I’m sure you’ve heard about the various new browser releases from ChatGPT & Perplexity, while Edge and Chrome rush to load your web browsing experience with features, automations and promises of more autonomy.
How Agentic Browsers Are Rewriting the Privacy Contract
In just three weeks this October, two AI giants changed the way we interface with the internet. On October 2, Perplexity released its Comet browser to the public for free. Nineteen days later, OpenAI launched ChatGPT Atlas on macOS. Neither event captured the media attention they probably should have, but both mark a fairly tectonic shift in how humans experience the web. The rules of the internet are changing at this scale for a 5th time (by my counting):
Key Eras of the Internet
1. Web 1.0: The Static Web (1990s–early 2000s)
Marked by read-only HTML websites, basic hyperlinking, and limited interactivity.
Internet use focused on accessing static information from content providers and labs, with users as passive consumers.
2. Web 2.0: The Social Web (Early 2000s–2020s)
Transformation driven by user-generated content, interactivity, and social platforms (e.g., Facebook, YouTube, Twitter).
Enabled participatory communities, blogging, crowdsourced data, and the explosion of collaboration over the open web.
3. Mobile and Cloud Era (~2010s)
Shift to always-connected experiences via mobile apps, cloud-managed services, and APIs.
Created a new paradigm of on-demand access, personalization, and distributed digital environments, shaping modern commerce and communication.
4. Web 3.0: The Semantic/Intelligent Web (Emerging 2020s–2025+)
Defined by AI-driven recommendations, semantic search, “smart” assistants (voice, algorithmic agents), and decentralized architectures.
The web started understanding meaning, intent, and context, not just keywords, powering advanced automation and curation.
5. The Agentic/AI Browser Era (2025–present)
Agentic browsers (Atlas, Comet, etc.) shift the function of browsers from passive viewers to active digital agents that autonomously complete tasks, mediate access to information, and act on user intent.
We’ve just entered the era of the agentic browser: tools that don’t just render web pages, but read them for us, act on our behalf, and remember everything. They promise to free us from the tedious, but at the cost of ceding tiny micro-choices that once defined how we thought, searched, and discovered online.
THIS is the tradeoff now unfolding across the internet: the growing relief of automation VS the quiet erosion of choice VS the worry of intrusion of privacy.
And it’s not just the big players - there will be a LOT of independent agentic browsers popping up to lean on your greatest insecurity, fear or need.
Convenience architecture
Traditional browsers like Chrome or Firefox are neutral; they show what you ask for. Atlas and Comet are more participatory: they guess your needs, summarize text, suggest actions, and automate decisions. Their models depend on deep data access to do this effectively.
Comet, by default, collects URLs, content, and browsing history to “retrieve relevant information.” Unless you opt out, this also trains Perplexity’s AI models.
Atlas stores seven-day “browser memories” summarizing content to personalize your experience - essentially caching snapshots of your intent.
Turn off these features, and you lose the capabilities that define the product. The mechanism that powers the convenience is the surveillance itself.
A tense fight between autonomy and control
I’ve sampled each agentic offering and have CoPilot and Gemini Pro too, and I can safely say that agentic browsers do tend to make things easier for me. Right now that ease comes in retracing my steps across my computer, saving notes in a more organized way or helping me troubleshoot things that are going wrong. We’re not yet at the commercial application of agentic browsing: automated bookings, one-click research, agentic commerce; but the wheels have begun turning on those too.
In effect, they let us “put our brains in jars” for micro-decisions: what to buy, what to read, what’s “relevant.” The more the AI learns, the less we intervene. And that causes ripple effects between our own thoughts/desires and the world around us.
We can see AI Summaries directly squeezing traditional search - this will continue until it is complete.
