Summary of Key Points
OpenAI’s AI-native browser, Atlas, was shut down less than 10 months after its launch due to the high costs associated with developing an independent AI-native browser (including engineering maintenance and security measures) and limited commercial returns. However, its core capabilities (such as cross-site task execution) have been integrated into existing products like ChatGPT. The current industry trend is that large companies (such as Google and Microsoft) are choosing to embed AI within their existing browsers, while smaller players without a significant presence in the market (such as Meituan’s Tabbit) are focusing on specific use cases to create “secondary browsers.” The survival space for independent AI-native browsers has shrunk, but they have not completely disappeared.
1. Atlas’s Failure: Excessive Expenses vs. Meager Profits
The closure of Atlas was essentially due to an imbalance between costs and benefits:
- Exorbitant Costs: Atlas did not simply use the Chromium browser framework; it developed its own OWL architecture, which meant building both a browser engine and an AI assistant platform from scratch. Maintaining Chromium itself is like trying to keep up with a high-speed train—fast iteration and large codebase mean that a few engineers can handle AI plugins, but maintaining an independent browser requires ten to hundred times more effort. Additionally, security risks were a major concern: since AI could manipulate user web accounts, hackers could embed malicious code to trick the system into executing harmful actions, creating an endless cycle of defense efforts.
- Meager Revenue: Atlas was only available for Apple devices, with no Windows version released. Its AI features required a ChatGPT subscription, meaning it did not generate direct profits; it relied on directing users to other services and collecting data. The user experience was also unreliable—tasks like placing orders on Walmart would often fail, and adding items to an Amazon cart was slower than doing it manually. The additional revenue generated was not enough to cover the costs, leading OpenAI to decide to stop developing Atlas.
2. AI Browsers Are Not Dead; They’ve Just Moved to Other Platforms
The exit of Atlas does not mean that AI browsers are a failed concept. Instead, OpenAI has chosen to integrate these features into existing products:
- Core Capabilities Retained: Functions like cross-site task execution (e.g., making purchases) have been incorporated into ChatGPT’s desktop and web-based versions, as well as Chrome extensions. Advanced website logging-in features and cross-tab interaction have been transferred to Chrome’s Side Chat extension.
- Memory Function Reduced: Atlas would automatically remember users’ browsing history; now, users must explicitly tell ChatGPT what to remember. However, industry experts argue that memory is not the main barrier to replication, as capabilities like Chrome and Gemini can also retrieve this information from user history. The real challenge lies in the “execution environment” (user identity, login status, permissions, operation auditing), which requires a different browser and poses a high barrier for adoption.
3. Independent AI-Native Browsers: A Narrow Path Forward
For smaller players without a significant market presence (such as Meituan’s Tabbit), an independent approach is still possible, but their focus has shifted from trying to replace Chrome to becoming a “secondary browser” for specific use cases:
- Why Is There Still a Chance? Traditional browser extensions have limited permissions (e.g., they cannot freely manipulate tabs or manage user space). AI-native browsers, on the other hand, are designed with both humans and AI in mind, allowing for more advanced interactions. Although large companies can grant more permissions to their browsers, traditional browsers are designed around human needs, making significant reengineering difficult (for example, Google is hesitant to make drastic changes to Chrome to prevent AI from reducing ad clicks).
- Where Lies the Narrow Path? Independent browsers have a chance in niche scenarios—e.g., dedicated work browsers (helping with meeting notes or document management), enterprise security browsers (isolating sensitive operations), and task-specific browsers (for shopping or booking). Users are willing to pay for such specialized tools, giving smaller players a foothold in the market.
4. Why Don’t Large Companies Develop Independent AI Browsers?
Large companies make practical choices:
- Existing Platforms Are More Cost-Effective: Google has Chrome, and Microsoft has Edge; adding AI to these browsers is much cheaper than developing a new one. Anthropic, for example, does not have its own browser but develops extensions for Chrome, directly integrating into users’ existing work environments.
- Commercial Considerations: Google’s main revenue comes from search ads. If AI can complete tasks on behalf of users (e.g., providing direct answers instead of showing search results), ad clicks will decrease. Therefore, companies are cautious about making significant changes to their browsers that could impact their advertising revenue.
5. The Perplexing Issue of Permissions
The slow adoption of AI browsers is largely due to permission-related issues:
- Too Little Permission Is Ineffective: If AI only offers suggestions without the ability to perform actions (e.g., filling out forms or placing orders), users find it less useful than manual processes.
- Too Much Permission Is Dangerous: If AI has full access to user accounts and files, it can lead to serious security risks if hacked.
This contradiction between security and functionality is a major obstacle to the widespread adoption of AI browsers. Users want browsers that are both secure and easy to use, but current technology and trust mechanisms have not yet resolved this issue.
Smaller players like Tabbit are betting on specific scenarios where users are willing to grant more permissions to AI for enhanced security and efficiency. If these niche markets fail to take off, independent AI-native browsers may truly disappear.