One of the clearest signals from the Berkeley Haas AI Conference was that traditional product planning is dead. Not dying — dead. Across three separate panels, product leaders from Adobe, YouTube, Intercom, Perplexity, and GitHub Next all said some version of the same thing: if you're building a 3-5 year roadmap in AI, you're wasting your time.
What Changed
Chloe McConnell, Senior Director of PM at Adobe Express, put it most directly: "The roadmap we thought we had in January feels almost irrelevant because things have moved so fast by November."
This isn't the usual startup platitude about being agile. These are leaders at Adobe, YouTube, and Apple — companies that traditionally plan in multi-year cycles. The pace of change in AI has broken their planning frameworks.
Consider: in the last 12 months, Box's CTO noted that 15 major frontier models were released — and each one was arguably the most impressive piece of engineering ever created. Every one of those releases potentially invalidates product assumptions, opens new capabilities, and reshuffles competitive dynamics.
How do you roadmap against that?
What Replaces the Roadmap
The answer across every panel was the same: frameworks and principles, not feature plans.
YouTube's Framework: Fame, Fortune, or Fun.
Heather Christmann shared that YouTube evaluates every creator product through three lenses: does it help creators achieve fame (reach, discovery), fortune (revenue, monetization), or fun (creative satisfaction)? If a product delivers at least one of those AND has daily utility (the Google "toothbrush test"), it's worth building. The specific features change constantly, but the framework is durable.
Adobe's Leapfrog Principle.
Chloe described a planning exercise they use: "Think about what might solve a problem today, but will this be relevant in six months? You're likely gonna leapfrog yourself four times in the next year. Can you skip the interim steps and go straight to the next one?" Instead of building incrementally, they try to anticipate where the technology will be and build for that state directly.
Perplexity's Daily Question.
"Are we staying ahead of everyone by at least a couple of months?" That's it. Not a roadmap — a compass heading. Ship fast, measure constantly, adjust daily.
The Skills That Matter Now
If you're a PM whose career was built on crafting beautiful 3-5 year roadmaps — the kind that used to earn you a senior or principal title — the game has changed. The new planning skills are:
Rapid intuition. Can you quickly evaluate which opportunities are durable and which are hype? Michael Pratt from Apple framed it as being "super tight and aligned on your actual core metrics" — revenue or cost savings, regardless of whether AI is involved.
Iteration speed. Chloe from Adobe talked about the mindset shift: "We have to be really iterative and flexible. It's faster than some of us even expected." The ability to plan in 2-week cycles rather than 2-year cycles is now a core PM competency.
Framework thinking. Instead of predicting specific futures, develop durable decision-making principles. YouTube's fame/fortune/fun, Adobe's leapfrog principle, Snowflake's automate/augment/stay-human — these are the artifacts that replace the roadmap.
Kill speed. As important as shipping fast: knowing when to kill something fast. Multiple speakers mentioned the discipline of evaluating whether a feature will still be relevant when it ships, and cutting it if the answer is no — even if it's 80% built.
What This Means in Practice
I think the practical implication is that AI product teams need to operate more like newsrooms than factories. A newsroom has a permanent editorial perspective (the framework), a rapid daily cycle (ship/learn/iterate), and the ability to pivot completely when the story changes.
The factory model — plan the work, work the plan, measure against the plan — was built for a world where the inputs were predictable. That world is gone.
Annual strategy still has a place. Quarterly goals still make sense. But the 3-5 year roadmap with specific features mapped to specific timelines? That's a fiction now. The best AI product teams have replaced it with something more honest: durable principles, aggressive experimentation, and the humility to admit they don't know what they'll be building six months from now.