So let’s first be objective and look at the main selling points of each agentic browser offering - and break them down with a rating out of 5 for each area a business user and a personal user might be interested in. These ratings are collated from every major review site plus my own deep usage:
OpenAI Atlas
Deep ChatGPT integration (all web activity runs through AI)
Autonomous task execution (agents for bookings, research, workflows)
Proactive summarization and in-browser memory
Limited browser extension support (macOS focus)
Strong opt-out, less opt-in for privacy controls
Microsoft Copilot
Native in Windows/Edge with strong Office 365/Teams integration
Agent Mode automates internal and web tasks (multi-app workflows)
IT policy, permission, and compliance controls
Retains much user agency, leverages device-level access
Google Gemini
Runs inside Chrome; deeply integrated with Google Workspace and apps
Multi-tab, multi-task AI automation (bookings, form fills, cross-app actions)
Utilizes Google’s cloud ecosystem and past search/app history
Heavy data collection and cross-platform accessibility
Perplexity Comet
Standalone AI-native browser
Conversational agent for page summaries, shopping, open-tab actions
Revenue sharing for publishers, early-stage extension options
Data collection for personalization; opt-out privacy, not local-first
Here’s them all ranked out of 5 - 5 being how “good” they are at automating tasks, integrating with your internet ecosystem, respecting user privacy, being easy to learn, creating productive value and having secure policies.
How does each browser stack up right now when it comes to the tradeoff between automation/convenience and user control/privacy?
Platform | Automation | Workflow | Privacy | Ease of Use | Productivity | Security |
|---|---|---|---|---|---|---|
Atlas | 4.5 | 3.0 | 3.0 | 4.0 | 4.0 | 3.0 |
Copilot | 4.0 | 5.0 | 4.0 | 4.0 | 4.5 | 4.5 |
Gemini | 4.0 | 4.5 | 2.5 | 5.0 | 4.0 | 3.0 |
Comet | 3.5 | 3.0 | 2.5 | 4.0 | 3.5 | 2.5 |
What the ratings above show: Security ranks the lowest, followed by Privacy, whilst Ease Of Use ranks the highest. This is a combination of 12 different review sites plus my own ratings. It objectively means that Big Tech is solely focused on getting as many users in to the ecosystem as humanly possible right now.
Business users rank privacy concerns and data sensitivity as top priorities, driving cautious adoption and preference for features that protect corporate information. Personal users value automation and convenience, generally accepting more data exposure for productivity and ease of use.
Notes on why each one is ranked like they are
Atlas
Stellar for advanced automation; less flexible for highly customized workflows or restrictive business privacy requirements.
Business users will worry about “AI gatekeeping” and cloud-centric memory.
Personal users enjoy robust summaries, but face a learning curve with agent commands.
Copilot
The gold standard for businesses in Microsoft environments; unbeatable if workflow is Office-centric.
Some features are overkill for casual use; policy/permission management shines for IT admins.
Personal use is pleasant but less compelling if you’re outside the Microsoft ecosystem.
Gemini
The easiest to start using (native to Chrome); best cross-app integration and recall/query of past activities.
Business users face data privacy caveats, as Google’s monetization is ad-driven.
Excellent productivity booster for almost anyone; weaker where compliance or confidentiality is critical.
Comet
Strong on open-web automation, summaries, and publisher incentives, but early-stage on deep integration, security, and local privacy.
Businesses may hesitate on sensitive workflows as cloud data sharing is opt-out, not default-off.
Great curiosity/learning tool for personal users, less so for finance/legal/regulated sectors.
When does assistance becomes advertising?
The underlying business models reveal the deeper incentive - and as you might have guessed, the tertiary revenue stream of advertising starts to bleed in to becoming the primary one again:
OpenAI Atlas: designed to capture user data, tie activity to the ChatGPT ecosystem, and eventually layer in commerce and advertising.
Perplexity Comet: positions itself as a fair-share browser compensating publishers; but only by mediating every interaction through its ecosystem.
