The Agentic Browser Wars: Operator vs Comet vs Dia vs Arc Browse
Six browsers launched agent capabilities in the last 12 months. Operator, Comet, Dia, Arc Browse, Brave Leo, and Chrome extensions. Most will lose. Here's what's actually different about each, who they're for, and what to watch for the category to settle.
In the 18 months since OpenAI launched Operator, six browsers have shipped agent capabilities of meaningful quality. Operator (OpenAI), Comet (Perplexity), Dia (The Browser Company, Arc's successor), Arc Browse (legacy Arc with agent extensions), Brave Leo (privacy-focused), and a growing ecosystem of Chrome agent extensions led by Anthropic's Computer Use. The arms race is louder than the smartphone wars of 2010, and the assumption inside each company is that "the browser that browses for you" will be one of the most valuable consumer products of the next decade.
Most of these browsers will lose. The category will likely settle to two or three winners by 2028, and the dynamics of which ones win are not what the current narrative suggests. Here's a clear read on what's actually different about each, who they're for, and what to watch for the category to consolidate.
What's different about each
Operator (OpenAI) is the most-used agentic browser by approximately 4x according to third-party tracking. It uses vision-grounded execution — taking screenshots, reasoning about what's on screen, and clicking based on visual understanding. This is slower than DOM-scraping but vastly more robust across sites. Operator's strength is transactional tasks: booking restaurants, ordering food, comparison shopping. Its weakness is anything requiring complex judgment or that crosses friction barriers like captchas. The product feels engineered for the use case OpenAI sees as the highest-volume, which is consumer transactions.
Comet (Perplexity) is the second-largest by usage and the most thoughtful for research-heavy workflows. Comet was built on top of Perplexity's existing search and answer infrastructure, which means it's exceptional at synthesizing information across multiple tabs and websites into a single coherent answer. Where Operator excels at "do this for me," Comet excels at "understand all of this and tell me what to do." The agent mode in Comet is real but less developed than Operator's; the product's emphasis is research-as-browsing, not browsing-as-execution.
Dia (The Browser Company) is Arc's successor and the most consumer-design-conscious of the major options. Dia's agent is built into the address bar — you type a request and it interprets whether you want to navigate, search, or have an agent complete a task. The UI is the most polished consumer browser experience, and the company's design pedigree shows. The agent capability itself is meaningfully behind Operator and Comet on task completion benchmarks, but the user experience for the tasks Dia can handle is better. It's the right browser for people who want the AI-first browser experience without the technical feel of Operator.
Arc Browse is what users still call the legacy Arc browser, which The Browser Company is winding down in favor of Dia. Arc Browse exists primarily because of its loyal user base who object to the Dia direction. Agent capabilities are limited to extensions. The browser is unlikely to be a long-term competitor in the agentic category but maintains relevance through its devoted niche.
Brave Leo is the privacy-focused option, built on top of Brave's browser and using a combination of Brave's local AI infrastructure and optional cloud LLM integration. Leo's agent capabilities are competent but understated, and the product positioning is explicitly privacy-first. Brave's user base is small relative to the other options, but high-quality, and the privacy positioning resonates with a specific demographic.
Chrome agent extensions are a distributed ecosystem of agentic capabilities built on top of vanilla Chrome through extension APIs. Anthropic's Computer Use, several open-source projects, and a growing number of vertical-specific agents (for sales prospecting, recruiting, e-commerce research) live in this layer. The strength of the extension approach is that users keep their existing browser, bookmarks, and workflow; the weakness is that no single extension can match the integrated experience of Operator or Comet.
Who's actually using what
Adoption patterns by use case have begun to stabilize. Operator dominates transactional use cases: roughly 60-70% of users who run agent-completed bookings, orders, or form submissions are using Operator. Comet dominates research: roughly half of users who run multi-source research workflows are using Comet. Dia has the strongest consumer-first usage in design and marketing communities where the polished UI matters more than the deepest agent capability. The verticals beyond consumer have started forming around the use case rather than the browser: sales teams adopting Operator for prospecting research, support teams using extensions for ticket triage, procurement teams using Comet for vendor evaluation.
Newsletter
Like this decoded read? Get the next one in your inbox.
One sharp take on the latest in AI for PMs, founders and operators. Every few days. Free.
Free forever. Unsubscribe anytime. No spam.
The geographic split is also worth noting. Operator and Comet skew heavily US/EU. Dia has surprisingly strong India and Southeast Asia adoption among tech-forward consumers. Brave Leo is strongest in Europe, where privacy concerns translate to actual product preference more than they do in the US.
What's holding the category back
Three structural blockers remain across all agentic browsers. First, trust and authentication. Users are willing to let agents browse but reluctant to let them log in to authenticated services. Most production agent workflows still require humans to handle the actual login step before the agent takes over. The trust gap will close only as identity standards (like passkeys, agent-specific OAuth scopes) emerge.
Second, payment friction. Agents are good at navigating to a checkout page and reasonably good at filling card details. They are bad at handling 3D Secure verification, OTP flows, and the cascade of bank-specific friction that Indian and European users encounter every transaction. Until payment standards explicitly support agentic flows (which UCP and similar protocols are working toward), conversion through agents will materially trail conversion through humans.
