Wardley Mapping for AI Products: Visualize Your Strategy Before the Market Shifts

What Wardley Mapping Is and Why AI PMs Should Care

A Wardley Map is a two-axis visual that shows your value chain (the components your product depends on) plotted against their evolutionary stage (how mature they are in the market). The x-axis runs from Genesis (invented, novel, custom) through Custom Built, Product, to Commodity/Utility. The y-axis shows visibility to the user, with user-facing components at the top and infrastructure at the bottom.

Simon Wardley developed the technique at Canonical in the late 2000s after noticing that most strategy frameworks describe what to do without telling you where you are. A Wardley Map forces you to answer: where is each component in its evolution, and therefore, what strategies make sense for it?

Why it matters for AI products specifically

AI capabilities are evolving faster than any previous technology. LLM APIs went from Genesis (GPT-3, 2020) to near-commodity in roughly four years. A strategy built on a capability in the Product stage looks very different from one built on a Genesis capability. Most AI strategies fail to make this distinction.

The core insight Wardley provides

Components in the Genesis and Custom stages reward innovation and differentiation. Components in the Product and Commodity stages reward efficiency and cost management. Treating a commodity capability as strategic is how you waste engineering resources. Wardley Maps make the misclassification visible.

What it catches that frameworks miss

Porter's Five Forces and Blue Ocean Strategy describe competitive positioning but not the direction of change. Wardley Maps show movement: components inevitably evolve left to right, toward commodity. You can plot where things are heading, not just where they are.

The AI PM use case

Use a Wardley Map before a quarterly roadmap, a major build vs buy decision, a new product area, or a pricing strategy review. Any decision that depends on the maturity of your AI capabilities benefits from one.

The Four Evolutionary Stages Applied to AI Capabilities

Every AI capability your product uses sits somewhere on the evolution curve. Identifying the stage correctly is the whole exercise. Here is how each stage maps to what you actually see in the AI market.

Genesis: Novel, custom, uncertain

Current AI examples: Agentic computer use at reliable production quality. Sub-second multimodal reasoning. Long horizon planning agents. Physical AI that executes fine motor tasks.

Strategy: Invest experimentally. Small, time-boxed bets. If your product genuinely lives here, you have a window before competition arrives, but the window is short and the risk is high. Do not bet the company roadmap on a Genesis capability without a fallback.

Custom Built: Understood but still hand-crafted

Current AI examples: RAG pipelines tuned to proprietary data. Fine-tuned models on domain-specific corpora. Multi-agent workflows for specific enterprise processes. Evaluation frameworks for niche domains.

Strategy: Build if this capability is core to your differentiation. Outsource if it is not. Custom Built is where most of the AI PM build vs buy decisions live. The question is: is this the unique insight your product is built around, or is it a dependency?

Product: Packaged, sold, benchmarked

Current AI examples: Foundation model APIs (GPT-5.6, Claude Opus 5, Qwen 3.8 Max). Vector databases. Prompt management platforms. Standard RAG retrieval. AI guardrails and content moderation APIs.

Strategy: Buy, not build. Do not build your own vector database in 2026. Do not write your own content moderation pipeline. These are solved problems. Building them costs engineering resources that should go to the novel layers above them.

Commodity/Utility: Undifferentiated, invisible infrastructure

Current AI examples: Text embedding. Basic sentiment classification. Simple named entity recognition. Standard OCR. Token counting and context management.

Strategy: Automate, consume, ignore. These capabilities cost near zero, change rarely, and confer no competitive advantage. Do not staff engineers to manage them. Use the cheapest, most reliable provider and move on.

How to Build Your AI Product Wardley Map

You do not need specialized software. A whiteboard, a spreadsheet, or a shared Figma file works. The value is in the thinking, not the tool.

Step 1: Start with the user need at the top

Write your primary user need at the top of the y-axis. For a coding assistant: 'Developer writes and ships code faster.' For a customer support AI: 'User gets accurate answers without waiting.' This anchors everything below.

Step 2: Map the value chain downward

Ask: what does the product need to deliver this user need? List every component: the UI layer, the orchestration layer, the model layer, the retrieval layer, the data layer, the compute layer. Each component that depends on another sits below it.

Step 3: Place each component on the evolution axis

For each component, ask: is this novel and custom, or is this a commodity anyone can buy? Most AI product maps show that the model layer is Product stage, the data and fine-tuning is Custom Built, and the core insight or workflow is Genesis. Place accordingly.

Step 4: Add competitor positions

Where is your competitor's version of each component? If they are buying what you are building, or vice versa, the map reveals a strategic mismatch that needs examination.

Step 5: Draw movement arrows

Every component moves right over time. Add arrows showing where you expect each component to be in 12 to 18 months. This is where the map becomes a planning tool rather than a snapshot.

Go Deeper in the AI PM Masterclass

The masterclass covers strategic frameworks for AI products, including when and how to apply tools like Wardley Mapping to real product decisions. Taught live by a Salesforce Sr. Director PM.

Strategic Patterns AI PMs See in Their Maps

Once you start mapping AI products, you see recurring patterns. Recognizing them early is the whole point of the exercise.

The Commoditizing Moat

Your differentiation is built on a capability that is rapidly moving from Custom Built to Product. Copilot-style code completion was a moat in 2022 and a commodity by 2024. Map movement arrows to see if your moat is 12 months from being a commodity.

The Invisible Dependency

A commodity component you assumed was stable suddenly becomes unavailable or changes pricing. Your AI product has a hidden dependency on it. Mapping forces you to see every component, including the infrastructure ones you assumed were invisible.

The Wrong Layer War

Two competitors are fighting over a Product-stage capability while the Genesis stage above them is uncontested. Classic example: two teams building better RAG retrieval while the actual moat is in the orchestration layer above. The map shows where the real battle is.

The Build Trap

The team is building something that is already in the Product stage, available for purchase. They are reinventing the wheel because the map was never made explicit. Wardley Maps surface this before sprint planning, not after six months of engineering.

The Timing Window

A Genesis capability is about to industrialize and become a Product. Teams that build on it now, before it commoditizes, can capture the transitional period when others are still building custom versions. Spotting these windows is where the map pays off most.

The Platform Trap

Your product depends on a capability controlled by a single vendor in the Product stage. When that vendor raises prices or changes terms, you have no fallback. The map reveals single points of failure in your value chain before they become crises.

Where AI Capabilities Sit on the Map in Mid-2026

Here is a reference positioning of major AI capabilities as of August 2026. Your specific product context will shift these, but this gives you a starting point for your own map.

Foundation model APIs (GPT-5.6, Claude Opus 5, Qwen 3.8 Max)

Product stage

Multiple competitive providers, benchmark parity emerging, price war underway. Do not build your own. Buy and route intelligently.

RAG pipelines with standard vector search

Product to Commodity

Commoditizing rapidly. Pinecone, Weaviate, pgvector, Qdrant. Use a managed service unless your retrieval problem is unique.

Domain-specific fine-tuning on proprietary data

Custom Built

Still requires significant craft. This is where product differentiation lives for many AI products. Worth building if your data is genuinely unique.

Multi-agent orchestration for novel workflows

Genesis to Custom Built

Frameworks exist (LangGraph, LlamaIndex) but production reliability at scale is still hand-crafted. This is currently the most valuable layer to own.

AI evaluation frameworks and evals

Custom Built to Product

Braintrust, Patronus, Arize are emerging as products. General evals are productizing; domain-specific evals are still custom built. Plan to transition.

AI guardrails and content moderation