Agentic browsers shift power: Comet goes free, Neon charges

The moment the browser learned to act

In October 2025, Perplexity removed the paywall from its Comet browser and effectively handed an agentic web to anyone willing to try it. For months, Comet lived behind a premium tier and an invite list. Going free turns a niche experiment into a mainstream distribution play. It is the clearest signal yet that the next battlefield is not answers. It is actions.

Comet is not another sidebar chatbot. It reads, plans, and clicks. It fills forms, compares products, and completes bookings in the middle of your browsing flow. Perplexity paired the shift with pragmatic growth levers. A referral program pays cash bounties per activated install, with payouts that vary by country. A set of commerce integrations means checkout can happen in the chat or on the page, so the path from intent to purchase becomes one continuous motion. When a browser can execute, distribution shifts. Traffic matters less than completion.

Opera’s Neon launched into the same weather system with a different bet. Neon’s early access is paid. The company is charging a monthly fee to try agentic browsing that includes isolated workspaces, reusable prompt blocks, and a local navigation agent that can click through sites and forms. In other words, two strategies are visible. Perplexity is racing for ubiquity through a free core and referral economics. Opera is testing whether a premium product can monetize power users before agents become a commodity. The contrast tells you what phase we are in. The platform is not settled, and pricing is a lever as much as a feature.

From search to action

For two decades, search engines were the front door to the internet. You typed, clicked a result, and repeated. Ads and search engine optimization sat on top of that loop. Agentic browsers reroute this loop. The browser becomes the front door and the concierge. You still express intent, but now the software follows through. It does not hand you ten blue links. It drafts a multistep plan, tries steps, checks its work, and asks for confirmation only when needed.

Consider a concrete example. You ask your browser to find a cabin near Big Bear for the first weekend in December, under a set budget, with a pet fee cap and a flexible cancellation policy. A conventional search returns listings and sponsored links. An agentic browser opens two or three sites, applies filters, fills dates, expands the map, and exports a shortlist with total prices that include fees. You approve one. The browser completes the booking and adds the calendar event. There is no search results page to scroll, no ad slot to bid on, and no tab hopping. The entire journey begins and ends in the browser agent.

This is a distribution change. Whoever owns the browser runtime that can perceive the page, reason over it, and act on it, owns the path to conversion. That runtime is what Comet, Neon, and a handful of startup browsers are racing to define. We have seen a similar control shift when publishers reclaimed some search surface with initiatives like publisher owned AI search. Browsers are now taking that logic one step further by collapsing intent, discovery, and completion into a single agentic loop inside the client.

The economics shift underneath

Two new rails are visible inside this distribution shift.

First, checkout. Perplexity spent 2025 stitching payments into its agent experience so a user could buy with a familiar wallet without context switching. The agent can collect shipping, surface total price, and confirm the purchase in a single flow. Once checkout is native to the browser, the agent optimizes for completion, not clicks. Conversions no longer depend on whether a merchant’s site has a perfect funnel because the agent can navigate friction in real time. That flips the incentive structure for both merchants and ad platforms. Merchants no longer chase micro-optimizations for each landing page. They expose clean, transparent flows that the agent can drive reliably.

Second, referrals. Perplexity now pays creators and users for distribution tied to agent activation. The unit is not a page view or a newsletter click. It is an installed, signed in, first query on Comet with optional default browser set. This is classic growth marketing applied to an agent. You can predict what comes next. Bounties for completing a first shop task. Bonuses for driving high quality downstream actions. Region based pricing for activations. It is a performance marketing model centered on the agent, not on an external landing page.

When checkout and referrals become default components in a browser runtime, advertising and search engine optimization lose leverage. The agent is incentivized to complete an action and to learn which flows convert. That puts pressure on the old ad auction and on ranking games. It also rewards merchants who expose structured data and predictable steps that an agent can parse on the first try.

Why startups have the edge right now

Startups can ship agent features without the weight of legacy ad businesses or entrenched extension ecosystems. They can design the product around actions from day one. Comet’s decision to go free is the classic growth land grab. Get the agent into as many hands as possible, then monetize heavy use and vertical add ons. A news bundle, higher model limits, and background assistance are revenue layers that do not blunt the distribution flywheel. Referral cash sweetens the loop by recruiting creators, student ambassadors, and hobbyist communities to seed the install base. We saw similar momentum when the warehouse agent went free, which unlocked adoption in operations without heavy procurement cycles.

