A roadmap for AI adoption in product development.
In the previous post, we looked at how AI amplifies the practices you have in place — good or bad. Feature factories get faster. High WIP gets higher. The “yes man” PO gets a backlog that looks immaculate and means nothing.
What does effective AI adoption actually look like? This post lays out a practical roadmap — four foundations that need to be in place for an effective AI adoption in product development.
A roadmap, not a checklist
These four foundations aren’t steps to tick off in sequence. They’re interdependent — each one makes the others more effective. You can start anywhere, but the further you progress across all four, the more value AI will deliver.
Think of it less as a setup process and more as a description of what a mature, AI-ready product practice looks like.
Foundation 1: Build a Unified Knowledge Platform
Every product team generates knowledge continuously — through strategy workshops, user interviews, discovery sessions, roadmap reviews, backlog refinement. The problem is that this knowledge rarely lives anywhere coherent. It’s scattered across slide decks, Miro boards, Confluence pages nobody maintains, and the heads of people who’ve been on the team longest.
This creates two problems. First, your team operates without a shared, current understanding of the product. Decisions get made on assumptions, context gets lost when people leave, onboarding takes months. Second, when AI enters this environment, it has nothing real to work with. It fills the gaps with plausible-sounding content that has no grounding in your actual product context.
The answer is what I call a Unified Knowledge Platform — a consolidated, living repository of product knowledge that spans the entire development lifecycle: strategy, discovery, ideation, and delivery. Not a static document, not a dump of meeting notes, but a maintained source of truth that both your team and your AI tools can draw from.
This isn’t a new tool you need to buy. It’s a practice — the discipline of documenting what you know, keeping it current, and making it accessible. What matters is that it exists, that it’s maintained, and that it’s the place everyone goes when they need context.
In the next post, we’ll go hands-on — building a Unified Knowledge Platform and connecting it to an AI Product Agent. For now, the key point is simple: without it, every AI prompt you write lacks the context to produce anything useful.
Foundation 2: Embed your practices into your AI tools
AI tools are generalists. Out of the box, they know nothing about how your team writes user stories, what your definition of done looks like, how you structure a roadmap, or what splitting patterns you follow.
If you want AI to produce outputs that fit your workflow, you need to give it that context explicitly — through system prompts, custom instructions, or templates. Your AI Product Agent can then generate user stories in your format, draft roadmap items that align with your goal-based approach, and ask the right clarifying questions before generating anything rather than making assumptions.
The practices and conventions you’ve documented become the skills you build into your AI Product Agent.
Foundation 3: Develop your AI evaluation skills
AI generates confident, well-structured, plausible output. That’s exactly what makes it dangerous without the right skills to evaluate it.
Get Ricardo Maia’s stories in your inbox
Join Medium for free to get updates from this writer.
Working effectively with AI requires a specific set of competencies: guiding it with the right context and constraints, spotting hallucinations and incorrect assumptions, identifying when output is misaligned with product strategy, and knowing when to override it entirely.
This is a shift in role for everyone working with AI — moving from producer of artefacts to curator and evaluator of AI-generated work. It requires more judgment, not less. AI doesn’t reduce the need for expertise — it makes expertise more important, not less.
Foundation 4: Define clear policies for AI use
Without clear guidance, teams default to one of two extremes: using AI for everything without questioning outputs, or avoiding it entirely out of uncertainty.
Clear policies eliminate that ambiguity. They define approved use cases, establish boundaries in sensitive or high-stakes contexts, and clarify what data can and cannot be shared with AI tools. They also cover ethics — fairness, transparency, and accountability in AI-assisted decisions.
This doesn’t need to be a lengthy governance document. Even a simple, shared agreement on how your team uses AI — what it’s for, what it’s not for, and what always requires human judgment — removes the friction that gets in the way of confident adoption.
Four foundations, one starting point
Getting value from AI in product development isn’t about finding the right tool. It’s about having the right foundations in place. The Unified Knowledge Platform gives your agent context. Embedded practices give it skills. Evaluation competencies ensure you stay in control. Clear policies ensure the team uses it confidently and responsibly.
Without these four foundations, AI adoption stays shallow — isolated productivity gains here and there that have little impact on the organisation as a whole. With them, AI stops being a collection of individual tools and becomes a coherent capability that changes how your team builds products.
This is the difference between AI that delivers real, organisation-wide value and AI that merely keeps individuals busy.
This is the second post in the AI for Product Owners series. In the next post, we’ll go hands-on — building an AI Product Agent that embedds you organisations frameworks, guidelines and templates and connecting it to a Unified Knowledge Platform.
Missed the first post? Start here: AI won’t fix a broken product practice. It will accelerate it.
