News
White Paper
Why Enterprise AI Fails at Scale
On Wednesday, Feb. 25, Perplexity introduced Perplexity Computer, a new system it describes as a general-purpose digital worker designed to create and carry out complex workflows across tools, models and time.
“Frontier AI models are getting smarter,” the company said in its official announcement. “The best are becoming so capable that the products built around them are a bottleneck for showing their true potential.”
Perplexity Computer is an answer to that bottleneck — a system that goes beyond chat-based responses or one-off agent tasks. According to the company, it can reason through large objectives, break them down into subtasks and deploy sub-agents to complete them, running for extended periods of time if needed.
“Chat interfaces have answers, while agents can do tasks,” the company said. “Perplexity Computer is a system that creates and executes entire workflows, capable of running for hours or even months.”
Table of Contents
- From Prompt to Workflow
- A Multi-Model Strategy
- Reviving the Old Definition of 'Computer'
- Perplexity Computer Availability and Enterprise Focus
From Prompt to Workflow
Perplexity Computer operates much like a human knowledge worker, by interacting with the same software stack and interfaces that people use.
Users describe an outcome rather than a step-by-step instruction list. From there, the system decomposes the goal into tasks and subtasks, spawning sub-agents to handle different components of the job. One agent might gather data through web research, another could draft documents, while others process data or call APIs connected to the user’s services.
The work unfolds asynchronously. Users can step away or run multiple instances in parallel. If the system encounters a roadblock, it generates additional sub-agents to troubleshoot — searching for missing API keys, conducting supplemental research or even writing code to build an application required to complete the task.
Each job runs in what Perplexity describes as an isolated compute environment, complete with access to a real file system, browser and tool integrations. The goal, the company says, is to provide “a safe harness for powerful AI,” eliminating the need for complex local setup while giving users access to cutting-edge model capabilities.
A Multi-Model Strategy
One major feature of Perplexity Computer is its model-agnostic architecture. Rather than relying on a single foundation model, the system orchestrates multiple leading models, selecting them based on task specialization.
Experts argue that AI models are becoming commoditized. Perplexity says the opposite: as frontier systems improve, they are diverging into highly specialized tools. Some excel at long-context reasoning, others at research synthesis, coding, image generation or speed. Perplexity Computer aims to leverage fragmentation.
As of launch, the system runs Opus 4.6 as its primary reasoning engine while routing subtasks to other models depending on the job. Gemini is used for deep research and sub-agent creation, Nano Banana for image generation, Veo 3.1 for video, Grok for fast lightweight tasks and ChatGPT 5.2 for long-context recall and broad search.
The orchestration layer itself remains flexible. As models evolve, Perplexity says, the harness can swap in new systems. Users can also manually choose which models power specific subtasks, a feature the company frames as increasingly important as token usage and cost management become more central concerns in enterprise environments.
Reviving the Old Definition of 'Computer'
To frame the launch, Perplexity looked back at the original meaning of the word “computer.”
In 1757, mathematician Alexis Clairaut enlisted two human “computers” — apprentices tasked with complex calculations — to refine Edmond Halley’s prediction of a comet’s return. Dividing the labor and working for months, the trio predicted the perihelion of Halley’s Comet within two days of accuracy.
The terminology has changed, but the underlying idea remains consistent: dividing complex work into coordinated parts to achieve accurate results.
“What they’ve lacked is a system that can orchestrate capability across tasks, tools, and time,” the company said of modern AI models. “Until today.”
Webinar
Feb
26
Do More with Less: Modernizing the Cloud Contact Center for 2026
Learn how to leverage cloud platforms without adding a single hire to personalize every customer interaction.
Register
Webinar
Mar
25
Content Leaders Collective: Navigating Content Decisions at Scale
Discover how content leaders are modernizing content operations, avoiding costly missteps and preparing for scale and AI.
Register
Webinar
On demand
Content Strategy Leaders Live: Scaling for Speed, Complexity and AI in High Tech
A candid roundtable on how high-tech leaders are rethinking content at scale.
Watch Now
Webinar
On demand
Cut the Noise: Deploying AI That Actually Moves the Needle
Learn how to turn AI experimentation into concrete revenue operations.
Watch Now
Webinar
On demand
Ditch the Desk Phones: How Modern Teams Drive AI-First Communications
Find out how one team finally pulled the plug on a legacy phone system. And built something smarter.
Watch Now
Webinar
On demand
Rebrand. Migrate. Optimize. How to Do It All (Without Slowing Down)
Cresta leveled up site speed, design flexibility and marketer sanity (in record time). Find out how.
Watch Now
Perplexity Computer Availability and Enterprise Focus
Perplexity Computer is initially available to Perplexity Max subscribers, with Enterprise Max access planned soon.
It uses usage-based pricing, with optional sub-agent model selection and spending caps that allow you to choose different models for different sub-agent tasks and control token spend.
Perplexity Computer uses usage‐based pricing with optional sub‐agent model selection and spending caps so you can choose different models for different sub‐agent tasks and control token spend.
Max users get 10,000 credits/month included with their subscription.
The company is also offering a one‐time bonus of 20,000 extra credits to Max users, granted at launch for existing users and at signup for new users, that expires 30 days after it’s granted.