Marty Cagan says your product coach should be a foundation model. Here is what that actually looks like.

Cagan's thesis: models need your strategic context

In February 2026, Marty Cagan published an article on SVPG arguing that foundation models are ready to serve as personal product coaches. Not generic chatbots - coaches that understand your company's strategy, your product's positioning, and your team's constraints. The key insight: models only become effective product coaching AI when given "project instructions and your company's strategic context." Without that context, you get generic advice. With it, you get coaching that is specific to your situation, your product, and your goals.

This is not a minor point. It is the entire thesis. The raw capability of foundation models is not the bottleneck - the bottleneck is structured context. The team that gives a foundation model the richest strategic context gets the best product coaching. The team that types "help me write a PRD" into a blank chat window gets platitudes.

The connection to spec-driven product management

Spec-driven product management (SDPM) is the methodology of encoding product thinking into structured documents that both humans and AI agents can consume. Vision documents, strategy canvases, OKR hierarchies, persona definitions, discovery findings, and feature specifications - each one a layer of strategic context that makes foundation model product management effective.

• Vision tells the model where you are going and why
• Strategy tells the model what you are prioritizing and what you are deliberately not doing
• OKRs tell the model what success looks like this quarter
• Personas tells the model who you are building for
• Discovery findings tell the model what you have learned from users
• Specs tell the model exactly what to build and the constraints that apply

When Cagan says a model needs "project instructions and strategic context," he is describing the SDPM document hierarchy. The more layers you encode, the better the AI product coaching gets.

What this looks like inside Delvyn Studio

Delvyn Studio is built around the idea that every product decision should flow from strategic context. Vision feeds strategy. Strategy feeds OKRs. OKRs feed discovery. Discovery feeds agent specifications. Specs feed AI coding agents. Each document references the ones above it.

When you ask Delvyn's AI coaching features for help - whether that is validating a key result, critiquing a spec, or suggesting a discovery experiment - the model has access to the full chain. It does not start from zero. It starts from your vision, your strategy, your current OKRs, your personas, and the discovery insights you have already gathered. This is what a Cagan AI product coach looks like in practice.

A worked example: coaching during a discovery spike

A product creator is running a discovery spike on checkout abandonment. They have three competing hypotheses, five user interview transcripts, and a key result that says "reduce cart abandonment from 12% to 8%." They ask the AI coach: "which hypothesis should I test first?"

Without strategic context, a model would give generic prioritization advice. With Delvyn's context chain, the model knows the Q2 strategy prioritizes activation over retention, the persona is a first-time buyer with high price sensitivity, and a previous learning capture showed that payment-method friction outweighed UI confusion 3:1 in the last spike. The coaching response is specific, grounded, and actionable - not because the model is smarter, but because the context is richer. This is why AI agents need product context, not just prompts.

Why this matters now

Cagan is the highest-authority voice in product management. When he says foundation models should be your product coach, the industry listens. But the SVPG AI article leaves an open question: how do you actually give a model your strategic context in a structured way? That is what spec-driven product management answers, and that is what Delvyn Studio implements.

The teams that invest in encoding their product thinking into structured documents today are building a compounding advantage. Every vision doc, every OKR, every discovery finding makes their AI coaching better. Teams that skip this step will keep getting generic advice from generic prompts.