This episode of “Impact of AI: Explored” is all about AI browsers—the good, the bad, and the ugly—where we, James O’Regan and Gerjon Kunst, unpack why these new “agentic” browsers are exciting for productivity but also deeply challenging for security and privacy. It is a hosts-only episode, so it is just the two of us sharing hands-on experiences and concerns from the field, aimed at IT pros, architects, and anyone curious about the next evolution in how we use the web.
Introduction
In this episode of “Impact of AI: Explored,” we dive into AI browsers: full browsers with built-in AI agents that can browse, click, buy, and automate workflows for you. Instead of a guest, it is just James and Gerjon having an honest conversation about how these tools boost productivity, where they can go wrong, and what that means for both consumers and enterprises.
Meet the hosts
Impact of AI: Explored is a podcast initiative by and for the developer and IT professional community, hosted by James O’Regan and Gerjon Kunst. Our goal with the series is to help people make sense of the constant wave of AI developments and understand how they affect day-to-day life and work in IT.
- James O’Regan is an end-user computing and workplace specialist, podcast host, and product marketing professional who has been hosting technical podcasts and discussions around EUC, AI, and digital workspaces for years.
- Gerjon Kunst is CTO at RawWorks, Microsoft MVP, and co‑host of multiple podcasts, focusing on AI, cloud, and modern workplace technologies, helping organizations turn AI from hype into practical value.
Setting the stage
AI has already moved into search bars and productivity suites, and AI browsers are the next logical step: they do not just answer questions, they act on the web for you. That shift from “chatting with an LLM in a tab” to an AI that clicks, navigates, and completes tasks introduces huge productivity potential but also a whole new attack surface for security and compliance.
In this conversation, we explore what AI browsers are, how they work in practice, where they shine for everyday users, and why IT admins should be nervous. Listeners can expect concrete examples (sneakers, travel, expenses), real-world risks (browser hijacking, prompt injection, data leakage), and some predictions for how Microsoft, Google, and others will bring this tech into the enterprise.
Episode highlights
- From AI tab to AI browser
- We unpack the difference between “LLM in a tab” and a true AI browser that can log in, click around, and complete multi-step tasks on your behalf. In the episode, we talk about using Comet and OpenAI Atlas as daily browsers, including moments where the browser logs into sites, searches, and fills flows while you are doing something else in another tab.
- The productivity–security whiplash
- A key turning point in the conversation is when we move from “this is a productivity hack” to “this is an IT security headache,” especially once you log into corporate environments with an agentic browser. We discuss scenarios where AI could effectively hijack the browser session, blur the line between human and AI actions, and make it very hard for IT to tell who (or what) is touching company data and systems.
Deep dive: What makes an AI browser different?
A traditional browser with an LLM in a tab is just an interface to a model: you paste a URL, ask questions, and get summaries or explanations. An AI browser goes further—it sees the page (often via screenshots), interprets layout, decides where to click, navigates forms, and can carry out end‑to‑end workflows like planning trips, finding deals, or filling in web apps.
We talk about browsers like Perplexity Comet and OpenAI Atlas that ask for access to your keychain or passwords, highlighting both the convenience (auto-login, checkout, form filling) and the alarm bells this rings for security. In theory, if you let it, such a browser could also access your credit card details and execute purchases, which is incredibly powerful but not something you want in a V1 product without mature guardrails.
On Windows, Gerjon uses Comet; on macOS, Atlas, and both show the “agent at work” experience: the AI logs into sites, opens Google Maps, finds EV charging stops, and returns when it is done. This agentic behavior introduces latency (screenshotting, interpreting, clicking, repeating) but also demonstrates the future of AI-supported browsing and why it is more than just “ChatGPT inside Chrome.”
Real-life stories and examples
Throughout the episode, we ground the discussion in practical use cases:
- Everyday shopping: Searching for Nike Air Jordan sneakers within a price and color range and letting the browser not only find options but also put them into the cart so you can compare them later. This is a great example of how e‑commerce will be reshaped when AI agents can not only recommend but also act.james-gerjon-5.txt
- Travel planning: Asking the browser to plan a trip from Enschede to Paris with four stops, including two EV charging locations, and watching it drive Google Maps, find charging points, and propose an itinerary for you.james-gerjon-5.txt
We also touch on enterprise-flavored use cases that are both tempting and terrifying:
- Expenses automation: The dream scenario where you tell the browser, “Here are my receipts, here is the expense app—file everything,” and it grinds through forms while you keep working in another tab.james-gerjon-5.txt
- Corporate environments: Logging into Microsoft 365 through an AI browser and theoretically asking it to create emails or perform tasks in your tenant, which makes it nearly impossible for IT to know whether actions were taken by a user or by an agent.james-gerjon-5.txt
Key takeaways
- AI browsers are the next evolution after “LLM in a tab”: they do not just answer questions, they take actions on the web for you.
- Tools like Perplexity Comet and OpenAI Atlas already offer agentic behavior, planning trips, researching products, and running multi-step flows across sites.
- Productivity potential is huge—think automated expenses, travel planning, and shopping—especially when you can let the AI handle boring browser tasks in the background.
- Security and privacy risks are equally huge: browser hijacking, prompt injection, unclear data storage, and the difficulty of distinguishing human vs. AI actions in corporate environments.
- Enterprise adoption will depend on strong controls: role-based access, clear policies, data boundaries, and deep integration into existing security and compliance stacks.
- Expect the major players—Microsoft with Edge and Copilot, and Google with Chrome and Gemini—to push agentic features and bring more enterprise-grade controls over time
- For now, AI browsers are best treated as an experimental, consumer-first technology: great to play with on personal devices, but not ready for sensitive data or unrestricted corporate use.
Closing thoughts
Our verdict in this episode is simple: AI browsers are not a revolution yet, but they are a clear evolution in how AI will interact with the web and our daily workflows. They combine serious productivity gains with serious security questions, and both IT professionals and everyday users need to approach them with curiosity and healthy skepticism.
If you are experimenting with agentic browsers like Comet, Atlas, or upcoming versions of Edge and Chrome, we would love to hear your stories and use cases. Reach out if you have a cool scenario, lessons learned, or if you want to join us as a guest in a future episode on AI browsers and agents.
