For AI companies, browsers offer a unique opportunity to gather vast amounts of data, and engineer a setup where users never have to leave their ecosystems. For users, AI agents on AI-native browsers could become the access points for not just finding information but executing essential tasks. In turn, this could critically disrupt long-standing business models for publishers, vendors, and everyone else who relies on search traffic. While widespread adoption appears so way off, if it materialises at all, the stakes are high, as they were during the first "browser war" in the mid-1990s. Therefore, it is important to be up to speed on what AI browsers are, why AI companies are developing them, and what the benefits and risks could be for companies and consumers.

Google was unchallenged for too long

For many years, browsers have been an unfashionable corner of the internet, garnering little attention and seeing relatively little change. The sector's dominant force is undoubtedly Google, whose browser, Chrome, currently enjoys a 68% market share and has commanded more than 50% of the market for almost a decade. Apple's Safari is its closest competitor, with a host of other browsers making up the rest, including Edge, Firefox, and Opera.

Chrome's dominance means that Google Search, Chrome's default search engine, handles close to 90% of all worldwide searches. This fact has led to an antitrust case being brought against Google, which could result in it having to sell off Chrome or decouple it from the rest of its product offering.

While Google's dominance in the browser market remains undisputed, its wider position is looking shakier due to the rise of Generative AI (Gen AI). Google execs have reportedly predicted an "inevitable decline" in Google search engine traffic as users increasingly turn to tools like ChatGPT, Claude, and Google's own Gemini to search the internet. This appears to be already happening, with Google Search's market share recently dropping below 90% for the first time in over a decade.

The first AI browsers arrive

In May of this year, The Browser Company launched Dia, the first high-profile attempt to create an AI-native browser. While this was an invite-only beta launch, it set the ball rolling. In early July, Perplexity launched Comet, its AI browser, which is currently available to users with Perplexity's $200-per-month Max subscription. Both Comet and Dia are based on Google's open-source engine, Chromium, and both include built-in AI assistants. Comet uses Perplexity's AI search engine by default.

The launch of Comet was shortly followed by widespread reporting that OpenAI is developing its own AI browser that will feature a "native" ChatGPT interface. While OpenAI has not commented on this, the fact that it hired top talent from Chrome last year has fuelled speculation that it will make a move in this area.

This new generation of browsers, or rather the AI agents integrated within them, can process natural language, automate actions, and adapt their behaviour based on context. Users can direct their browser to manage their tabs, book meetings, make purchases, or research a topic. In other words, they have the potential to fundamentally shift the way users already interact with the web, although realising this potential at any significant scale remains some way off.

AI companies aiming to contextualise data and control defaults

According to Business Insider's chief media and technology correspondent Peter Kafka, OpenAI and others are actively interested in this field because browsers provide an almost unrivalled opportunity to "peer into your life."

There is a lot of data available in browsers, and crucially, this data is instrumental when it comes to providing context. This can help AI browsers to act more like their users because they can draw from this context when executing an action they have been asked to perform. For example, if a user has a spreadsheet with details about an upcoming work open in one tab of their browser, the AI browser can use relevant information like the budget, event date, and number of participants to execute actions on the web, such as finding and booking a suitable conference room. This contrasts with current AI tools, where the user would have to input all of these contextual details into a prompt manually.

A second factor for the AI companies is access and defaults. At a basic level, the companies that own the most popular browsers (Google and Apple) can limit the access other AI companies have to these browsers. As Peter Kafka points out, this would hinder other companies’ ability to develop effective AI agents.

A more subtle but potentially more important point is that browsers are set with defaults for particular actions. For example, the default search engine on Chrome is Google Search. Managing these defaults is a valuable way for companies to keep users within their ecosystem. Comet uses Perplexity's AI search engine by default and has Perplexity's AI agent built in. This means a user browsing the web stays inside Perplexity's universe, which helps with data gathering and, potentially, monetisation. Perplexity's CEO admitted as much in a recent interview: “We plan to use all the context to build a better user profile, and maybe, you know, through our ‘discover’ feed we could show some ads there.”

"Vibe browsing" can improve user experience

When AI educator and enthusiast Mathew Berman trialled Perplexity's AI-native browser, the term he coined to describe the experience was "vibe browsing." Just as vibe coding puts the technical task of writing code into the hands of AI, so vibe browsing entrusts an AI agent with navigating the internet and performing tasks on a user's behalf.

