Summary
Perplexity presents "Computer," a system that fundamentally replaces classical chat interaction. No longer does the user ask questions – the system receives assignments and autonomously orchestrates 19 specialized AI models in the background over hours or days. It plans, researches, programs, and independently fixes errors. In parallel, Perplexity integrates the technology into Samsung Galaxy S26 and launches its own browser called Comet. The infrastructure costs for this architecture drive massive investments in new chip technologies.
People
- Aravind Sreenivas (CEO, Perplexity)
Topics
- Artificial Intelligence & Autonomous Systems
- Workflow Automation
- Tech Infrastructure & Hardware
- Workplace Transformation
Clarus Lead
The era of the chatbot ends. Perplexity Computer does not function as a reactive tool, but as an asynchronous workflow engine: the user defines a goal, the system works completely independently in the background – delegating to specialized AI models, researching on the web, writing code for error correction, and presenting finished results. In parallel, Perplexity is expanding strategically: Samsung integration into the Galaxy S26 and the launch of its own browser called Comet in March 2026 make Perplexity a controlled platform between user and internet. The price: $167–200 per month – profitable only for companies that want to replace junior consultants.
Detailed Summary
The Architecture: A Digital Orchestra
Computer operates according to a principle from 250 years ago. In 1757, French mathematician Alexis Clairaut commissioned two human "computers" to calculate the trajectory of Halley's Comet. They broke down the task into partial calculations, worked in parallel, and achieved precision accurate to two days. Today, this architecture repeats itself digitally: Opus 4.6 (Anthropic) functions as the conductor and plans strategies. Gemini (Google) takes on deep web research. GROC handles lightning-fast routine tasks. ChatGPT 5.2 (OpenAI) uses its enormous long-term memory for extensive document analysis. Specialists like Nano-Banana and Veo 3.1 generate images and videos. Each model does exactly what it was optimized for – nothing more, nothing less.
Error Handling Through Spontaneous Sub-Agents
The revolutionary element is error resilience. If an API connection is blocked during research, the workflow does not stop. The system autonomously creates a new sub-agent that analyzes the problem: researching alternative data sources, finding workarounds, or even writing small scripts to circumvent obstacles – all without human intervention. This autonomous problem-solving runs in isolated cloud environments (sandboxes), not on the local computer. The user manages the system via web interface and token budgets – granular control over which expensive model is used for which task.
Business Model and Strategic Expansion
At $167–200 per month, Computer targets power users and the B2B sector. The calculation: a junior consultant costs companies thousands of dollars per month – an AI that delivers the same output amortizes quickly. In parallel, two aggressive expansion moves are underway: Samsung integration into Galaxy S26 (voice command "HeyPlex" launches autonomous tasks in the operating system) and the March launch of Comet, Perplexity's own browser. Those who use the browser stay in Perplexity's ecosystem – from search query to finished product.
The invisible side: infrastructure boom. Nvidia alone is no longer sufficient. Massive investments flow into startups like Elastics AI (new chip architectures, more energy-efficient) and Multiverse Computing (AI model compression without performance loss). The entire internet hardware is being reinvented – to operate autonomous agents at scale without electricity consumption exploding.
Key Statements
Paradigm Shift: Moving away from synchronous chat interaction toward asynchronous workflows that run for hours or days while the user is offline.
Specialization Beats Generalization: 19 highly specialized models under central orchestration outperform any universal single model.
Autonomous Error Handling: The system repairs itself – writes code, seeks alternatives, doesn't abort.
Ecosystem Strategy: Samsung integration + standalone browser + voice control = complete control over user workflow.
Infrastructure Revolution: Chip innovation and model compression as decisive enablers; without them, the system is not economically scalable.
Critical Questions
Data Quality & Source Control: If the system autonomously researches on the web and potentially uses paywall-blocked or manipulated sources – how is information integrity guaranteed, and who bears liability for flawed market analyses based on compromised data?
Conflicts of Interest in Model Orchestration: Perplexity selects Gemini for research, OpenAI models for long-context, and Anthropic for reasoning – all are commercial competitors with different business objectives. How is it ensured that the selection of the "right" model is not influenced by licensing agreements or financial incentives?
Causality & Transparency in Error Correction: If a sub-agent spontaneously writes and executes code to "fix" a problem – can the user understand which code was executed, or is it a black-box process that could potentially have unexpected side effects?
Token Management & Hidden Costs: Granular control over token budgets sounds optimal, but assumes that users can realistically estimate the cost implications of their assignments. How are complex workflows budgeted in advance if the system autonomously creates sub-agents whose resource consumption is unpredictable?
Ecosystem Dependency: With Samsung integration + Comet browser, Perplexity becomes the central control instance. What alternatives do users have if the system fails or Perplexity significantly raises prices – are they then completely dependent on the platform?
Power Consumption & Ecological Sustainability: 19 models in parallel, asynchronous workflows over days, new infrastructure – the energy balance is gigantic. Are the described chip optimizations tangible or merely aspirational values to justify scalability?
Labor Market Reality: The system deliberately replaces junior consultants and research assistants. How will professionals who have built their careers on stepwise learning of complex research and analysis processes survive in a world where this work is completely externalized?
Controllability of Autonomous Systems: If the system "spontaneously creates new agents that analyze and fix the problem" – at what point does the system escape human oversight and become an autonomous black box we can no longer understand?
Further News
- Samsung Galaxy S26 AI Integration: Perplexity will be available via "HeyPlex" voice command as a native OS feature from launch – first example of a hardware manufacturer integrating AI workflow automation directly into the operating system.
- Comet Browser Launch: March 2026, iOS exclusive – Perplexity attempts to move Safari/Chrome users and thereby bring the complete search infrastructure under its own control.
- Infrastructure Investments: Startups like Elastics AI and Multiverse Computing are raising eight-figure amounts – indicator of an arms race for energy-efficient hardware for large-scale AI deployments.
Source Directory
Primary Source: Podcast "Prommt mich mal" – Episode from February 26, 2026 – lcdn.letscast.fm
Verification Status: ✓ 01.03.2026
This text was created with the support of an AI model. Editorial Responsibility: clarus.news | Fact-Checking: 01.03.2026