Creating Intelligent Products

Marty Cagan argues that the products most PMs will build in the coming years are "intelligent products" that blend deterministic and probabilistic behavior. He traces the history of AI from 1980s expert systems, whose rule-based approach failed partly because expert physicians relied on probabilistic judgment rather than rules. He contends that probability is central to intelligence, not an edge case, and that B2B and regulated teams dismissing probabilistic solutions are mistaken. He points to examples like the Waymo Driver, recommendation engines, and generative AI tools to show that the future will mostly involve hybrids of both approaches, with probabilistic behavior designed in rather than bolted on.

01Key takeaways

• Treat probabilistic behavior as a core design element of intelligent products, not a bolted-on feature.
• Expect most valuable products to blend deterministic logic with probabilistic models.
• Don't dismiss AI for regulated or mission-critical B2B contexts; probabilistic judgment is what experts already use there.
• Look beyond generative AI to recommendation, translation, and autonomous systems when assessing AI opportunities.
• Learn from history: rule-based approaches struggled because expert judgment is largely probabilistic.

02Key sections

A long view of AI product history

Cagan recounts that his first 1986 article described AI applied to software development, and that the expert-systems approach of that era was built on capturing human experts' rules.

Why rule capture fell short

A Stanford physician interview revealed that experts make educated guesses from probabilities and then test them, showing that rule-based systems could not capture the scale or nature of expert judgment.

Probability as the core of intelligence

Cagan argues that predicting what is most likely, and being mostly right, is what we call intelligence across nearly every domain of expertise.

Examples of intelligent products

He cites Waymo's fleet learning, Google Translate, Spotify, Netflix, and generative tools like Cursor and Shopify Magic as examples of products that blend AI into real value.

Implications for product teams

Teams should adopt a nuanced view that treats deterministic and probabilistic components as complementary, designing probabilistic behavior as a key feature rather than an add-on.

03From the post

“A partnership dedicated to teaching best practices to product teams and product leaders”

04Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.