Context

Why this matters

Melissa Perri, host of the Product Thinking Podcast and founder of Product Institute, surveyed 309 product leaders across 40 countries with the practitioner community Product Circle for the State of AI in Product 2026 report. The question wasn't which AI tool teams use, it was whether AI is actually changing how product decisions get made. This is a solo, data-driven episode, not an interview, and Melissa also announces the show is pausing new episodes after this one.

The Big Idea

AI adoption is nearly universal, but it hasn't fixed how product organizations make decisions, because AI multiplies whatever operating model already existed rather than replacing it.

Across the 309 leaders surveyed, adoption of AI tools tops 85%, yet only 36% say it's strengthening their operating model, and the largest organizations, despite investing the most, convert AI into results the least.

Key Insights

Tools changed, decisions didn't

87.7% of respondents use AI coding assistants, 85.4% use AI for research, writing, or analysis, 71.5% have built internal AI tools, and 69.9% have shipped AI-powered features. But only 36% say AI is strengthening their product operating model, about a quarter say it's exposing weaknesses that were already there, and 6% say it's making things actively worse.

The bottleneck moved upstream

About half of respondents say AI is having high impact in engineering, and 45% say the same for design and prototyping, but the number drops fast for strategic planning, QA, customer research, and cross-team collaboration, all single or low double digits. One surveyed PM put it directly: delivery of designs and code got fast, and delivery of good decisions became the new bottleneck.

AI multiplies, it doesn't equalize

Organizations that rated their operating model as mature before AI are 1.7 times more likely to say AI is strengthening it, and three times less likely to say it's making things worse. Teams of 1 to 50 people report AI strengthening their operating model at 48%, while organizations of 500 or more collapse to 20%, despite investing more in training, hiring, and tooling.

Execs and PMs disagree on strategy

62% of product managers say the lack of a clear AI strategy at the top is a big challenge, versus only 19% of C-level respondents, a 43-point gap. Melissa's read: the strategy likely does exist at the investment level where the CEO and board operate, but it hasn't been translated into the operating rules a PM needs on a Monday morning, what to use AI for, what not to, who reviews the output.

Leaders fear moving too slow

35% of respondents said their biggest fear is moving too slow and falling behind, versus 19% who feared moving too fast without enough guardrails. Over 28% named a third, different fear: overinvesting in AI tools without changing how the organization actually works, the exact pattern Melissa says the survey's own data already shows playing out at scale.

Mental Models & Frameworks

AI as multiplier, not equalizer

Whatever operating model existed before AI determines what AI does to it: mature, healthy models get stronger, weak ones get exposed or get worse. Before making any AI investment, audit the existing decision-making and review process rather than assuming a new tool will fix a broken one, since a low-maturity team investing heavily in AI tends to end up worse off than a high-maturity team investing lightly.

Strategy needs a translation layer

An AI strategy that exists only at the executive or board level isn't actually operating until someone translates it into concrete rules: what to use AI for, what to avoid, who reviews AI-assisted output, how it changes review cadences. If a PM can't state these rules for their own work, the strategy hasn't reached them yet, regardless of what leadership believes about its existence.

Common Mistakes

Measuring adoption instead of outcomes

Treating AI adoption rate as the success metric, when adoption is only the input, not the goal. Melissa's recommendation is to measure cycle time, decision quality, and customer insight velocity instead, since the point of adopting AI was always to move those outcomes, not to maximize how many people opened the tool.

Overinvesting in tools without changing work

Pouring budget into new AI tools while leaving the underlying workflow, review cadence, and decision process untouched. This is the fear over 28% of surveyed leaders named directly, and the survey's own size-based data, large organizations investing the most while converting the least, shows this pattern already producing weak returns.

Practical Application

Audit your decision workflow first

Before buying another AI tool, map how decisions actually move in your org: who reviews what, how customer signal reaches the roadmap, where prioritization actually happens. Make that operating model explicit before layering AI on top of it, since AI will only make a broken process visible and faster, not fix it on its own.

Translate strategy into daily rules

If leadership has an AI strategy, turn it into guidance a PM can use this week: what to use AI for, what to avoid, who reviews AI-assisted output, and how it changes prioritization. A strategy that lives only in a boardroom deck isn't actually operating, no matter how real it is at the top.

Replace adoption metrics with outcomes

Stop reporting how many people use an AI tool as a success metric. Track cycle time, decision quality, and customer insight velocity instead, the outcomes AI adoption was supposed to produce in the first place, and the numbers that actually tell you whether the operating model improved.

Train product, design, engineering together

Run AI training across functions instead of separate, tool-specific sessions per team. Melissa's survey data shows function-specific training widens the fluency gap between roles, while training product, design, and engineering together on the same AI workflows compresses that gap instead of deepening it.

Questions to Consider

• Are we measuring AI adoption as a success metric, when cycle time, decision quality, and customer insight velocity are what actually indicate whether AI is helping?
• Does our AI strategy exist only in a leadership deck, or has it been translated into operating rules a PM could state for their own work this week?
• If our decision-making process was weak before AI, are we assuming AI will fix it, when this survey's data suggests it's more likely to expose or worsen it?
• Are we training product, design, and engineering on AI tools separately by function, and if so, is that widening the exact fluency gap Melissa's survey found?

Bottom Line

AI adoption is nearly universal and still hasn't closed the gap between fast delivery and good decisions, because AI multiplies whatever operating model and strategy translation already existed rather than creating either one from scratch.

Audit your own decision-making workflow before buying the next AI tool, and measure outcomes, not adoption.

Resources Mentioned

Notable Quotes

"Delivery of designs and code got very fast. Delivery of good decisions became the new bottleneck." (a surveyed product manager, quoted by Melissa Perri)

"The next AI investment in your org is not another tool. It's a workflow redesign." (Melissa Perri)