AI Publications

🔗Who Wins The Browser Wars? - by Evan Armstrong

1. Introduction

The article opens by forecasting a coming “tech war” that will reshape how humans interact with digital tools, particularly through browsers enhanced by artificial intelligence. Armstrong introduces two such AI-powered browsers he tested: Dia from The Browser Company and Comet from Perplexity. These represent a new class of browsers that aim to not just serve as access points to the internet, but as intelligent agents capable of doing tasks for users.

Drawing a parallel with the past, Armstrong recalls how Google Chrome overtook Internet Explorer by optimizing for ad revenue—prioritizing speed, user volume, and data collection. But in the new AI age, advertising alone is insufficient. The competition has shifted from grabbing attention to owning user cognition and workflows. Armstrong hints that OpenAI may soon release its own browser, signaling intensifying competition.

He emphasizes that his analysis is unbiased, though he was given early access by both companies. This sets the stage for a deep exploration of how AI is redefining what browsers are and what it means to “win” in this new technological era.

2. Vectors of Competition

Armstrong describes how traditional browser competition (based on page load speed) has evolved. Now, “speed to complete a task” is the new battlefield. He identifies two key vectors of competition in the AI browser space:

2.1. “Do stuff for” (Application Agents)

AI browsers like Comet can take actions across different applications on behalf of the user. These include tasks such as replying to emails or managing calendars. For example, one could instruct Comet to plan a date night—buy movie tickets, find a restaurant, make a reservation—without performing those steps manually. This model positions the browser as a proxy user, operating inside application-specific workflows.

However, current implementations are limited. Despite impressive moments, Armstrong notes that success rates are inconsistent. AI agents fail often, especially in complex or ambiguous scenarios. This is due to the massive complexity of handling interactions across millions of varied and ever-changing websites.

2.2. “Do stuff with” (Multi-Tab Agents)

This second vector refers to helping users synthesize and interact with content across tabs—such as during research, writing, or learning. Armstrong explains how Dia excels here, allowing him to chat with his notes, essay outlines, and academic texts all in one place. This dramatically accelerates research and comprehension, such as comparing his own arguments with Heidegger’s philosophical writings.

Despite this utility, Armstrong notes significant UX challenges—like cluttered interfaces with overlapping AI chat windows—and technical limitations of current LLMs. For instance, LLMs often make subtle but important errors when interpreting complex philosophical material. This reflects the broader limits of current AI models, which both Dia and Comet rely on.

Additionally, Armstrong underscores why AI agents have seen success mainly in software development: clear feedback loops, rich training data (e.g., GitHub), and straightforward ROI. In contrast, building an agent that navigates the broader internet is exponentially more difficult and expensive.

Still, he notes that even when the agents partially succeed, they deliver a sense of "magic." The bet these companies are making is that LLMs will catch up to the task capabilities that browsers are already being designed for.

3. Implications and Incentives

The final section addresses the business and strategic implications of AI browsers. Armstrong points out that companies like The Browser Company and Perplexity are currently funded by venture capital, operating at significant losses (e.g., Perplexity had -$69M in net income). Their success will depend on monetizing either through subscription models or, less likely in the short term, advertising.

3.1. Monetization Possibilities

• Subscriptions: Already in play—Comet is part of Perplexity’s $200/month Pro Tier. Users may be willing to pay if the browser saves significant time and effort.

• Advertising: More uncertain. Traditional ad models depend on user clicks, but AI agents may bypass or automate user interaction. It’s unclear what agent-targeted ads might look like (if they’re possible at all). This uncertainty, coupled with the cost of AI inference (i.e., GPU time), points toward subscriptions as the near-term path to revenue.
  

3.2. Why the Browser is the New Battleground

Browsers are not just gateways; they’re data collectors. The more people use these AI browsers, the more companies can gather behavioral data, which includes:

• User history (frequent sites, visit times, duration)

• User profiles (interests, job, preferred communication)

• Common prompts or “skills” (frequent commands that inform future features)
  

These data sets will be critical for training and fine-tuning future models, potentially allowing companies to sidestep the massive costs of building foundational models from scratch. Essentially, whoever controls the browser and its AI agent may eventually control the entire user interaction layer of the internet.

4. Conclusion

AI browsers may render current software tools obsolete. If AI agents can handle end-to-end workflows, applications become less important than the portal used to access them. Browsers will be that portal. The company that owns this interface—and the agent behind it—could own the future of digital labor.