AI-Assisted Roadmaps: Where Hype Meets Reality
I've used AI to help prioritize features. It works great—until you realize it doesn't understand your users.
Everyone's talking about AI agents building roadmaps now. I get it. The appeal is obvious: throw your feature requests into Claude, get back a prioritized list, ship faster. Done.
I tried this on a Seven CMS customer project last month. We had ~40 feature requests, no clear prioritization, and a PM who was drowning. I fed everything to an AI with some context: user segments, revenue impact, implementation cost. The output was clean. Organized. Ranked.
It was also wrong in three different ways.
The AI ranked a complex integrations feature higher than bug fixes in the admin panel. Why? Because I didn't tell it that the admin panel handles 10x more daily users. The AI optimized for what looked good on paper—high-value enterprise clients requesting integrations—not for what would reduce support tickets by 30%.
Here's what I learned: AI is an excellent second opinion, not a decision-maker.
The real value came when I used the AI output as a starting point. Our PM reviewed it with actual usage data, looked at churn metrics, and reordered things. That took an hour. The AI saved us from a blank page—the mental paralysis—not from hard thinking.
For my own projects (Seven Shop, Seven Suite), I now use Claude differently. I ask it to:
- Help structure how we think about prioritization
- Identify gaps in our reasoning
- Organize what we already know into frameworks
But the actual decisions? Those still need humans who understand the business.
The problem with "AI agents built our roadmap" narratives is they skip the part where someone spent time feeding the AI good inputs. Garbage in, garbage out still applies. You can't automate away domain knowledge.
If you're considering this: sure, experiment with AI-assisted prioritization. It's genuinely helpful for structure and avoiding cognitive biases. But treat it like a code review tool, not a compiler. The output needs human judgment.
And if anyone tells you their AI built their entire product strategy with zero human input? They're either exaggerating or they didn't notice the mistakes yet.