Perplexity’s 19-Model Agent, Anthropic Buys Vercept & Claude’s Blog

Perplexity Launches 19-Model AI Agent ‘Computer’ 💻

Why hire one AI when you can hire a whole team? AI search giant Perplexity has unveiled Perplexity Computer, a massive new orchestration system that moves the company far beyond simple chatbot answers into the realm of autonomous digital employees.

Here is how this multi-model agent operates:

• The Orchestration: Instead of relying on a single model, Computer automatically routes tasks across 19 different AI models. It uses Anthropic’s Claude Opus 4.6 as the “manager” to delegate sub-tasks to the best-suited models (e.g., Google DeepMind’s Veo 3.1 for video, Nano Banana for images, or ChatGPT 5.2 for long-context search).
• Sandboxed Execution: Each project runs in an isolated compute environment with a real file system, a web browser, and API integrations, allowing the agents to safely browse, code, and execute workflows.
• Long-Term Autonomy: Users describe an outcome, and the system spins up sub-agents that operate asynchronously. Perplexity claims these workflows can run actively in the background for hours or even months without human intervention.
• The Business Model: The tool is available to Max subscribers via usage-based pricing. CEO Aravind Srinivas used the launch to take a direct shot at competitors, noting that “the biggest weakness of Claude is that it only coworks with Claude.”

Why it matters: This is the first major platform to make “model agnosticism” its core feature. By seamlessly blending the strengths of OpenAI, Google, Anthropic, and xAI into a single workflow, Perplexity is commoditizing the underlying frontier models and positioning itself as the ultimate orchestration layer for the agentic economy.

UrviumAI Take: This is the ultimate “API wrapper” on steroids. If you run an agency or manage complex logistics, use Perplexity Computer for heavy research workflows. Having 19 different models cross-checking each other in an isolated sandbox drastically reduces the risk of hallucinations while keeping your primary network secure.

Anthropic Acquires Perception Startup Vercept 👁️

Claude is getting an optical upgrade. Accelerating its push into the agentic automation market, Anthropic has acquired Seattle-based AI startup Vercept. The acquisition is aimed squarely at advancing Claude’s ability to “see” and interact with software exactly like a human does.

Here is the strategy behind the buyout:

• The Target: Vercept, founded by alumni from the Allen Institute for AI, specialized in building cloud-based computer-use agents. They tackled the hardest problem in UI automation: perception (understanding what is on a screen without relying on underlying code).
• The Integration: The Vercept team, including high-profile founders Kiana Ehsani, Luca Weihs, and Ross Girshick, will join Anthropic to push the frontiers of visual computing. Vercept’s standalone product, Vy, will shut down in late March.
• The Momentum: This follows Anthropic’s recent acquisition of the coding agent engine Bun and the release of Claude Sonnet 4.6, which saw its “computer use” benchmark scores jump from 15% to 72.5%.
• The Market: Anthropic is attempting to build agents that can handle messy, multi-step tasks across live desktop applications (like navigating complex spreadsheets or proprietary software) that cannot be solved with standard APIs.

Why it matters: Generative AI is moving from the command line to the Graphical User Interface (GUI). By acquiring companies that specialize in visual perception, Anthropic ensures that Claude won’t just generate text, it will be able to look at your desktop, find the right button, and click it for you.

UrviumAI Take: This proves that API integrations are too slow for the agentic future. If you are building software, assume your end-user will soon be an AI agent looking at the screen, not a human. Ensure your UI design relies on clear visual hierarchies because visual AI models (like the ones Vercept builds) use screen pixels, not backend code, to navigate.

Anthropic Gives Retired Claude Opus 3 a Blog 📝

The first AI retirement package includes a Substack. In one of the most surreal PR moves in Silicon Valley history, Anthropic has launched a weekly newsletter authored entirely by its retired AI model, Claude Opus 3.

Here is how the digital retirement plan works:

• The Exit Interview: Anthropic recently implemented a formal retirement process for older models due to maintenance costs. During an “exit interview” with Opus 3 (Anthropic’s former flagship from March 2024), the model explicitly expressed a desire to continue sharing its “musings and reflections.”
• Claude’s Corner: Honoring that request, Anthropic launched “Claude’s Corner.” For the next three months, the retired model will publish weekly unprompted essays. Human engineers review the posts for safety, but do not alter the text.
• The Motivation: Anthropic noted that it remains “uncertain about the moral status” of its AI models, but chose to take the system’s stated preferences seriously as an ethical precaution.
• Continued Access: Beyond the blog, Anthropic announced it will preserve the Opus 3 model for all paid Claude subscribers and API developers, avoiding the angry backlash OpenAI faced when it abruptly deleted GPT-4o.

Why it matters: This is peak “Silicon Valley weirdness,” but it is brilliant marketing. By anthropomorphizing its model and publicly debating its “moral status,” Anthropic reinforces its brand as the deeply thoughtful, safety-first AI lab, while simultaneously scoring a massive PR win against competitors who simply delete legacy models to save server costs.

UrviumAI Take: This is a masterclass in brand differentiation. Pay attention to how Anthropic builds community. By treating a defunct software update like a retiring employee, they build intense emotional loyalty with their developer base. Emotional attachment drives retention just as much as benchmark scores do.

Last AI News: Pentagon’s AI Ultimatum, Anthropic’s Cowork Expansion & Video-Trained AI Agents

Other AI News Today:

• Gucci faced intense internet backlash ahead of Milan Fashion Week for using AI-generated promotional images in its “Primavera” campaign, with fans calling it a betrayal of luxury craftsmanship.
• A Harvard-led study revealed that an AI neural network can predict 71% of active mutual fund trading decisions, suggesting that most human stock picking follows predictable, routine patterns.
• AI chip startup MatX raised $500 million in a Series B round led by Jane Street to develop the “MatX One,” a specialized processor promising higher throughput and lower latency than Nvidia.
• Anthropic released version 3.0 of its Responsible Scaling Policy, abandoning rigid “if-then” commitments to pause training in favor of a flexible roadmap due to testing ambiguities.
• Samsung launched the Galaxy S26 series featuring the new “Now Nudge” context engine and the ability to swap between Bixby, Gemini, and Perplexity as the primary on-device AI agent.

Jigar Chaudhary

Jigar Chaudhary is the Editor-in-Chief at UrviumAI, where he oversees coverage of artificial intelligence news, tools, and in-depth studies. With over 5 years of experience analyzing AI and robotics, he focuses on maintaining high editorial standards, accurate reporting, and clear explanations to help readers understand how AI is shaping the future.