AI Made Building Faster. It Didn't Make You More Sure What to Build.

Marty Cagan published The AI Productivity Paradox at SVPG in July 2026. The headline statistic is uncomfortable: 89% of executives say AI has increased the speed of work at their organisation, but only 6% say they are confident they can point to specific, measurable, org-wide AI ROI. Speed up, confidence flat. That is the gap.

Speed is not the scarce input

When you read the 89/6 split, the tempting interpretation is that AI implementation is immature — that if teams just used the tools more consistently, the ROI numbers would follow. Cagan's argument, and the one worth sitting with, is that this misses the point. Output outrunning outcomes is not new. It predates AI by decades. What AI did was make the existing problem faster and therefore louder. You can now ship the wrong thing at a pace that was previously impossible.

What the tooling category is selling

Look at the response from the product tooling market. Productboard launched Spark — an AI layer that helps teams analyse feedback and surface themes. Pendo is building agentic workflows that move work through the product pipeline automatically. These are genuine improvements to execution speed. They deserve credit for solving real coordination pain. But none of them ship the operating model that tells the agent what to optimise for. They accelerate the downstream work. The upstream judgment — which customer segment, which problem, which definition of success — is still assumed to already exist.

• Spark surfaces signal from feedback. It does not decide which signals belong to a strategy worth funding.
• Pendo agents move tasks through a pipeline. They do not define what the pipeline is solving for.
• Every PM tool now has an agent mode. None of them ship the operating model that governs what the agent builds.

Why ROI confidence stays low

ROI confidence is not an AI question. It is a decision-quality question. To be confident that AI investment is generating returns, you need a clear answer to: what were we trying to achieve, and did we achieve it? That question requires a live chain between strategy, discovery evidence, and the work that shipped. Most organisations do not have that chain in a form any tool — human or AI — can actually read. The operating model is missing, diffuse across slide decks and wikis and people's heads. So when the board asks whether the AI investment paid off, nobody can answer with confidence because nobody can trace the decision back to a measured outcome.

The operating model is the missing layer

Cagan and Baxley published A Fresh Definition of The Product Role in August 2026. The three skills they land on are all judgment: knowing what customers value, knowing what the business needs, knowing what is feasible. AI does not provide those judgments. It executes on them — well, at speed, at scale — but only if they have already been made clearly. If the vision, strategy, discovery constraints, and success metrics are sharp and connected, AI becomes a powerful multiplier. If they are vague or missing, AI is a fast path to building confidently in the wrong direction.

What closing the gap looks like

The answer is not a new AI tool. It is a governed operating model — one where the question "why are we building this?" has a live, readable answer that is connected to everything downstream. That means vision and strategy documents that are not slide decks rotting in a drive. It means discovery evidence linked to the bets it informed. It means OKRs that trace back to strategy and forward to the work in the roadmap. When that chain exists, ROI becomes traceable. Not perfectly, not automatically, but systematically — which is all you need to answer the board's question with more than 6% confidence.

Where Delvyn Studio fits

Delvyn Studio is built on the premise that the operating model is a governing chain, not a slide deck. Vision, strategy, discovery, and OKRs are connected in one place, readable by your team and by any agent working on the product. We are not adding another AI layer on top of delivery. We are building the layer that tells the AI — and the product team — what to deliver and why. The 89/6 gap is not a usage problem. It is a governance problem. And governance is what we ship.

AI made building 10× faster. That is real and it is not going away. But fast building without a governing operating model does not produce ROI confidence — it produces faster shipping to the wrong destination. The teams that close the gap will not be the ones with the best AI tools. They will be the ones who decided, before they deployed the tools, what they were building toward.