Building an AI-First Product Organization | Paul Adams (CPO Intercom)
Why AI challenges traditional PM beliefs and how Intercom built an AI agent that can resolve 50% of customer support questions by itself
Dear subscribers,
Today, I want to share a new episode with Paul Adams (CPO Intercom).
Watch now on YouTube, Apple, and Spotify.
Paul and I had a great conversation about:
(00:00) Will AI make PMs dumber and lazier?
(09:00) The rise of "AI PMs" and how to learn this stuff without the BS
(21:05) Why customer support mostly sucks and how AI will fix it
(33:00) Why AI challenges "start with the customer problem" and other principles
(38:26) Think big, start small, ship to learn
Keep reading for the edited interview takeaways.
How PMs and designers can skill up on AI
Welcome, Paul! To start, what are your favorite AI use cases?
I rely heavily on ChatGPT to:
Do research on new topics, including understanding AI itself (e.g., “Explain retrieval augment generation”).
Summarize info such as detailed research and posts into a concise list of takeaways.
With how good AI is becoming, can it make PMs like us lazier and dumber?
Ha! I do worry about that. AI will further separate great PMs from mediocre ones.
The best product people have built deep product sense by talking to customers over many years.
Using AI to summarize user interviews just isn’t the same as sitting in on each interview. There are no shortcuts to direct customer interaction.
There’s a lot of hype right now, with many people claiming to be “AI PMs.” What are some good ways to learn about AI without the BS?
People always gravitate to buzzwords at the start of new tech waves. However, “AI product” and “AI PM” are meaningless because this tech will be used in all products—it’s just like “mobile first.”
To genuinely skill up on AI, I recommend:
Reading AI papers and practical guides.
Trying AI products to build hands-on intuition.
Talking to AI/ML colleagues to learn the tech.
When hiring, I look for people who are curious, hands-on, and open-minded about AI products than those who only have traditional credentials.
How to build an AI-first product organization
What have you found non-intuitive about building AI products vs. traditional products?
So, we have 3 product principles at Intercom:
Start with the customer problem. Talk to customers to understand if the problem is real and whether they want you to solve it.
Think big, start small, ship to learn. Have a vision, but ship the smallest solution first and iterate based on what you learn.
Deliver outcomes. Instead of only focusing on shipping features, deliver outcomes that help the customer and grow the business.
However, AI has challenged my thinking around these principles. For example, when it comes to starting with the customer problem:
Because AI is moving so fast, it’s sometimes better to explore what the tech can do before mapping it back to a customer problem.