Both models depend on full mediation of the open web - essentially you have to give the agent access to everything because it needs to do the things you used to so your brain can sit in a jar, right? In a world of agentic browsing, we’ve explored that the balance shifts from the user asking things to the AI offering things. What once was an infinite space of exploration now risks becoming a closed system of curated outcomes; super efficient… but not controlled by you.
The illusion of privacy
“Incognito mode,” “privacy mode,” and “limited interaction with the cloud” sound protective but are largely performative. Even anonymized metadata can reveal detailed behavioral patterns. Privacy in agentic browsers is not so much eliminated as it is redefined around acceptable levels of exposure.
Usefulness now depends on trust: that the vendor’s goals align with the user’s long-term interest. Regulation hasn’t yet caught up to define what that trust should look like, but it’s pretty clear that data privacy and utility are flip sides of a coin that each individual is going to make themselves, daily with their micro behaviours - as opposed to one rule being applied to technology providers to protect users. As mentioned above, Big Tech is going to push forward with ways of making these browsers more and more proactively helpful, whether you like it or not. And one day you may well just accept it, but the main objective of this piece is show you that micro choices matter.
The micro-choice deficit
What’s vanishing isn’t just privacy - it’s friction, deliberation, and discovery. Each automated step saves seconds but removes decisions that once shaped cognition and creativity.
Browsers filter what’s worth clicking
Agents auto-summarize what’s worth knowing
Recommendation loops prioritize what’s most efficient
Over time, these small delegations accumulate into a quiet pattern of behavioral outsourcing, where convenience collapses curiosity.
So we have a double-edged future ahead
Agentic browsers will democratize access, they’ll simplify complexity, and they will increase productivity. That’s for certain. But they also embed new power hierarchies between users, platforms, and publishers.
Final Closing Note
In the agentic era, nearly everything you do online - every search, every click, every delegated task - will be turned into part of a revenue ecosystem between businesses, at a speed and scale far what we’ve seen to date with adtech. But there's a deeper shift happening: big tech companies are racing to lock users into their ecosystems before they’ve realized what’s happening.
The race for better features and the building of an entire productivity-commerce-browsing ecosystem is about control over the next 20 years.
OpenAI is building an operating system, not a browser. Atlas eliminates the traditional address bar, forcing every web interaction through ChatGPT first. The company is positioning itself to become "the substrate of intelligence itself," mediating not just searches but every complex decision you make online. With 800 million users, Atlas doesnt need to compete with Chrome as a browser, as much as it redefines what a browser is: what the Google search bar used to be to Google.
Microsoft is embedding Copilot into Windows at the OS level, turning every PC into an "AI PC" where agents have device-level access to launch apps, manage workflows, and execute tasks across your entire system. Google is weaving Gemini throughout Chrome and Workspace, using your search history, emails, and documents to create predictive profiles that feed its advertising engine. Even newer players like Perplexity are using revenue-sharing models that sound fair; but only work if they become the mandatory gatekeeper between you and what you’re browsing.
As The Atlantic reported, these companies are intentionally designing systems that are harder to leave. Proprietary APIs, bundled premium services, exclusive fine-tuning tools, and controlled memory systems all create "vendor lock-in", where switching becomes so costly and complex that you're effectively trapped.
So the platform wars in Internet 5.0 aren’t necessarily about who has the best LLM. It’s once again about distribution, lock in and monetization. Who controls the interface between your intention and every action you or your agent takes. When your browser becomes an AI agent with full privileges across all your logged-in sessions (banking, email, shopping, work systems) the company controlling that agent controls your digital life. There is a huge flurry of circular revenue deals being done by AI businesses right now to boost the top line revenue, but the real race is for your money, hitting their pockets.
So much like we all naively entered internets 1, 2, 3 and 4 - we’re about to walk in to v5 and it’s clear that things aren’t going to slow.
