Sendbird,
or the invisible middleman.

The plumbing nobody applauds

Inside a glass-walled office on First Avenue in San Mateo, engineers are watching a dashboard tick. The number on it has 10 digits. By the time you finish reading this sentence, it has moved a few thousand times. That number is messages. Sendbird carries roughly seven billion of them every month, threaded through 4,000 different apps, into the pockets of about six billion people. Almost none of those people have heard of Sendbird, and Sendbird is fine with that.

The company sells what plumbers sell: the bit you only think about when it leaks. Their APIs let any app slot in chat, voice, video, and lately, AI agents - the way you'd slot in a power outlet. Match Group, Reddit, DoorDash, Noom, Yahoo Sports, Hinge, Paytm. Different products, different audiences, same pipes underneath.

There is a small irony here. A company whose entire business is making conversation easier spends most of its energy ensuring you don't notice the conversation is happening on its rails. The success metric of infrastructure is invisibility, and by that yardstick Sendbird is, embarrassingly, very good at its job.

A small leak in every app

In 2013, if you were a product team trying to add chat to your app, you had two options. Option one: build it yourself. This required a team that understood real-time sockets, concurrency, persistence, moderation, push notifications, presence indicators, delivery receipts, and the specific way iOS likes to murder background processes at 3 a.m. Option two: cobble together a vendor stack that did some of those things, badly, and pray.

Both options ended in the same place. Engineers spent six months building chat. Then nine months maintaining it. Then a customer support manager somewhere asked why the messages didn't sync across devices, and everyone went quiet.

The tension was simple: chat looked easy and was, in fact, brutally hard. Every social app, every marketplace, every healthcare portal needed it. Almost none of them wanted to build it. Someone had to.

Four founders, one pivot, no ego

Sendbird was not the founders' first company. John S. Kim, Brandon Jeon, Harry Kim and Kevin Yoo had previously built SmileMom, a community app for mothers in Korea. It did fine. It did not do enough. By 2015, the team noticed that the messaging engine inside SmileMom - the part they'd quietly over-engineered for fun - was more interesting than the app itself.

So they did the thing every founder dreads and everyone, in hindsight, claims was obvious: they tore the engine out of the car and sold the engine. SmileMom became Sendbird. The Y Combinator Winter 2016 batch took them in, wrote the first $120,000 check, and pointed them at the United States.

The bet was specific. Most chat tools at the time were either consumer-grade (Slack, WhatsApp) or barely-functional B2B widgets that broke at scale. Sendbird's pitch was games-grade engineering applied to ordinary messaging. The founders had built real-time systems for environments where lag of 200 milliseconds was a failure. Apply that discipline to support chat and live commerce, and the result is something nobody else can easily copy.

From SDK to agent platform

For a long time, Sendbird's catalog read like a developer's grocery list. Chat SDK. Calls API. Desk (a chat-first helpdesk). Business Messaging for outbound SMS, WhatsApp, and KakaoTalk. UI Kits for teams that didn't want to design their own message bubbles. The pieces fit together, but they sold themselves as plumbing.

In 2023 and 2024, the pitch sharpened. Sendbird started describing itself as an omnichannel AI agent platform - which sounds like exactly the kind of phrase a marketing department would produce, and yet, surprisingly, means a real thing. The idea: support and sales conversations should be handled mostly by AI agents, escalated to humans when the agent admits it's stuck, and stitched across whatever channel the customer happens to be using that minute.

The receipts

It is easy to say a chat platform is big. It is harder to make the case land. So: numbers. Sendbird sits inside the apps you use without noticing, which is the whole point, but the receipts are real.

Reddit uses Sendbird for direct messaging and Subreddit chat. DoorDash uses it for the back-and-forth between customers, drivers, and merchants. Match Group runs flirty first hellos through it. Noom uses it to keep coaches and clients in sync. Yahoo Sports leans on it for fan chat. Different shapes, same plumbing.

Conversations that don't feel digital

The official mission, polished by twelve marketing meetings, is to build connections in a digital world. The unofficial mission, pieced together from interviews, conference talks, and the body language of founder John Kim, is more pointed. Make digital conversations feel like they are happening between real people, even when one of those people is, increasingly, a machine.

That is harder than it sounds. A support chatbot that escalates the wrong question to a human is annoying. One that holds onto a question it can't solve is worse. One that sounds like it is reading off a help center FAQ is the worst of all. Sendbird's bet is that the next decade of customer experience is won by whoever makes those handoffs feel natural - and whoever's voice AI doesn't sound like a 2011 phone tree.

The next bird

Communications infrastructure is the kind of business that gets boring in a hurry and then, every five years or so, is briefly the most important business in the world. WhatsApp was boring before it sold for $19 billion. Twilio was boring before pandemic-era telehealth made it indispensable. Sendbird is currently in its boring phase, which is exactly when you should pay attention.

The bet for 2026 and beyond is that every consumer-facing company becomes, by accident, a communications company. Banks message. Hospitals message. Grocery delivery messages. The companies that get this right will look like they hired an extra ten thousand support reps overnight. They didn't. They wired in AI agents. The wiring, if Sendbird's roadmap holds, runs through San Mateo.

Back to the dashboard on First Avenue. The number is still ticking, faster now than when this article started. Somewhere a DoorDash driver is asking a customer if they want the package by the door or in the lobby. Somewhere a Match user is sending the first message of what will become a relationship, or, more likely, not. Somewhere a Voice AI agent is rerouting a flight. None of these people are thinking about Sendbird. That, still, is the point. The middleman did its job, which was to disappear.