Opera’s paid early access tells a different story. A browser company with an existing user base is trying to segment by willingness to pay for premium agency. Early revenue validates the idea and funds further development. It also buys a testing ground with power users who will tolerate rough edges in exchange for capability. The downside is obvious. Free agents get mindshare. Price sensitive users avoid the paywall. In a world where Google, Apple, and Microsoft will bundle baseline agents into default browsers and operating systems, paid access must deliver striking value to stand apart.

A useful framing is to separate distribution mode from monetization mode.

• Distribution mode: ubiquity through free installs, cash referrals, and viral workflows that demonstrate a clear before and after.
• Monetization mode: metered usage for heavy users, vertical bundles for power tasks, and enterprise features that wrap identity, audit logs, and support.

Perplexity is optimizing for distribution mode today and reserving multiple monetization options for later. Opera is testing whether monetization mode can be the wedge that funds distribution. Both can work, but they produce very different ecosystems for developers and partners.

The tool registry era begins

Agents are only as useful as the tools they can safely call. That is why a quiet but pivotal change landed this fall. An open protocol for tool definition and discovery arrived in preview, giving developers a way to publish standard tool servers and giving agent clients a canonical place to find them. Think of it like a package manager for actions. A travel tool that wraps multiple flight search sources. A retail tool that aggregates up to date stock and store hours. A calendar tool that grants scoped, revocable access. Once these are in a registry, agents can discover capabilities at runtime rather than relying on hard coded integrations.

This matters for browsers. If Comet, Neon, and their peers speak a common tool language, the agentic web becomes more interoperable. A user can bring the same tool set between browsers or across devices. Enterprises can publish internal tools to the registry and restrict them to approved clients. Security teams get a simpler surface to audit. Most importantly, the pace of agent feature growth accelerates because teams do not have to rebuild the same connectors. Tools become portable, composable building blocks.

The first wave will expose obvious capabilities like travel search, retail inventory, and calendar control. The second wave will carry enterprise specific tasks, from invoice approvals to role based knowledge access. We already see early beachheads in customer experience where agentic AI makes CX the first autonomy beachhead. Browsers can pull those same patterns into the general web by letting an agent chain tool calls with visual verification.

On device control changes what is possible

A long standing problem with web agents has been brittle automation through site specific scripts. The new generation is different. Models can now perceive the screen directly and manipulate interfaces with a virtual mouse and keyboard. These computer use systems are landing in production. They can scroll, click, type, and verify the result with visual feedback. They also ask for confirmation before sensitive actions and pause on authentication. In practice, this closes the gap between nicely structured application programming interfaces and messy real world sites.

Browsers are the natural host for this capability. Opera Neon’s local navigation agent lives inside the browser and acts within a contained workspace. Comet performs its own page level actions. Over the next year, expect on device computer use to integrate deeply with password managers, passkeys, and profile systems. That will let agents act inside logged in sessions without leaking credentials. It also means agents can complete enterprise workflows behind the firewall when the browser is connected to corporate identity and device management.

There are real risks to manage. Agents need clear permission windows, visible action previews, and rollbacks for actions that go wrong. They also need per site profiles, so an aggressive click strategy on a shopping site does not carry over to a banking site. The agent that can act confidently and safely will beat one that still asks the user to shepherd every step.

How this upends search engine optimization, ads, and app discovery

Search engine optimization that is optimized for snippets and ranking assumes a human in the loop who will click through and skim. Agents do not skim the way humans do. They map a plan to the specific result you asked for and only retrieve what is needed to complete that plan. That punishes fluff. It also rewards sites that expose structured data and predictable flows. If your site buries the important details behind tabbed accordions and dark patterns, an agent will either fail or take the user elsewhere.

Advertising inventory shrinks in the most valuable moments. If a browser agent compares options and books in one flow, there are fewer opportunities to show ads. Some ads will move into the agent’s suggestions and product cards. That will raise policy and trust questions. The larger shift will be toward affiliate and referral economics where the browser collects a fee for completed purchases. Transparent disclosures and user control will be essential for credibility. Expect browsers to offer universal receipts that declare referral terms and a unified panel where users can opt in or out of referral driven suggestions.

