The San Francisco startup Tilion has been selected for Y Combinator’s Fall 2026 batch. Three of its four co-founders are Bangladeshi.
They are Md Shahriar Islam Arham, Kazi Fardinul Haque, Arman Luthra and Abdul Basit Tanmoy. What they are building is infrastructure that lets AI agents work on the web as smoothly as people do.
The problem they picked
Think about a morning that involves logging into ten websites, filling in the same form repeatedly, and copying information from one page into another. None of it is hard. All of it is dull and slow, and handing it to an AI agent is something the whole industry is now attempting.
The trouble starts at scale. One or two agents are fine. When thousands go out onto the web at once, things jam — a site is slow to load, a browser session becomes unstable, and the job stops halfway.
The reason is structural. Every site on the internet was built on one assumption: that behind the screen sits a person, clicking from one page to the next. Nobody designed for agents. They are, as the founders put it, second-class users of the largest thing humanity has built.
What Tilion built
A separate browser infrastructure for agents. Y Combinator’s own profile calls it "self-healing": it can run several million agent operations a day without human supervision, and when one session breaks another picks the task up and carries on.
Customers use it for large-scale data collection, automated workflows and continuous monitoring. Many browsers can be rented at once, and connecting it to an existing system takes a single line of code.
At the centre of it is an open-source engine called Fortress, built on the Chromium browser engine.
What getting in means
Y Combinator is among the best-known startup accelerators in the world. Airbnb, Stripe and Reddit all started there, and Bangladesh’s own Chaldal is on the list. Various reports put the acceptance rate at about 1% of applicants.
That is the part worth noting from here. The interesting thing is not that a Bangladeshi team was admitted; it is what they were admitted for. Tilion is not building an app for a local market. It is building a piece of infrastructure the global AI industry needs and does not yet have, which is a different kind of bet, and a harder one.
Why this problem is about to get much bigger
Agent traffic is growing faster than the web’s ability to tell it apart from people. Sites built to serve a human reader now receive automated visitors that log in, fill forms and read pages at machine speed, and most of the defensive furniture on the web — rate limits, bot detection, session handling — was designed to stop exactly that behaviour when it was hostile.
So there are two directions this can go. Either the web gains proper interfaces for agents, or agents keep pretending to be browsers and the infrastructure underneath them has to absorb the mismatch. Tilion is a bet on the second, which is the safer bet, because it is already what is happening.
The thing worth watching locally
Bangladesh’s technology reputation abroad rests heavily on services — freelancing, outsourcing, development work done to someone else’s specification. Those are real industries and they employ a great many people, but they are priced as labour.
Infrastructure is priced differently. A company that sells a component other companies build on earns from their growth rather than from hours worked. Whether Tilion succeeds is a separate question from whether the attempt is the right shape, and the attempt is the right shape.
