Beginning this week, subscribers to Perplexity will gain access to a new autonomous tool.

According to Perplexity Computer, it “integrates all existing AI capabilities into a singular system.” More precisely, Perplexity describes it as a user agent that can carry out intricate workflows autonomously through the use of 19 distinct AI models, even generating subagents for specific tasks.

This tool is currently available, but only for the company’s premium subscription level, the $200/month Perplexity Max. It operates completely in the cloud, which may alleviate some security issues associated with other autonomous tools such as OpenClaw.

While TechCrunch hasn’t conducted a hands-on demonstration of the new tool, example workflows presented on Perplexity’s website show it managing tasks that entail gathering statistics, financial, or legal information; performing analysis; and presenting its results as completed websites or visual representations. 

Last week, Perplexity invited the media to a background briefing with its executives to discuss the product and outline plans for the year. The event was supposed to include a demonstration of the tool, but the company canceled it due to identified issues in the product just hours prior.

This tool signifies Perplexity’s progression, which initially gained attention early in the AI surge by integrating advanced models into user-friendly interfaces, especially its search-engine-like answer service. It later introduced its Comet web browser last summer. An executive remarked that competitors like Google have since adjusted their offerings to resemble those created by Perplexity, viewing it as both a compliment and a potential challenge.

The company is adapting to a changing market: One of the early AI firms to provide advertising, it withdrew from that field late last year, stating last week that it damaged users’ trust in the accuracy of their responses. However, Perplexity’s total user base — numbering in the tens of millions — is significantly smaller compared to OpenAI, which boasts 800 million weekly users and began testing advertisements in ChatGPT this year.

Currently, Perplexity’s executives claim they are targeting a more niche group of users, with products that cater to individuals making “GDP-impacting decisions.” During the briefing, executives who requested anonymity detailed a focus on enterprise subscriptions, especially for in-depth research.

“We don’t often discuss MAUs because we’re not fundamentally pursuing a strategy to acquire as many users as possible,” one of the executives said.

Perplexity has recently introduced a new benchmark for complex research tasks, known as Draco, where, unsurprisingly, its own deep research offering outperforms rivals like Gemini. 

Perplexity asserts that it is no longer dependent on external companies’ APIs for its web index and has developed its own AI-optimized search API. Nevertheless, the company is committed to packaging cutting-edge models within a consumer-friendly user experience, arguing that there is significant value in orchestrating multiple third-party LLMs to find the most cost-effective and accurate answers to inquiries.

“Multi-model is the future,” one Perplexity executive asserted. They believe models are becoming more specialized rather than commoditized. The company has observed that its users frequently switch between models to achieve their desired results, with queries in December 2025 for visual outputs predominantly directed to Gemini Flash, software engineering tasks conducted with Claude Sonnet 4.5, and medical research utilizing GPT-5.1.

If one LLM excels in coding tasks while another is more effective at creating marketing text, Perplexity’s software can automatically select the best one. Another instance noted by executives involves utilizing Perplexity’s own modified open-source LLMs developed in China to answer queries at a lower cost, a tactic for which the company faced criticism last year for not being transparent with its clients. However, when done openly, this method could offer an efficient means to enhance LLM queries.

The company also provides a feature named Model Council, allowing users to query several models simultaneously. However, the economic viability of offering multiple queries at fixed subscription rates remains uncertain. 

Yet, with no costly infrastructure commitments and, as the executives asserted, high margins on user fees, Perplexity is confident it will stay competitive by directing tokens to the most suitable model for each task.

Additionally, there are upcoming developments: the Perplexity Comet browser is set to launch on iOS next month, and the company is organizing a developer conference, Ask, on March 11 in San Francisco to encourage third-party utilization of its API.

One executive mentioned that instead of reviewing the number of queries from the previous day each morning, he now examines the latest revenue figures. Some clients are noticing this shift towards a focus on profitability, with the Perplexity subreddit frequently featuring complaints regarding new rate limits on both free and paid product tiers.

Nevertheless, the executives at the briefing dismissed these concerns: “Any claims regarding the free tier becoming worse or rate-limited are entirely unfounded,” one remarked.