Perplexity Gives Users More Control Over How Computer Solves Tasks

Perplexity is adding effort controls to its Computer AI agent, allowing users to choose how much model coordination and reasoning a task receives. The feature is available on the web now, with mobile and desktop support planned for a later release.

A new control layer for AI agents

Perplexity announced the change in a post on X, saying the controls are being added to Computer’s model selector. The company described the new options as “effort presets” that combine two variables: the orchestrator model responsible for coordinating the agent’s work, and the depth of its reasoning.

That distinction matters because Computer is designed to do more than generate a single response. An agent can break a request into steps, use tools, navigate information, make decisions and check the results. Each additional step can improve the quality of the outcome, but it can also require more time, model calls and computing resources.

Perplexity’s effort controls appear intended to make that tradeoff visible to users. Rather than relying on one fixed operating mode, Computer will offer different preset combinations that determine how intensively it works through a task.

The company has not disclosed the names of the presets, the models included in each option or the precise differences in reasoning depth. It also has not said whether users can separately select an orchestrator and a reasoning level, or whether those variables are locked together inside each preset.

Those omissions leave the commercial value of the feature difficult to measure for now. A control that merely changes labels would have limited significance. A control that meaningfully changes the number of steps an agent takes, the models it invokes and the reliability of its output could become an important product differentiator.

Why effort controls matter commercially

For Perplexity, the feature addresses a central problem in the economics of agentic AI. More capable models and longer reasoning processes can produce better results, but they are also more expensive to operate. If every request receives the maximum level of effort, the provider absorbs higher inference costs and users may face slower responses or stricter usage limits.

Preset effort levels provide a potential way to align resource consumption with the value of the task. A user asking for a quick factual lookup may prefer a lighter option that prioritizes speed. A user asking Computer to research a market, compare products, organize information or complete a complex workflow may accept longer processing in exchange for deeper planning and verification.

That creates a clearer link between user intent and infrastructure spending. Perplexity can reserve its most expensive model combinations for tasks where users are willing to wait, while handling simpler requests with less computational overhead. If the company eventually ties effort levels to subscription tiers or usage allowances, the controls could also support a more structured pricing model.

The feature may therefore be as much about cost management as user experience. AI companies are under pressure to make agents useful without allowing every interaction to become an open-ended sequence of model calls. Giving users a visible choice could help reduce frustration on both sides. Users gain more predictability, while Perplexity gains a mechanism for managing demand.

Positioning Computer in a crowded market

The move also reflects a broader shift in how AI products are being presented. Chatbots traditionally compete on answer quality, speed and access to information. Agents compete on whether they can reliably complete a process from beginning to end. That makes the system’s internal allocation of time and computation part of the product experience.

Perplexity is positioning Computer as a system that can work through tasks, rather than simply respond to prompts. The effort selector reinforces that positioning by treating depth as something users can configure. It suggests that the company wants Computer to serve both casual users seeking immediate answers and more demanding users who need a research or automation partner.

This could help Perplexity distinguish its product from general-purpose chat interfaces. The company’s advantage has historically been tied to search, research and answer delivery. Computer extends that proposition toward execution. Effort controls give the extension a practical interface and may make the product easier to understand as its capabilities become more complex.

However, the feature will only create a durable advantage if the presets produce consistent differences in performance. Users need to know what they are gaining from a higher effort level. That could include better source evaluation, more complete task execution, stronger error checking or improved performance on complicated instructions. Without clear feedback, effort settings risk becoming an abstract technical choice that most users ignore.

Web availability comes first

Perplexity said the controls are available now on the web. Mobile and desktop applications are expected to receive the feature later, but the company did not provide a timetable.

Starting on the web is a logical distribution choice. Browser users are often more likely to conduct extended research or complete multi-step tasks, and Perplexity can test how people use the controls before expanding them to other platforms. The web rollout may also allow the company to gather data on which settings users prefer, how much latency they tolerate and whether higher effort leads to greater engagement.

That information could influence future product and pricing decisions. If users consistently choose deeper settings for valuable workflows, Perplexity may have evidence that Computer can support premium plans or specialized business offerings. If most users select lighter modes, the company may need to focus on speed and cost efficiency rather than maximum reasoning depth.

The announcement is limited in scope, but its direction is significant. Perplexity is making the agent’s operating intensity part of the user relationship. The next test will be whether that control translates into measurable improvements in reliability, speed and cost. Until the company reveals the presets and their performance differences, the rollout is best understood as an important interface and infrastructure experiment, not yet a demonstrated competitive breakthrough.