How to become an AI-first designer
When ChatGPT launched in late 2022, Paul Adams, chief product officer at Intercom, saw it as a turning point. Within a weekend, Intercom scrapped its roadmap and pivoted to focus on AI-first product development. Since then, the company has launched Fin, an AI-powered chatbot that now resolves up to 45% of customer queries without human support, and expanded AI across its entire customer service platform.
In a conversation with Peter Yang, Adams explained why designing for AI requires breaking old habits and adopting a mindset of experimentation, speed, and open-mindedness.
Why AI changes the starting point
Intercom has long followed the principle of starting with a clear customer problem. But Adams says that AI has upended that approach.
“We don’t really know what’s possible with AI,” he said. “Sometimes we need to explore what the technology can do first, then find real user problems it can solve.”
That reversal has become necessary in a world where AI capabilities evolve weekly. Rather than locking into a solution too early, teams at Intercom test feasibility first, then anchor their ideas in customer value.
How Fin redefined the chatbot experience
Before large language models, Intercom offered a resolution bot that required heavy manual setup. Adoption was low, even though it could be effective in the right conditions. But once ChatGPT arrived, the team realized they could deliver a better out-of-the-box experience using just a customer’s help content.
“It was a before and after moment,” Adams recalled. “We knew instantly that everything had changed.”
A small, focused team shipped the first version of Fin in four months. With no manual training required, it resolved 25 percent of support queries immediately. A year later, that number had grown to 45 percent, with some customers seeing results as high as 60 percent.
Shifting product principles for an AI world
Intercom traditionally builds around three product principles:
Start with the problem
Think big, start small, ship to learn
Deliver outcomes
In the AI era, Adams believes the second principle is now the most critical. AI products are hard to predict and easy to get wrong. The key is to scope small, ship fast, and iterate based on what users actually experience.
“We are building things we’ve never built before,” Adams said. “The tech is new, the patterns are new, so you have to keep learning.”
He treats strategy docs as living documents, updated frequently based on what the team is learning from experiments.
Why product judgment still matters
Adams is a heavy user of AI tools like ChatGPT for research and writing, but he’s cautious about overreliance.
“The best product people have deep intuition,” he said. “And that comes from talking to customers over many years.”
AI can summarize feedback and surface ideas, but it can’t replace the nuanced understanding that comes from direct conversations. Adams worries that relying too much on AI might erode product judgment and lead to weaker decision-making.
Small teams, fast learning
Intercom favors small, independent teams of 8 to 10 people. Even when building Fin, only a few teams worked on the first version. This allows fast iteration and avoids the coordination drag of large reorganizations.
“You don’t need to reorg the company,” Adams said. “You just need a few curious people building things.”
He believes generalists with curiosity and a maker mindset are best suited to building in this era. AI experience helps, but the willingness to learn is more important.
The new era of software design
Adams sees AI as a change on par with the internet or mobile, possibly even more significant.
“We might be talking about something like the industrial revolution,” he said. “A complete shift in how humanity works and lives.”
His advice to designers and PMs is simple: stay curious, keep building, and be ready to unlearn what used to work.
“The PRD may not be the job of the future,” Adams said. “Trying AI tools on the weekend might be.”
Listen to the full episode here.
Source: Peter Yang
You might also like
