With today’s (Dec. 11) launch of Opera’s Neon artificial intelligence (AI) browser, available for $20 a month, the field of competitors is expanding, promising an AI assistant that can browse, click, and reason on a user’s behalf. These new tools, like Neon’s ability to analyze videos, draft text, and perform multi-step actions, are positioning themselves as proactive alternatives to traditional browser extensions, moving from simple assistance to full-blown autonomy.
But as more entrants arrive, a central question becomes harder to ignore: Are these tools ready for the high-stake financial and business workflows they were meant to support, or are they falling short of their transformative promise?
Bloomberg reported that companies building agentic browsers hoped to break decades of platform lock-in by offering assistants that could carry out multistep tasks on the open web. Instead of typing queries, clicking links or switching tabs, users would simply issue instructions for the browser to interpret and execute.
The promise appealed to developers who saw an opportunity to automate the manual work people still perform online. But tests show that the technology does not consistently understand page layouts, decision points or the logic behind basic user intent. As covered by PYMNTS, early adopters describe stalled actions, looping behavior or results that miss the point, which slows down workflows that the tools were meant to accelerate.
Expectations Mismatch
Websites are built for human cognition, visual scanning and direct manipulation do not always translate into machine-readable workflows. That gap forces the models to guess the meaning of buttons, fields and menus, which increases the likelihood of error.
A common use case for AI browsers is summarizing long YouTube videos and querying the assistant about topics mentioned in them, representatives from OpenAI and Chrome said. Comet users are also asking “six to 18 times more questions” than they did with the regular Perplexity chatbot, according to Jesse Dwyer, the company’s head of communications as reported by Bloomberg.
And consumer demand is expanding. Adam Fry, a product lead on OpenAI’s Atlas browser, said the team has received many requests from power users who want to “schedule tasks so that the browser can repeat them on a regular basis.”
That capability could support scenarios where, for example, a finance professional sets the browser to automatically generate a report or dashboard from an online tool each month.
Security and Trust Barriers
The trust problem extends beyond capability. AI browsers introduce new security risks that traditional platforms have spent years hardening against, and those gaps could expand as the tools become more autonomous. That concern centers on prompt-injection attacks that can manipulate AI reasoning through hidden instructions embedded in websites. If an AI agent misinterprets malicious content as a legitimate command, it could perform actions that compromise user safety. Companies working in this space acknowledge the need for new guardrails, but the security model is not yet mature.
Enterprise buyers remain cautious for the same reason. Corporations cannot risk deploying tools that might execute unintended actions on financial platforms, internal systems or customer accounts. The browser has long been a controlled surface where user intent is explicit and verifiable.
Agentic workflows introduce ambiguity that many organizations are not prepared to manage. Even consumer users hesitate to let an automated system fill forms, complete purchases or interact with sensitive services. Until security frameworks catch up, AI browsers will remain limited to low-stakes experimentation.
Meanwhile, legacy browsers are adapting quickly. As reported by PYMNTS, Google has integrated Gemini capabilities into Chrome, giving users AI summarization, search refinement and writing tools inside the interfaces they already trust.
Microsoft is building similar capabilities into Edge. These developments shrink the perceived benefit of fully agentic browsers. If users can access AI features without learning a new tool or tolerating instability, they have little incentive to switch.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
