Mint Explainer | AI agents take the wheel in the new browser wars—what it means for users – Crypto News

The battle over agentic artificial intelligence (AI) browsers—tools that can book meetings, make purchases, and compare products on your behalf—is heating up, echoing the Netscape vs. Microsoft Internet Explorer clash of the 1990s. This time, Perplexity has thrown down the gauntlet by making its Comet browser free, joining rivals such as Opera Neon, DuckDuckGo, Dia, Bravo, and OpenAI’s AI shopping agent.

These autonomous browsers promise both convenience and disruption, while raising new questions about privacy, security, and the risks of unchecked automation.

Mint unpacks the latest developments.

What’s this battle all about?

Web browsers have long been gateways to the internet—but traditionally, browsing is manual. Users click links, type searches, compare products, and make purchases themselves. Agentic AI browsers change that. They can navigate websites, fill forms, execute transactions, negotiate prices, and even complete purchases automatically. The underlying web infrastructure remains the same, but user actions are increasingly delegated to AI.

Perplexity AI, founded by Aravind Srinivas, is raising the stakes with its free Comet browser. Like other agentic browsers such as Opera Neon and Dia, Comet can summarize pages, answer questions, automate searches, and still offer standard features like bookmarks, page translation, and support for most Chrome extensions.

History offers a cautionary tale: the 1990s browser war ended with Microsoft discontinuing Internet Explorer in 2022 and Netscape being acquired by AOL, which just recently shut down its dial-up service.

What exactly is Agentic AI?

Agentic AI, also known as AI agents, refers to models that can make autonomous decisions and act without human input. They are adaptable, decision-capable, and now even shop for you. OpenAI’s shopping agent in the US, for instance, enables users to buy via Instant Checkout, powered by Stripe’s Agentic Commerce Protocol—an open standard that allows AI, people, and businesses to complete purchases together.

As per Precedence Research, the global agentic AI market size accounted for $5.25 billion in 2024 and is forecast to rise from $7.55 billion in 2025 to about $199.05 billion by 2034. Gartner predicts that by 2028, one-third of GenAI interactions will involve autonomous agents.

What about limitations like security, privacy, and accuracy?

Autonomy comes with trade-offs. Handling sensitive data makes agentic AI browsers vulnerable to misuse, leaks, or hidden sharing. AI agents can churn out code quickly, but errors without human oversight remain likely. Even advanced agents can hallucinate, miss mistakes, or sabotage systems they’re meant to improve. A Model Evaluation & Threat Research (METR) randomized controlled trial found developers using AI tools took 19% longer to complete tasks.

Security is another concern: 80% of breaches involve compromised identities, and generative AI amplifies this risk by enabling more sophisticated phishing and social engineering, according to Eric Kelleher, president and COO of Okta, a cloud-based identity and access management services company.

What are the other big concerns?

Many vendors, according to research and advisory firm Gartner, are rushing to add agentic features to their roadmaps, driven by market pressure and competitive noise but few are pausing to validate use cases or model the operational complexity.

As the coexistence of AI and human workers enhances efficiency, it will also require a re-evaluation of job roles. Ethical and security risks add to the concerns, as AI agents could be weaponized for phishing, misinformation, or cyberattacks. Moreover, as AI agents continue to evolve, governments must step in with regulations to ensure accountability, or we may see a rise in lawsuits over flawed AI-driven decisions.

As AI agents advance, governments will need to regulate for accountability—or risk a surge in lawsuits over flawed AI decisions.

How to address the issue?

Due to rapid adoption, the number of agentic identities is expected to exceed 45 billion this year—over 12 times the global workforce—yet only 10% of executives surveyed by Okta have strategies to manage them.

Kelleher recommends limiting agent access to defined time windows, enforcing interoperability standards, and implementing strong monitoring and auditing to maintain control.

Agentic AI browsers can mitigate privacy risks by limiting data collection, giving users control over AI actions, ensuring strong security, and being transparent about data use, often detailed in their frequently asked questions (FAQs) section. Compliance with regulations and industry standards further strengthens safety and accountability.

Gartner advises building explainability into the user experience, restricting autonomy to controlled environments, and implementing fail-safes and oversight. The firm encourages vendors to tie pricing to measurable outcomes, like time saved or error reduction.

Further, AI agents require unique identities to access data and workflows, managed via application programming interface (API) tokens or cryptographic certificates rather than human passwords or multi-factor authentication (MFA).

Without strict governance, these identities risk excessive access and autonomy. Robust frameworks must control permissions and continuously monitor activity to prevent hidden vulnerabilities.