App discovery changes because the browser becomes the agent operating system. You will not go hunt for a diet app. You will ask your browser to design a weekly meal plan and populate a grocery order. The browser will pull recipes, generate a shopping list, map it to store inventory, and ask at the end whether you want to subscribe to a meal planning assistant. That assistant is not an app you installed. It is a tool bundle the browser can load on demand.

By 2026, default browsers will feel like bundles of assistants. Tool registries will supply the catalog. Affiliate rails will supply monetization. On device control will remove the need for every merchant and service to ship a perfect application programming interface. Organic discovery will still matter, but not in the old sense. The question will be which agents your browser trusts to handle a category of tasks and what incentives shape those defaults.

What to do now if you build for the web

• Expose structured data and predictable flows. Agents thrive on consistency. Use clear field labels, visible totals, and standard checkout steps. Avoid clever but nonstandard interfaces that break automation.
• Publish tools to a registry as soon as your category emerges. Give your agent friendly endpoints a name, a schema, and versioning. Treat tool quality like an application programming interface product, with observability and rate limiting.
• Prepare an affiliate and referral layer that reflects the agent reality. Move from top of funnel pixels to post purchase signals. Offer browser partners clear payout terms, regional pricing, and fraud controls based on completion, not clicks.
• Benchmark agent flows the way you benchmark page speed. Record how many steps an agent needs to complete your most common tasks, how often it misreads an element, and where it pauses for approval. Tighten the weak points.
• Give users control and transparency. If your service is recommended by a browser agent, disclose the referral relationship inside your own receipts and emails. Show the logic behind suggestions and allow opt outs.
• Design for safe action. Add confirmation checkpoints for money movement, health decisions, and identity changes. Provide an action log with an undo window that your support team can access.
• Build per role experiences. Expose standard flows for consumers, but also define administrator and auditor views so enterprise agents can act without breaking compliance.

The competitive shape heading into 2026

• Perplexity: Free core for distribution, layered subscriptions for heavy use, native checkout, and cash referrals to activate new markets. The bet is that ubiquity wins, and economics are driven by usage, conversions, and premium models.
• Opera: Paid early access as a wedge for premium agent workflows, designed for power users who will pay for end to end capabilities. The bet is that a durable browser business can be built on differentiated agent features and a reputation for privacy and control.
• The platforms: Google is weaving agentic features into its browser. Open model providers are pushing computer use and agent kits deeper into their stacks. If default browsers ship competent agents, startup browsers must stay faster, more open, and more aligned with power users to hold share.

If these bets hold, browser strategy will look like mobile strategy in the early 2010s. The winners will own identity, tool distribution, and payments. The difference is that action now flows through a general purpose interface that understands screens, not just APIs. That puts the browser, not a walled app store, at the center of the value chain.

What could still derail this shift

Three friction points could slow agentic browsing.

1. Privacy debt: If referral economics and on device control are not matched with clear consent and data minimization, users will push back. Browsers must treat action history like sensitive data with strict retention controls.
2. Site adversarialism: Some sites will attempt to block agent traffic through detection and obfuscation. Browsers will need a dialogue with merchants to define acceptable automation and to highlight how agents can actually increase conversion.
3. Capability cliffs: Agents will fail in long tail edge cases. The browser that recovers gracefully, explains what went wrong, and lets the user finish quickly will win loyalty. That requires better error handling and human factors design, not just stronger models.

The browser is becoming the agent operating system

The browser’s job used to be to render documents and run scripts. Now it also hosts perception, planning, and control. Comet going free made that future feel immediate. Neon charging for access showed there is room for different models. Tool registries are giving agents a common vocabulary. On device computer use is letting them act in messy reality. Checkout and referral rails are paying for the party.

The web is not vanishing. It is gaining a runtime where intent flows into action. The winners will be the teams that design for that runtime now, with the humility to let the browser do the clicking and the courage to align economics with the finish line. If you build for the web, your new success metric is simple and demanding. Optimize for agent completion rate and the trust that makes users say yes when the agent asks to proceed.