That’s why AI Literacy is key. At the very least, you can make sure you enter this new world with your agency intact. Here are a few areas you can quietly make a choice:
Understand the lock-in playbook: Companies want you to build habits, store data, and delegate tasks inside their ecosystem, because leaving becomes exponentially harder with each integration. Recognize when convenience is actually dependency.
Demand local-first processing: you may not want to do this as a personal user which is fair, but if privacy is #1 for you then insist on browsers and AI tools that run on your device, not their servers. If the AI must phone home to function, you're renting access to it, not owning your tools.
Choose true opt-in, not opt-out: reject platforms that collect your data by default and hide privacy controls in settings. Look for systems where data collection requires your explicit, informed consent.
Prioritize interoperability and portability: Use platforms that let you export your data, switch providers, and avoid proprietary formats. If a company makes it hard to leave, that's a red flag.
Watch for ecosystem creep: Be wary when one tool starts integrating with everything you use. This is a strategic play under the guise of convenience in order to make you dependent across multiple surfaces.
Support open standards and independent audits: if you’re a business user reading this, push for transparency in how data is stored, used, and deleted. Regulators in the US, EU, and UK are already investigating AI lock-in - so you can demand accountability now.
Build hybrid strategies: for large enterprise businesses, the smart approach is to use APIs for experimentation, but on-premise or open-source models for production. Don't let your entire operation depend on one vendor - swathes of businesses who have relied too deeply on Google to run their marketing, advertising and content delivery are beginning to regret it. I personally would rather see a healthy competitive ecosystem at play and that requires businesses to build their own journey instead of just subscribing to Open AI’s journey.
The convenience is real and the productivity gains are genuine. But the cost is the small decisions you make - and that’s without even going deeply in to the upcoming world of Agentic Commerce, where an AI agent buys things on your behalf and you can choose to dispute it or send it back - instead of making the initial decision.
Once these platforms achieve dominance inside a business (or your personal life), reversing course becomes nearly impossible. We saw that with Google in the media industry and we' see it with ChatGPT in our personal lives. It remembers us, so it’s valuable right? As one analyst put it, Atlas and similar browsers represent "the first step toward AI operating systems". The next few years will determine whether that future is built on open standards and user control, or closed gardens and corporate surveillance.
I’m not here to fear monger; I love AI. I just wanted to write this so we all enter the agentic era with both optimism and lessons learnt. The companies building these systems are betting that most users won't read the terms, won't understand the implications, and won't care about the tradeoffs until it's too late to choose differently. Don't let that be you.
Brave Security Team. (2025, July). Security analysis of Perplexity's agentic browser vulnerability. Brave Security Blog. Bloomberg. (2025, June). Search traffic and publisher economics in the AI era. Bloomberg Tech Report. Cyber Protect LLC. (2025, August). Assessment of privacy features in Perplexity Comet browser. Cybersecurity White Paper. Forbes. (2025, May). Agentic browsers and security risks for educational institutions. Forbes Technology Council. Google Analytics, UC Davis. (2025, March). GenAI browser assistants and out-of-context personalization risks. University of California, Davis. OpenAI. (2025, October). Atlas agentic browser launch & privacy documentation. OpenAI Documentation Archive. Perplexity AI. (2025, October). Comet browser privacy policy and technical documentation. Perplexity AI Help Center. Simon Willison. (2025, Spring). Architectural privacy gaps in agentic assistants. Personal blog and conference proceedings. StatCounter Global Stats. (2025). Browser market share data. Retrieved from https://gs.statcounter.com/ Similarweb. (2025). Web traffic and ecosystem lock-in analysis. Retrieved from https://www.similarweb.com/ The Atlantic. (2025, July). Platform lock-in tactics and ecosystem entrenchment. The Atlantic Magazine. University of California, Davis. (2025, March). Research on out-of-context personalization in GenAI browsers. UC Davis Reports. Yahoo News. (2025, February). Zero-click search and publisher visibility decline. Yahoo News Media Analysis.
