Browser Evolution Trends Seen from Arc

Arc is releasing “II”

Josh Miller announced last week that The Browser Company will launch a brand-new browser next year, while the existing Arc browser will continue to be maintained as a stable version. Arc has experienced rapid growth over the past few years, but it’s clear that Arc cannot become a mainstream browser.

This decision was driven by three types of users Arc faces. One group are power users who are particularly bought into the product; they also prefer Lego-style, well-designed tools like Notion and Obsidian. Another group are mainstream users: Arc conveys interesting ideas, but it’s still somewhat complex to use well, with a high cold-start cost—its incremental value often doesn’t justify the cost of switching browsers. The final stakeholder is The Browser Company itself, which doesn’t see Arc as the final form of a browser. To satisfy all three groups, launching a new browser is wiser than trying to morph Arc into something else.

What Arc taught us about innovation and growth

• After defining STP, an effective outreach method is to identify and penetrate specific communities—this is often more effective than targeting based on user personas. Even if personas match your STP, users targeted that way aren’t connected to one another, whereas people within a community are inherently tightly linked. Arc engaged with specific communities during development, interacting with potential users, and after launch used a clever invite-and-referral mechanism to achieve viral growth among precise user segments. Our vision is to help millions of merchants run their businesses well. Rather than saying “we want to reach small businesses with a sense of operating awareness,” it’s clearer to say “we want to penetrate a forum where restaurant owners discuss making money.”

• Arc is an excellent example of the “faster horse or car” puzzle: if you ask the market what they expect from a browser, almost no one will describe something like today’s Arc.

• Arc internally promotes a culture based on hypothesis testing rather than A/B testing, with strong emphasis on prototyping and experiment-driven work. Arc has no PRD process and no formal product managers. If someone has an initial idea, it isn’t discussed ad nauseam; they prototype and ship quickly within 48 hours to validate their intuition.

• Arc maintains steady, continuous user relationships through two distinctive teams: the Membership Team and the Storytelling Team (internally they never use the word “user”). The Membership Team handles service from a user’s first encounter with the product through their entire lifecycle; the Storytelling Team focuses on reaching people who haven’t yet had a chance to be served, marketing by telling their stories. A small detail: Arc ships new features almost weekly; the changelog is edited with Arc’s Easel feature, and each feature has a person to explain and recommend it—often the feature’s developer.

• Culturally, Arc is anti–Silicon Valley and can also be seen as anti–big tech. “Silicon Valley” here refers to a place optimized for efficiency, productivity, and profit, but with less warmth, soul, or feeling. I’ll paste a passage from Josh Miller on a podcast so you can get a sense of it.

  When we step back and think about it, we believe that feeling may be the most important thing in the world.

  Allow us a moment of imagination: picture your favorite neighborhood restaurant. Ours is a corner spot in Brooklyn’s Boerum Hill. It has abundant natural light, handmade woven seat cushions, caramel wood grain, and colorful decorations hanging from the ceiling. Can you picture your place? Can you feel its warmth and spirit?

  A Silicon Valley optimizer might say, “Well, their coffee isn’t brewed at exactly 200 degrees. The seating looks a bit shabby. The ceiling decorations serve no functional purpose.”

  But we think that’s precisely the point. Those small, handcrafted touches give our environment humanity and spirit. Without them, we’re left with something generic but utterly monotonous—a space that may “perfectly” meet our functional needs while neglecting our emotional needs.

Speculation on the next-generation browser evolution

We can divide Web applications into three layers. The content layer corresponds to various sites. The navigation layer evolved from portals in the ’90s to today’s search engines. The control layer is the browser.

The rapid development of LLMs will change many rules of the game. Last week OpenAI’s launch of its own GPT search touched Google’s core interests. Remember, Google pays Apple about $20 billion a year to remain Safari’s default search engine. But the web’s control layer has not yet been fully challenged; the next-generation browser evolution will unfold under the game-changing influence of large models. The largest incumbents in this layer remain Google: there are about 5.3 billion internet users worldwide, and Chrome’s daily active users exceed 3 billion—note, daily active users. BTW, Google in this wave of technological change feels like a “Janus-faced” actor; many intersections between old and new occur there, so following and studying Google will be interesting.

Back to what the next-generation browser might look like: one hypothesis is an AI-centered Web application platform where AI runs on-device—an “action intelligence.” There are two bases for this. First, GitHub’s recently published Octoverse 2024 shows a trend toward smaller on-device models for LLMs. On-device AI will fundamentally reshape the competitive landscape of technology, with the key element being AI agents. Second, endpoints will shift from mere content generation to completing concrete user tasks—that is, from “content intelligence” to “action intelligence.” Early signs of this appear in Claude’s released “computer use” features and several recently launched domestic OSs; for example, Honor’s Magic OS can place an order for Luckin Coffee on Meituan with a single sentence.