The big idea
The episode's central symmetry writes itself: FSD works because driving is a constrained task with observable state and clear feedback — and browsing has the same shape. Pages are the road, links are turns, forms are intersections. An agent that can perceive the page and plan actions can "drive" the web the way FSD drives Lombard Street.
The deeper point is about who browsing is for. Most of what we do online — comparing, booking, filling, checking — is chore traffic, not exploration. Agentic browsing splits the web into experiences you want to have and errands you want done, and quietly automates the second category.
Why this matters now
Browsers are the most contested territory in AI right now — every major lab and browser vendor is racing to put an agent behind the address bar. The web's chore layer (booking, comparing, form-filling) is enormous, measurable, and monetizable, which makes agentic browsing one of the clearest land-grabs of the platform shift.
Key takeaways
- Browsing has a self-driving equivalent. Perception (reading the page), planning (choosing actions), execution (clicks and forms) — the FSD loop maps one-to-one onto web tasks.
- Errands are the wedge. Nobody wants an AI to read their favorite blog; everyone wants one to handle the seventeen-tab comparison shop. Retriever targets the web's chore layer.
- The hard part is the long tail. Like driving, the demo is easy and the edge cases are brutal — broken layouts, logins, CAPTCHAs, ambiguous UI. Reliability, not intelligence, is the product bar.
- Supervision is a feature, not a failure. Early FSD keeps a human in the loop; early browsing agents should too. Trust is earned in stages, per task type.
- Two technical cofounders, one interface bet. A CEO/CTO pair betting that the browser — not a new app — is where agents meet everyday life.
Inside the conversation
The FSD-to-browser analogy, stress-tested
The ride pushes the metaphor hard: what's the browsing equivalent of an intervention? Of a disengagement report? The founders' answers sketch how agentic browsing will actually be measured and trusted.
What the web looks like when agents are users
If a meaningful share of traffic becomes agents acting for humans, sites face a choice: fight them or serve them. The conversation explores a near future of agent-readable interfaces and machine-negotiated errands.
Building rtrvr.ai
From naming a company after a dog that fetches, to deciding which verticals' errands to automate first — a look inside the earliest product decisions of an agentic-web startup.
About Retriever AI
Retriever AI (rtrvr.ai) builds agentic browsing — an AI that navigates and acts on the web for you. Founded by Arjun Chintapalli (CEO) and Bhavani Kalisetty (CTO).
Questions this episode answers
What is agentic browsing?
Agentic browsing is an AI operating your web browser for you: reading pages, clicking, filling forms, and completing multi-step tasks like comparisons or bookings. You state the goal; the agent drives the web — the way FSD drives the car while you supervise.
What is Retriever AI?
Retriever AI (rtrvr.ai) is a startup building agentic browsing — an AI that navigates and acts on the web on your behalf. It was founded by Arjun Chintapalli (CEO) and Bhavani Kalisetty (CTO).
Will AI agents replace manual web browsing?
For errands, largely yes: research-and-book, compare-and-buy, and fill-and-submit tasks are exactly what agents automate first. Browsing you enjoy — reading, watching, exploring — stays human. The episode frames it as splitting the web into experiences and errands.
