EmTechMIT, Cambridge, MA: Sudheesh Nair, Cofounder & CEO of TinyFish, opened his talk with a bold premise: AI agents are approaching a pivotal transformation, ready to change the very shape of the Internet by evolving from simple assistants to fully capable “associates”, collaborators and actors that handle real-world tasks across domains like hospitality, commerce, and transportation.

He highlighted a core tension: “There are certain things that the models are really good at doing and there are certain things these models would struggle.” For instance, agents excel at deterministic tasks, such as solving math problems, but falter when confronted with the ambiguous, multi-dimensional outcomes inherent to human goals, like booking travel or finding products that suit nuanced preferences.

The Problem of Browsing and Information Depth

Browsing, he argued, is a task humans intuitively master but one that AI finds challenging. “The internet has gotten too big, too massive for humans to do anything with it anymore. We have to revive the internet… we are only seeing 10% off the internet today.” Google indexes less than 10%; systems like ChatGPT see even less, which leaves vast amounts of data, the so-called “deep web”, untapped by current agent technology.

Nair explained that most businesses and consumers are unsatisfied with the superficial answers delivered by search engines. “Most businesses are not looking just for [a price range]. They need something more.” Modern agents, like TinyFish’s Atlas, must not only fetch surface data but also navigate captchas, APIs, logins, personalization, and dynamic content flows, all while reasoning about intent and context.

Five Layers of Internet Complexity

Sudheesh classified web data into five layers:

• JavaScript-embedded and API-gated data (e.g., real-time airline pricing),
• Appointment-based information (not trivially accessible),
• Contextual personalization (what you see depends on your identity and history),
• Logins and social media barriers,
• Deep content repositories hidden beyond surface search indexes.

He underscored that “social media… all of them are behind logging in… Reddit is the place where everything happens.” In the age of personalized, dynamic content, the static “top 10 links” model is obsolete.

The Executable Web Vision

Nair’s session advanced the concept of an “executable web,” shifting away from websites designed for human browsing toward MCP endpoints: programmatic interfaces meant to interact with agents directly and answer user intent. “Webpage and the site maps don’t matter, what matters is this MCP endpoint that basically says come to me and ask me what you wanna know, or tell me what you want to do.”

TinyFish’s model, therefore, doesn’t aim to compete for consumer traffic. Instead, it enables enterprise clients to program fleets of agents that comb 100% of the web, including hidden layers, for actionable insights. For example, a retailer can use agents to track competitive pricing and feature differences across platforms to ensure their own products aren’t missing key information like amenities, which can drastically affect bookings.

Infrastructure Innovations and Technical Underpinnings

A major advance in recent years, according to Sudheesh, has been the maturation of cloud platforms and agent orchestration: “We have a customer spinning up 800,000 browsers, even within a matter of literally minutes and shut them down and then bring them back up again.” This radical scalability, combined with improved reasoning and context understanding in LLMs, allows for browser agents capable of traversing complex web environments, not just scraping but acting intelligently, akin to human users.

He demystified the technical approach: “Instead of just visual screenshots for LLMs, we combine what a machine and a human see. Like a doctor using both sight and MRI. We flatten the website, extracting checkout times, price changes, promotion logic, the sort of high-quality signal useful for reinforcement learning.” AI now leverages both real-time and historic data, continuously learning and adapting its workflow for higher precision.

The Battle for Internet Power, and Democratizing Visibility

Sudheesh addressed the looming platform wars, specifically between Google and OpenAI: “Google index everything and they get keep and then monetize. What OpenAI is doing is summarizing it and then answering without sending anything back.” TinyFish’s position is to sidestep these giants and empower enterprise and small businesses who historically have been invisible (“TinyFish need to be seen”) by surfacing their offerings deeper into the digital stack.

He relayed that currently, “most valuable use cases”, clinical trial recruitment, retail comparison, boutique hotels, live in the web’s long tail. AI-powered agents must reach into the farthest corners, finding that 8-room mountain hotel or organic nail salon not surfaced on Google or main aggregator platforms.

The Internet of Tomorrow

Sudheesh mused on the future: “Websites all becoming MCP endpoints, the webpage and the site maps don’t matter. What matters is this MCP endpoint.” He believes that, similar to the transition from Yahoo’s flat directories to Google’s algorithmic organization, another leap is coming where AI agents personalize, traverse, and act across a web optimized for executable intent rather than passive consumption.

Advertising and monetization models will face disruption. “When monetary starts talking, everything stops. Amazon blocked agentic browsing, not because to make any shopping, but, because agents will not see the ads.” Nair predicts these entrenched models will have to yield to the more direct, outcome-focused offerings powered by agent technology.

Final Thoughts & Quotable Moments

TinyFish’s mission is not to be another aggregator or LLM indexer but an infrastructure provider: “You just need the tool and you spin out as many these agents as you want. We will go out to the internet and bring you the insight in actionable days on CSV format.” Agents become personalized associates, “who knows what I am about, what am I looking for, who am I”, working across all of the web instead of the shallow tip.

Sudheesh concluded with humility and vision: “Just because I’m on stage, my opinion in this area is probably not going to be any better than all of you because we are all wishing it. We definitely don’t think internet will continue the way it is because this is a massive disruption.”

For technologists, the message is clear: the era of agents-as-associates presages profound change, not just for web architecture and business, but for the logic of online existence itself.

For more information, please visit the following:

Website: https://www.josephraczynski.com/

Blog: https://JTConsultingMedia.com/

Podcast: https://techsnippetstoday.buzzsprout.com

LinkedIn: https://www.linkedin.com/in/joerazz/

X: https://x.com/joerazz