Atlas & Comet: Unlikely Winners in the AI Browser War
▼ Summary
– OpenAI and Perplexity have launched new AI browsers, Atlas and Comet, which they frame as transformative tools that could reshape internet use.
– The author argues the companies’ claims are overblown, suggesting their strategy may be influenced by survivorship bias, imitating successful tech giants like Google rather than solving a clear user need.
– These “agentic” browsers automate web tasks but create significant problems for analytics, advertising, and security, including new vulnerabilities like prompt injection attacks.
– While potentially useful for developers and technical users for testing and automation, the browsers lack a clear, safe value proposition for mainstream consumers to compete with established browsers.
– The future of AI-powered browsing is more likely to be shaped by incumbent browsers like Chrome or Edge, which have existing user bases and integrated AI models, rather than by these new entrants.
A year ago, predicting that a web browser would become the next major battleground for leading artificial intelligence firms would have seemed far-fetched. Yet, that is precisely the arena where OpenAI and Perplexity have chosen to compete, introducing their respective AI-infused browsers, Atlas and Comet. Both companies present these tools as revolutionary steps toward redefining the everyday internet experience. OpenAI’s Sam Altman has called it a rare chance to completely reimagine browsing, while executives from both firms have liberally used the term “operating system” to describe their ambitions, framing Atlas and Comet as foundational platforms for digital life and cognition.
Despite the transformative language, the current reality feels less groundbreaking. Calling a browser an “operating system” is more of a marketing flourish than a technical reality. Established browsers like Chrome, Edge, and Safari are deeply integrated into vast ecosystems of digital tools and services, creating sticky workflows for users. The new entrants, in contrast, appear to be replicating a familiar playbook. This raises a question: in the urgent search for a sustainable business model, have these AI pioneers fallen prey to a classic logical error?
The concept of survivorship bias offers a compelling lens. Famously illustrated during World War II, military analysts initially recommended reinforcing areas on bomber planes that showed the most bullet holes. Statistician Abraham Wald corrected this, pointing out that they were only examining planes that survived; the critical vulnerabilities were on the planes that never returned. This bias, focusing only on successes while ignoring failures, permeates business strategy. It’s seen when people cite billionaire college dropouts while ignoring the multitude who did not find success, or when self-help gurus promote the habits of the wealthy without acknowledging the many who share those habits but not the fortune.
In a similar vein, OpenAI and Perplexity seem to have observed that tech giants like Google, Microsoft, and Apple all maintain their own browsers and concluded that ownership of this tool is a key to enduring dominance. This imitation became particularly clear last year when both companies explored acquiring Chrome during Google’s antitrust proceedings. With that avenue closed, they built their own. It is no surprise, then, that Atlas and Comet feel familiar; both are built on the same Chromium foundation as Chrome and mirror its core interface of tabs, address bars, and extensions.
Their differentiating feature is the integration of “agentic browsing,” where the AI takes actions on the user’s behalf, researching, booking travel, managing emails, or shopping online. However, this functionality isn’t entirely new. Large language models like ChatGPT and Perplexity’s own AI have long used background, headless browsers to fetch web data. These new tools simply make that process visible to the user. For many, watching an AI click through web pages may be less appealing than the promise of a completed task, raising the question of whether this visibility adds real value or just visualizes the “sausage-making.”
The rise of agentic browsing introduces significant challenges for businesses and security professionals. When an AI agent visits a website via Comet or Atlas, it likely appears in analytics as ordinary human traffic from a Chromium browser. This obscurity undermines reliable metrics for marketers and opens the door to sophisticated advertising fraud, where ads are served to non-human agents at immense scale. The security implications are even more severe. Research firms like Gartner have advised blocking AI browsers due to risks like prompt injection attacks. Vulnerabilities dubbed “CometJacking” demonstrate how a single malicious link could commandeer an AI browser to steal all accessible user data, bypassing traditional credential theft entirely.
This does not mean agentic browsers are without utility. Niche professional applications could provide genuine value. Development and QA teams could use them to automate user journey testing at scale. SEO specialists might leverage them to analyze how AI agents interpret site structure and content, offering insights for optimization. Technical users may find them powerful for automating repetitive data aggregation or monitoring tasks. Ironically, these practical use cases cater to a specialized audience, not the mass consumer base whose behavioral data is crucial for training better AI models.
For Atlas and Comet to truly compete with entrenched incumbents, they must offer a compelling, secure value proposition to everyday users, a tall order as Chrome, Edge, and Safari rapidly integrate their own AI features. Google and Microsoft possess not only massive existing user bases but also their own advanced LLMs, giving them a formidable dual advantage. Even Firefox’s approach of offering user-controlled AI features could reveal what consumers actually want, rather than what companies wish to push.
The so-called AI browser war is in its earliest stages, and it will be a marathon, not a sprint. Victory will not come from flashy demos or lofty predictions about cognitive operating systems. The ultimate winners will be those that seamlessly integrate useful, trustworthy AI into workflows so effectively that the technology itself becomes invisible. Whether that future is shaped by Atlas, Comet, or the evolving giants of today remains to be seen.
(Source: Search Engine Journal)
