Cursor, the company that provides an artificial intelligence-integrated development environment, recently gained attention from the industry after suggesting that it had developed a fully functional browser using its own artificial intelligence agents, which is known as the Cursor AI-based development environment. In a series of public statements made by Cursor chief executive Michael Truell, it was claimed that the browser was built with the use of GPT-5.2 within the Cursor platform.
Approximately three million lines of code are spread throughout thousands of files in Truell's project, and there is a custom rendering engine in Rust developed from scratch to implement this project.
Moreover, he explained that the system also supports the main features of the browser, including HTML parsing, CSS cascading and layout, text shaping, painting, and a custom-built JavaScript virtual machine that is responsible for the rendering of HTML on the browser.
Even though the statements did not explicitly assert that a substantial amount of human involvement was not involved with the creation of the browser, they have sparked a heated debate within the software development community about whether or not the majority of the work is truly attributed to autonomous AI systems, and whether or not these claims should be interpreted in light of the growing popularity of AI-based software development in recent years.
There are a couple of things to note about the episode: it unfolds against a backdrop of intensifying optimism regarding generative AI, an optimism that has inspired unprecedented investment in companies across a variety of industries. In spite of the optimism, a more sobering reality is beginning to emerge in the process.
A McKinsey study indicates that despite the fact that roughly 80 percent of companies report having adopted the most advanced AI tools, a similar percentage has seen little to no improvement in either revenue growth or profitability.
In general, general-purpose AI applications are able to improve individual productivity, but they have rarely been able to translate their incremental time savings into tangible financial results.
While higher value, domain-specific applications continue to stall in the experimental or pilot stage, analysts increasingly describe this disconnect as the generative AI value paradox since higher-value, domain-specific applications tend to stall in the experimental or pilot stages.
There has been a significant increase in tension with the advent of so-called agentic artificial intelligence, which essentially is an autonomous system that is capable of planning, deciding, and acting independently in order to achieve predefined objectives.
It is important to note, however, that these kinds of systems offer a range of benefits beyond assistive tools, as well as increasing the stakes for credibility and transparency in the case of Cursor's browser project, in which the decision to make its code publicly available was crucial.
Developers who examined the repository found the software frequently failed to compile, rarely ran as advertised, and rarely exceeded the capabilities implied by the product's advertising despite enthusiastic headlines.
If one inspects and tests the underlying code closely, it becomes evident that the marketing claims are not in line with the actual code. Ironically, most developers found the accompanying technical document—which detailed the project's limitations and partial successes—to be more convincing than the original announcement of the project.
During a period of about a week, Cursor admits that it deployed hundreds of GPT-5.2-style agents, which generated about three million lines of code, assembling what on the surface amounted to a partially functional browser prototype.
Several million dollars at prevailing prices for frontier AI models is the cost of the experiment, as estimated by Perplexity, an AI-driven search and analysis platform. At such times, it would be possible to consume between 10 and 20 trillion tokens during the experiment, which would translate into a cost of several million dollars at the current price.
Although such figures demonstrate the ambition of the effort, they also emphasize the skepticism that exists within the industry at the moment: scale alone does not equate to sustained value or technical maturity.
It can be argued that a number of converging forces are driving AI companies to increasingly target the web browser itself, rather than focusing on plug-ins or standalone applications.
For many years, browsers have served as the most valuable source of behavioral data - and, by extension, an excellent source of ad revenue - and this has been true for decades. They have been able to capture search queries, clicks, and browsing patterns for a number of years, which have paved the way for highly profitable ad targeting systems.
Google has gained its position as the world's most powerful search engine by largely following this model.
The browser provides AI providers with direct access to this stream of data exhaust, which reduces the dependency on third party platforms and secures a privileged position in the advertising value chain.
A number of analysts note that controlling the browser can also be a means of anchoring a company's search product and the commercial benefits that follow from it as well.
It has been reported that OpenAI's upcoming browser is explicitly intended to collect information on users' web behavior from first-party sources, a strategy intended to challenge Google's ad-driven ecosystem.
Insiders who have been contacted by the report suggest they were motivated to build a browser rather than an extension for Chrome or Edge because they wanted more control over their data.
In addition to advertising, the continuous feedback loop that users create through their actions provides another advantage: each scroll, click, and query can be used to refine and personalize AI models, which in turn strengthens a product over time.
