1. Intercom: The Buttons That Survived the AI-Native Rebuild

Brian Donohue, Intercom’s VP of Product, on thinking big enough to reinvent a company.

Intercom tore their support product down to the studs and rebuilt it around large language models. Two buttons survived underneath every AI reply — “Yes” and “Talk to the Team” — quietly enforcing a turn-based mental model the rest of the system had outgrown. Brian Donohue, the 11-year VP of Product for Fin, walks me through what it took to finally see them, and what changed when they were gone.

Under every answer Fin gave, for two years, sat two buttons.

The AI would resolve a customer’s question, list its sources, and then ask: Did this answer your question? Below it, the only two ways to respond. Yes. Or: Talk to the team.

“It was, like, crazy bad UX,” Brian Donohue tells me, walking through it the way you walk through a photo of yourself from 15 years ago. “It was, like, the wrong question to ask at the wrong time.” Listen to the shape of the thing. Did this answer your question? Yes. Or: talk to the team. “You couldn’t say no and continue it,” he says. The interface presumed the conversation was over at the exact instant the whole point was to keep it going. “It was, like, a way no one would ever ask this.”

Brian is the VP of Product for Fin, Intercom’s AI agent for customer support — the thing he’s spent the last two and a half years going, in his own words, “progressively, progressively, progressively, more and more and more” all in on. He’s been at Intercom for 11 years. He’s watched the company carry its support product from the old rule-based bots, “where there’s a little bit of ml” and “mostly deterministic stuff going on,” all the way to a system rebuilt from the ground up around large language models. He’s proud of it. You can hear the pride in the way he narrates the early days — the summer of 2022, before ChatGPT, when one of his ML engineers stood up at a show-and-tell to demo summarizing a support conversation and the verdict was “it just still doesn’t work that well.” And then, a few months later, it did. “It immediately was good enough there,” he says. “That opened the door.”

So when he says the company built Fin from scratch, he means it. They threw away the old bots. They restarted on the new system. “That was ground up, from rebuilding.”

Which is what makes the buttons such a strange thing to find still standing.

They weren’t a corner of the product nobody looked at. “This is anchored into our whole system,” he says. “This is how we calculate resolutions. This is how we charge. It’s a fundamental UX part of the system.” The buttons weren’t a cosmetic afterthought. They were load-bearing. The company’s revenue ran through them.

And the team that built Fin from scratch — the team that prides itself on having thrown the old world away — did not see them. For two years.

“We were blind,” he says. “It was inherited product we were blind to.” He keeps circling the same disbelief, the way you do when you’re still metabolizing it in real time. “Oh, we built Fin from scratch.” A beat. “But no, we didn’t.” His voice catches the way it does when a person hears himself mid-sentence. “We still had parts of the system we were anchoring on, and two years later — oh my God, I can’t believe we still inherited that design and hadn’t challenged it.”

I tell him this is exactly the conversation I wanted. Earlier I’d said something I meant as a compliment and a confession of my own: I’m glad I’m talking to him and not to the CPO. The CPO is a fantastic storyteller, buttoned up and polished, and for good reason. “I want the documentary,” I tell Brian. “I don’t want the blockbuster.”

This is the documentary. A team rebuilds its entire product around the most disruptive technology of the decade, and the thing holding it back turns out to be a design decision they’d already declared finished. Not a hard technical problem. A button. A question phrased wrong, shipped fast, and never looked at again — because, as Brian puts it, “it’s slow, it’s deep in the process system, and so everyone’s reluctant to change it.”

What he says next is the first of three moments in this conversation where Brian admits that the thing he was most certain of was exactly the thing he had to unlearn. The button. Then a process he was sure would never die. Then the way he charged for the whole thing.

What this profile covers

The buttons are the first of three certainties Brian had to unlearn at Intercom. What it took for a team that prides itself on having rebuilt Fin from scratch to finally see the two-year-old design they’d been carrying — and what showed up on the dashboard the day they finally moved it.

The second is the operating model he and Basecamp once independently arrived at: the six-week cycle Basecamp made famous in Shape Up, the sweet spot Brian was sure would never die. He threw it out anyway, along with the triad and stable teams, in favor of a one-line mantra — follow the work — and a cost he names without flinching.

The third is how Intercom charges for Fin. Ninety-nine cents a resolution, billed only when the bot actually answers a customer’s question without escalating to a human. A pricing bet from a company with what Brian calls “a pretty bumpy history with pricing,” and what it collapsed about an entire category of product work he used to do.

Through-line: three things an 11-year intercom veteran was surest of, each one quietly the inherited debt. What he’s learned to distrust about the feeling of certainty itself.