In some respects, this is what existing AI agents – like OpenAI's Operator and its newly released ChatGPT agent – already do. They are capable of clicking through websites, navigating a user's calendar, and running code. However, these agents don't have access to all of the context that a browser can give. The value proposition stated in promotional material for Dia sums up the specific user advantage an AI browser can offer: "if it's in your tabs, Dia already knows."

Issues with privacy, security and accuracy

Entrusting browser data and permissions to an AI agent that executes actions on a user's behalf raises a host of questions around privacy and security. To be clear, these concerns apply to all forms of agentic AI, not solely AI browsers. However, because AI browsers are improving user experience and enhancing the effectiveness of AI agents, any growth in their usage will likely amplify the importance of privacy or security issues.

AI agents interact with live systems and handle sensitive information. Their developers, including the creators of AI browsers, are keen to point out that key actions such as making purchases require user confirmation. Nevertheless, questions remain.

Data leakage is a potential problem, with the risk that AI agents might inadvertently transmit sensitive information to external environments. Regarding financial, legal, and healthcare data, there may also be compliance risks if an AI agent does not conform to data protection regulations such as GDPR. More generally, it is vital for companies to ensure a clear log of the activity an AI agent has carried out for any work-related tasks to enable traceability, auditing, and effective root-cause analysis.

Naturally, AI companies are looking to build in protections that address these privacy and security concerns. OpenAI Operator, the company's AI agent which is available to premium subscribers, was able to correctly ask for user confirmation for 92% of actions that are classified as "high risk." Other approaches to enhancing the security of AI browsers include "supervision modes," where the AI agent pauses when a user has become inactive. This mode can be applied to sensitive platforms like email.

A more general question mark remains around accuracy and bias. LLMs are prone to hallucination, meaning an AI agent could simply make up a reasoning step or action, which could result in incorrect entries or miscommunication. Generative search tools have been known to invent citations used in research, or fail to cite sources, which could impact the output and decision-making of an agent in an AI browser. Placing power into the hands of AI agents to not only retrieve information but also actively interact with the web means the effects of any inaccuracies or biases will be amplified. For this reason, and due to wider concerns over intellectual property and plagiarism, one browser company, Vivaldi, has committed to never incorporating AI into its browser.

Accelerating economic pressure on publishers, vendors, and SEO companies

The rapid development of AI browsers raises more questions for already embattled publishers, vendors and marketers. The growing use of Gen AI tools for search is reducing traffic to websites. The recent addition of AI overviews on Google search has exacerbated this situation – 2024 research into Google search patterns suggests that 60% of all searches are now "zero-click".

In this context, any shift towards widespread AI browser adoption would likely accelerate this trend. That is because AI agents in AI-native browsers display search results and complete actions without the user having to visit any websites.

In terms of marketing and SEO, AI tools already circumvent existing Google search practices and prioritise different types of content. Marketers and agencies are turning to GEO (generative engine optimisation) to understand how Gen AI selects sources and shares links. Agentic AI already adds complexity to this process, and AI browsers will potentially take this up a notch. That is because the AI agents in a browser use context to generate results and execute tasks, meaning that each output is highly personalized.

Some time before AI browsers are widespread

We remain some way off from the mass adoption of AI browsers. There are currently two large-scale AI browsers available: one of them, Dia, is invite-only, and the other, Comet, requires a $200-per-month Perplexity Max subscription. It remains unclear when OpenAI's AI browser will launch, or when Google's AI Gemini tool will be rolled out to Chrome.

It is also worth remembering the reason why, in recent years, browsers were not the most cutting-edge area of software development. The average user spends little or no time thinking about their browsers, and simply elects for the default on the device or desktop they are using. Even if AI browsers offer a range of agentic bells and whistles, it may prove challenging to shift user habits, especially if there is a cost involved. On that topic, agentic AI requires significant computational power, which may present challenges to AI companies when it comes to making AI browsers affordable for mass adoption.

While this new instalment of the "browser wars" could have a fundamental impact on the way we interface with the web, we remain at the early stages of AI browser development. The changes AI browsers cause could be monumental, but they are not happening yet, and indeed, they may not be with us for some time.

For more startup news, check out the other articles on the website, and subscribe to the magazine for free. Listen to The Cereal Entrepreneur podcast for more interviews with entrepreneurs and big-hitters in the startup ecosystem.