Samsung and Perplexity Push New Agentic Tools Into Phones and the Cloud
It was a big week in The Prompt Economy, with big moves from big companies making big claims. None were bigger than Samsung, which used its user conference to herald what it says is “The Beginning of Truly Agentic AI.”
To back that up Samsung says its Galaxy Unpacked 2026 event was designed around a simple idea: Agentic AI should feel less like a separate feature and more like a dependable part of everyday phone use.
The company introduced the Galaxy S26 and Buds4 lines and positioned agentic AI as the next step in mobile AI, where the device can understand context, suggest actions and help users complete tasks without app switching. Samsung highlighted features such as call screening summaries, “Now Nudge” suggestions based on conversational context, and an updated Bixby that can search the web while keeping results inside the conversation.
Samsung also stressed that agentic AI requires a strong foundation in privacy and on-device protection. It described an on-device “Personal Data Engine” that learns user preferences locally, paired with multiple layers of security, including Knox Enhanced Encrypted Protection, Knox Vault and new privacy controls like “Privacy Display” to reduce shoulder-surfing for sensitive apps.
Samsung also pointed to its work with Google, including expanded Circle to Search and an early preview of a more “agentic” Android platform powered by Google’s Gemini 3. To close, Samsung’s Co-CEO TM Roh framed the goal as making AI dependable and broadly usable: “Infrastructure is responsibility. It must work for everyone, everywhere.”
Perplexity’s Agentic Shift
It was also announced last week that Perplexity is launching a new cloud-based agent product called Perplexity Computer, available to subscribers on its top-tier $200/month Perplexity Max plan. The company describes it as a “computer user agent” that can carry out complex workflows on its own, including creating subagents for parts of a task.
TechCrunch says Perplexity claims the system can route work across 19 different AI models, and the company’s examples show it handling jobs that involve collecting statistical, financial or legal information and turning the results into finished outputs such as websites or visualizations.
The article also frames the launch as part of Perplexity’s broader strategy shift. TechCrunch says Perplexity made its name by packaging top AI models into a search-like experience, then expanded into products like a web browser, and is now positioning itself for a more selective audience and higher-value use cases.
In a background briefing, executives said they are not trying to maximize user counts and instead want products for people making high-stakes decisions. A key theme is Perplexity’s belief that users will need access to multiple specialized models, and that software should automatically choose the best model for a given task. As one executive put it: “Multimodel is the future.”
Managing the Agentic Identity
Now, who is that’s using these agents? Cybersecurity Insiders argues that AI agents are already operating inside enterprises, but most companies are not treating them like real “users” from a security standpoint.
In the report, Paul Walker, field CTO at Omada, says agent frameworks are making it easier for software to take actions across many systems, which blurs the line between human and nonhuman actors. Walker’s point is that adoption is moving quickly, while governance is still being built, leaving gaps in oversight and accountability.
The report’s central claim is that identity management practices built for employees do not translate cleanly to agents. Walker says agents can create “authorization without oversight,” because an agent may decide what it needs and then request or assume access. That can lead to agents gradually accumulating permissions across tools and platforms that exceed the original intent. The recommended fix is to treat agents as first-class identities with formal provisioning, clearly defined entitlements, least-privilege access, life cycle controls, and monitoring for unusual activity across systems.
Walker also stresses that agent identity is not just about “who can log in.” It is about being able to prove what the agent did, what data it touched, and which systems it instructed. The report calls for clearer audit trails and tighter controls on what services agents can invoke, so security teams can trace actions end to end and stop risky behavior early. It frames this as a baseline requirement for scaling agentic AI safely: “AI agents are already inside your enterprise. The question is: Who’s governing them?”
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
The post Samsung and Perplexity Push New Agentic Tools Into Phones and the Cloud appeared first on PYMNTS.com.