In the meantime, advertising remains one of the few scalable monetization paths for consumer-facing artificial intelligence, and both OpenAI and Perplexity appear to be positioning their browsers accordingly, as highlighted by recent hirings and the quiet development of ad-based services.
Meanwhile, AI companies claim that browsers offer the chance to fundamentally rethink the user experience of the web, arguing that it can be remodeled in the future. Traditional browsing, which relied heavily on tabs, links, and manual comparison, has become increasingly viewed as an inefficient and cognitively fragmented activity.
By replacing navigation-heavy workflows with conversational, context-aware interactions, artificial intelligence-first browsers aim to create a new type of browsing.
It is believed that Perplexity's Comet browser, which is positioned as an “intelligent interface”, can be accessed by the user at any moment, enabling the artificial intelligence to research, summarize, and synthesize information in real time, thus creating a real-time “intelligent interface.”
Rather than clicking through multiple pages, complex tasks are condensed into seamless interactions that maintain context across every step by reducing the number of pages needed to complete them.
As with OpenAI's planned browser, it is likely to follow a similar approach by integrating a ChatGPT-like assistant directly into the browsing environment, allowing users to act on information without leaving the page.
The browser is considered to be a constant co-pilot, one that will be able to draft messages, summarise content, or perform transactions on the user's behalf, rather than just performing searches. These shifts have been described by some as a shift from search to cognition.
The companies who are deeply integrating artificial intelligence into everyday browsing hope that, in addition to improving convenience, they will be able to keep their users engaged in their ecosystems for longer periods of time, strengthening their brand recognition and boosting habitual usage. Having a proprietary browser also enables the integration of artificial intelligence services and agent-based systems that are difficult to deliver using third-party platforms.
A comprehensive understanding of browser architecture provides companies with the opportunity to embed language models, plugins, and autonomous agents at a foundational level of the browser. OpenAI's browser, for instance, is expected to be integrated directly with the company's emerging agent platform, enabling software capable of navigating websites, completing forms, and performing multi-step actions on its own.
It is apparent that further ambitions are evident elsewhere too:
The Browser Company's Dia features an AI assistant right in the address bar, offering a combination of search and chat functionality along with task automation, while maintaining awareness of the context of the user across multiple tabs.
These types of browsers are an indicator of a broader trend toward building browsers around artificial intelligence rather than adding artificial intelligence features to existing browsers.
By following such a method, a company's AI services become the default experience for users whenever they search or interact with the web. This ensures that the company's AI services are not optional enhancements, but rather the default experience.
Last but not least, competitive pressure is a serious issue.
Search and browser dominance by Google have long been mutually reinforcing each other, channeling data and traffic through Chrome into the company's advertising empire in an effort to consolidate its position.
A direct threat to this structure is the development of AI first browsers, whose aim is to divert users away from traditional search and towards AI-mediated discovery as a result.
The browser that Perplexity is creating is part of a broader effort to compete with Google in search. However, Reuters reports that OpenAI is intensifying its rivalry with Google by moving into browsers. The ability to control the browser allows AI companies to intercept user intent at an earlier stage, so that they are not dependent on existing platforms and are protected from future changes in default settings and access rules that may be implemented.
Furthermore, the smaller AI players must also be prepared to defend themselves from the growing integration of artificial intelligence into their browsers, as Google, Microsoft, and others are rapidly integrating it into their own browsers.
In a world where browsers remain a crucial part of our everyday lives as well as work, the race to integrate artificial intelligence into these interfaces is becoming increasingly important, and many observers are already beginning to describe this conflict as the beginning of a new era in browsers driven by artificial intelligence.
In the context of the Cursor episode and the trend toward AI-first browsers, it is imperative to note a cautionary mark for an industry rushing ahead of its own trials and errors. It is important to recognize, however, that open repositories and independent scrutiny continue to be the ultimate arbiters of technical reality, regardless of the public claims of autonomy and scale.
It is becoming increasingly apparent that a number of companies are repositioning the browser as a strategic battleground, promising efficiency, personalization, and control - and that developers, enterprises, and users are being urged to separate ambition from implementation in real life.
Among analysts, it appears that AI-powered browsers will not fail, but rather that their impact will be less dependent on headline-grabbing demonstrations than on evidence-based reliability, transparent attribution of human work to machine work, and a thoughtful evaluation of security and economic trade-offs. During this period of speed and spectacle in an industry that is known for its speed and spectacle, it may yet be the scariest resource of all.