Third, the long tail of broken sites. The sites agents perform best on are well-structured, accessible, semantically clean sites. The Indian web in particular has a long tail of poorly-built sites that work fine for humans but break agents in predictable ways. This is partly a tooling problem (agents need better recovery from unexpected state) and partly a sites problem (sites need to become more agent-friendly, which is a real engineering investment).
What to watch in the next 18 months
Apple's WWDC 2026 is the biggest single signal. If Apple announces meaningful agentic capabilities in Safari, the consumer browser category resets. If they don't, Operator and Comet continue eating share.
UCP (Universal Commerce Protocol) adoption matters for transactional agents. Google announced UCP at I/O 2025; the open question in 2026 is whether Amazon, Shopify, and the major Indian marketplaces (Flipkart, Meesho) adopt it. If they do, agent-driven commerce becomes meaningfully more reliable. If they don't, transactional agents stay limited to a smaller set of merchants.
Microsoft's Bing agent has been quiet but is reportedly under heavy investment. Microsoft's distribution advantage through Windows and Edge could matter more than product quality if they ship a competent agent.
Indian browser adoption is undercovered. India is the largest Chromium user base outside the US, and Indian consumer behavior around agent-mediated commerce (especially through WhatsApp-integrated workflows) could become its own category. Watch for Indian fintech and e-commerce companies announcing agent partnerships in 2026-27.
What this means for PMs
The category is real, the timing is faster than most product organizations are prepared for, and the second-order effects are larger than the first-order ones.
The first-order question — "should our product support agents?" — is increasingly straightforward. If your product is conversion-driven (e-commerce, booking, signup flows, SaaS trials), yes. The audit involves clean DOM, accessible markup, structured product schema, API parity for UI actions, and absence of agent-hostile patterns like required popups or aggressive captchas.
The second-order question is more interesting. When agent-mediated traffic becomes a meaningful share of your visitors, your conversion funnel becomes a different surface. You're optimizing for the human user's intent as interpreted by an agent, not the human user directly. This means your product copy, your category positioning, your structured data, and your error-handling all matter in new ways. Sites that have been quietly investing in AEO (Answer Engine Optimization) since 2024 will have a multi-year head start on sites that have not.
The third-order question is structural. If agents become the new front door for consumer transactions, the position of platform aggregators (Amazon, Flipkart, Booking, Uber) changes. Some aggregators will use agents to deepen their lock-in (Amazon controlling the agent layer too). Others will see agents commoditize their position (an agent that can compare across aggregators reduces aggregator pricing power). Either way, the next five years will see meaningful shifts in who captures value in transaction flows, and the agentic browser category is the leading edge of that shift.
The category is settling, but it isn't settled. Operator's lead is real but not unassailable. Comet's research strength is differentiated but narrower than the consumer mass-market opportunity. Dia is design-forward but capability-behind. The winner will likely be the company that solves trust and payment friction first, not the company with the smartest agent today. Whoever that is, it will reshape how Indian and global PMs think about web product design for the next decade.
Frequently asked
Which agentic browser is best in 2026?
Depends on the task. Operator (OpenAI) is best for transactional tasks like booking, ordering, and form filling because of its vision-grounded execution. Comet (Perplexity) is best for research and synthesis because it builds on Perplexity's search foundations. Dia (The Browser Company) is the most consumer-friendly UI but has the smallest agent capability of the major players. For builders, Operator has the cleanest developer story; for end users, Comet has the better browsing-with-agent experience.
What is OpenAI Operator?
Operator is OpenAI's browser-based agent that completes multi-step web tasks on behalf of users. Launched in early 2025, it became OpenAI's primary agentic surface by 2026. Operator uses vision-grounded reasoning (taking screenshots and reasoning about what's on screen) rather than DOM-scraping, which makes it more robust across sites but slower per step than purely text-based agents.
What is Perplexity Comet?
Comet is Perplexity's web browser, launched in 2025, that bundles Perplexity's AI search and answer capabilities directly into the browsing experience. It includes an agent mode that can complete multi-step tasks but its design emphasis is research and synthesis rather than transactional execution. Comet competes with Arc Browse and Dia as a consumer-AI-first browser.
Will Apple build an agentic browser?
Apple has not announced an agentic Safari product as of May 2026, but WWDC 2026 is widely expected to be where Apple shows its hand on agentic browsing. The likely path is some combination of Apple Intelligence (limited because it's behind), partnership with a third-party LLM (which Apple has historically resisted), or quiet integration of agentic capabilities into Safari extensions. The competitive pressure from Operator and Comet is making the hold-out position untenable.
Should my product be agent-friendly?
If your product depends on a user completing a multi-step flow (booking, checkout, signup, configuration), yes — within 18 months a meaningful share of your traffic will be agent-driven. The audit involves five things: clean DOM with semantic HTML, accessible ARIA labels, structured data (schema.org), API parity for every UI action, and absence of dark patterns like required popups or aggressive captchas. Sites that fail these checks will lose share to competitors that don't.