When I asked Jito Chadha where AI is headed inside real businesses, he kept returning to a practical problem: Most companies are still built for people clicking through screens.
A person opens a browser, types into a search bar, compares options, fills out forms, waits for approvals, and moves information from one system to another. Chadha believes that sequence is now being tested by AI agents that can search, request, verify, and complete work on behalf of a person. The technology is moving quickly. The business infrastructure around it is moving more slowly.
“As the world transitions to an agentic everything,” Jito Chadha told me, “you will see a lot of browser automation.”
That phrase, “browser automation,” is where Chadha sees the bridge between today’s human-designed web and the next phase of business software.
Chadha used hotel booking as a simple example. Today, a human searches for a room, clicks into a hotel website, checks dates, compares prices, and completes a transaction. An agent can begin to simulate that same process by operating the browser, keyboard, and mouse.
“It can even go into that target hotel’s website and do the browser automation and the UI automation to actually simulate like a human’s use of the keyboard and mouse,” Jito Chadha said.
The problem is that the agent is still entering through a doorway meant for a person. It is not interacting with the business through a direct, authorized path. Chadha described the current moment as “two completely different ways of interacting with the same type of data.” In one model, “AI is authorized.” In the other, “The AI is pretending to kind of navigate that website like a human.”
For founders and executives, that distinction matters. It suggests that AI adoption will require more than buying an agent. Businesses will need to think about how an agent proves identity, what data it can request, what permissions it has, and how much human oversight belongs in a workflow.
Chadha expects companies to eventually create cleaner routes for agents. Those routes may involve APIs, webhooks, or documented frameworks that allow agents to request pricing, availability, documents, or status information in a controlled way.
His reasoning is rooted in cost and efficiency. He pointed to the evolution from SOAP APIs to REST and then to GraphQL. With older structures, a system often had to request more data than it needed. That created unnecessary internet traffic, database queries, and infrastructure costs. GraphQL allowed a more specific request, where a system could get only what it needed.
“Why waste compute on something that could be far simpler?” Chadha said.
He expects a similar movement with AI agents. Browser automation may carry companies through the transition, but more deliberate agent interfaces will eventually become the efficient path.
Permission Becomes the Hard Part
For Jito Chadha, the value of an agent depends on what it can safely access. A generic chatbot may answer questions. A useful business agent needs context, data sources, and permissions.
“What kind of information would you want the agent to have?” Chadha said. “If you want databases attached to it, if you want every conversation to result in some kind of summary that then gets pushed to another queue where somebody can follow up at a later time or an agent can follow up at a later time. These are all considerations.”
That is where Chadha’s view becomes about AI as a business process. An agent has to know what it can touch, when it can act, and where the record of that interaction goes next.
Start Where the Work Is Obvious
Chadha does not recommend that companies start by automating whatever sounds most impressive. His framework begins with triage.
Across large operations, he said, the decision comes down to the number of people or cost associated with a process and then the complexity and viability of automation. A costly process may be too complicated to automate first. A simple task may not create enough impact.
“You kind of go for the lowest hanging fruit that has the biggest impact, always,” Jito Chadha said.
He described that formula as “like a fractal.” It can apply to one role or across a company. A marketing team might look at data collection, content production, or digital media interaction. A customer service team might look at verification, lookup, and status updates. An internal operations team might look at document retrieval, approvals, or recurring handoffs between systems.
One example came from benefits administration. Chadha described a company with inbound calls from people asking about the status of checks connected to a class-action matter. The process requires verification before specific information can be disclosed. His team is building an agent that can verify the caller as an authorized person, look up the person in a database, and provide the relevant status.
The goal is partly economic. Chadha said customer service systems are often measured by deflection rate, meaning how many calls are avoided before a human has to pick up the line. But he also pointed to a simpler benefit: “Having inbound callers get faster service.”
By the end of our conversation, Chadha had described a practical middle stage for business AI. Agents will use browser automation where they must. They will move toward authorized interfaces where companies build them. They will become more useful as data, identity, and permissions become more deliberate.
For executives, the immediate question is where work slows down, where access is safe, and where automation can create visible impact without overwhelming the business.
The agent-ready company, in Chadha’s view, begins there.
