf of users.[1] U.S. District Judge Maxine Chesney ruled on March 9 that Amazon had provided "strong evidence" of unauthorized access, and issued a preliminary injunction barring Comet from both public-facing and password-protected Amazon pages - including Prime accounts.[1] Perplexity called the suit a "bully tactic" and said it would "continue to fight for the right of internet users to choose whatever AI they want."[1]
The ruling is narrow in scope but broad in implication. Amazon is not alone: it has already blocked dozens of third-party AI agents from its shopping properties, including OpenAI's ChatGPT, while developing its own in-house assistant, Rufus.[1] The pattern is a preview of where the web is heading. Platforms that built their audiences on open crawling are now aggressively reasserting control at the moment AI agents - not humans - have become the primary consumers of their content.
The Local Bet
Personal Computer is Perplexity's structural answer to that problem, even if the company would not frame it that way. Rather than sending agents across a web that increasingly does not want them, the product plants a persistent agent inside the user's own environment.[2] The software runs continuously on a dedicated Mac mini - user-supplied, not manufactured by Perplexity - and connects to Perplexity's cloud orchestration layer, which routes tasks across 20 frontier models and integrates with more than 400 applications including Gmail, Slack, GitHub, Notion, and Salesforce.[2]
The product is software, not hardware. That distinction matters commercially: Perplexity takes no margin on the device, carries no inventory risk, and can eventually expand to other always-on platforms. The Mac mini is a starting point, not a constraint. At launch, access is limited to Perplexity Max subscribers at $200 per month - the company's highest tier, which includes 10,000 monthly compute credits - with a waitlist now open.[2]
Srinivas framed the ambition plainly at the conference: "A traditional operating system takes instructions; an AI operating system takes objectives."[2] On X, he was more direct: the single biggest disadvantage any person has, he wrote, is that they have to sleep. "A true personal computer has to be always-on."[3]
The Orchestration Argument
Perplexity's commercial thesis rests on a single claim: that the value in AI will accrue not to the companies that build frontier models, but to those that orchestrate them intelligently. Its own usage data, cited at the conference, illustrates the bet in practice. In January 2025, 90 percent of enterprise queries routed to just two AI models. By December 2025, no single model accounted for more than 25 percent of usage.[4] The company is wagering that enterprises will increasingly refuse to commit to a single provider's stack - and that the harness that deploys all models will be worth more than any one of them.
The Ask 2026 conference widened that argument in several directions. Perplexity brought its Computer platform to enterprise customers, adding SOC 2 Type II compliance, SAML single sign-on, isolated sandboxing per query, and native integrations with Snowflake, Salesforce, HubSpot, and direct access inside Slack.[4] It also expanded Perplexity Finance, giving Computer direct access to more than 40 live data tools drawing from SEC filings, FactSet, S&P Global, Coinbase, and LSEG - with every figure traceable to its source and no additional API license required.[4]
The Trust Problem
Personal Computer's security architecture is conspicuously emphasized in every Perplexity communication about it: a full audit trail for every session, a kill switch, and explicit user approval required before sensitive actions execute.[2] The emphasis is not accidental. A company under a federal injunction for concealing its agents from a major platform has an acute credibility problem when asking users to grant those same agents persistent, always-on access to their local files and private accounts.
The tension is not merely reputational. It is architectural. An agent that runs continuously on your machine, monitors triggers, and executes proactive tasks without requiring your presence is, by definition, difficult to audit in real time. The kill switch is a comfort; it is not a substitute for trust built over time. Perplexity's answer to that problem - the audit trail, the approval gates - is technically reasonable. Whether it is persuasive to the enterprise buyers the company needs to hit its internal revenue target of $656 million by end-2026 (against approximately $148 million annualized as of mid-2025) is a different question.[4]
A Turning Point for Agentic Commerce
The Amazon injunction and the Personal Computer launch together mark something larger than one company's strategy. They signal that the first phase of the agentic web - in which AI agents roamed freely across platforms built for human users - is ending. What replaces it is a more contested landscape: platforms asserting legal control over automated access, AI companies responding by moving deeper into controlled local environments, and courts adjudicating the boundary between user authorization and platform consent.
Perplexity is not the only company navigating this. It is simply the most exposed. The Amazon ruling is the first major legal test of agentic commerce, and Chesney's ruling implicitly holds that user intent does not override platform authorization - a principle every AI agent developer will now have to build around. Personal Computer, whatever its commercial merits, is also an implicit acknowledgment of that reality: if the web will not let you in, build something that does not need to knock.
Sources
