当前 AI 浏览器（如 Atlas、Comet）仍无法取代传统浏览器，测试中对基本任务表现不稳，且在多步骤指令上常出现延迟、循环或改以截图辨识等低效替代方案。AI 功能多被锁在付费层级，与传统浏览器不同的互动模式正在将网路分成面向人类与面向代理的两种路径。Mozilla 去年对 1,000 名美国用户与 400 名 Firefox 用户调查显示，60% 只愿在「低风险」情境使用生成式 AI，而时间节省会推动 AI 使用，但准确度需求迫使用户同时使用搜寻与 AI。
使用行为统计呈现显著变化：Comet 用户平均提问量为一般 Perplexity 聊天机器人的 6 至 18 倍；YouTube 影片摘要成为常见用途。行动端已占超过 67% 网路流量，使桌机浏览器缺乏快速演化压力。AI 浏览器遇到以视觉导向或叙事式结构建置的网站时最易失效，因代理需要高度结构化、可索引的资料；多家业者预期网站最终需分为面向人类与面向代理的双版本。
硬体与供应端摩擦加剧：Windows 测试中高阶代理功能提升 CPU 负载；Chrome 与 Microsoft 正量产自家代理工具以平衡竞争。大型网路服务仍不愿为 AI 代理打造专属版本，且多家公司（如 Reddit、Amazon、New York Times）采取法律行动反对资料再利用。信任度成为普及最大障碍：用户质疑代理是否会基于商业诱因而非最佳利益执行付款与预订等任务，使 AI 浏览器仍处于早期、未成熟阶段。
Current AI browsers such as Atlas and Comet remain far from replacing legacy browsers, often failing basic tasks and stalling on multi-step operations by defaulting to slow workarounds like screenshot-based recognition. Their most advanced features reside behind paid tiers, and their interaction models split the internet into human-oriented and agent-oriented paths. A Mozilla survey of 1,000 US users and 400 Firefox users found 60% were comfortable using generative AI only for low-stakes tasks; time savings encourage AI adoption, but accuracy needs drive users to combine search with AI.
Usage statistics show sharp behavioral shifts: Comet users ask 6–18 times more questions than users of Perplexity’s standard chatbot; summarizing long YouTube videos is a frequent case. With over 67% of web traffic now from mobile, desktop browsers have faced limited pressure to evolve. AI browsers fail most on visually heavy or narrative sites because agents require highly structured, indexable data; developers increasingly expect sites to split into human and agent versions.
Hardware strain and ecosystem friction persist: advanced agent features raised CPU load in Windows tests; Chrome and Microsoft are scaling their own agents to compete. Major online providers resist building AI-specific versions and several companies (including Reddit, Amazon, and the New York Times) have taken legal action over data use. Trust remains the primary barrier: users question whether agents making purchases or bookings act in their best interest rather than responding to hidden incentives, leaving AI browsers an early, still-immature category.