Wardley-Map Your AI Stack Before You Build Another Layer of It

Most build-vs-buy arguments in AI engineering are fought one component at a time. Should we build our own model gateway or use OpenRouter? Write our own agent loop or adopt a framework? Fine-tune or prompt? Each debate gets its own meeting, its own spreadsheet, its own loudest voice. And each one quietly assumes the landscape will hold still long enough for the decision to pay off.

It won't. The ground under your AI stack is moving faster than any stack has moved before. Inference prices for a fixed capability level have been falling somewhere between 9x and 900x per year depending on the capability threshold you track — GPT-3-level output that cost $60 per million tokens in late 2021 now costs pennies. Agent frameworks have gone through multiple breaking rewrites in three years. The gateway layer went from "we should build one" to a crowded commodity market with managed and self-hosted options in about eighteen months. Deciding component-by-component in a landscape like this is navigating a river by staring at your feet.

There's a sixty-year-old tool built for exactly this problem: the Wardley map. It won't tell you what to build. It will tell you something more useful — which of your components are about to become someone else's cheap product, and which are moving the other way.

The Map: Two Axes, One Uncomfortable Question

A Wardley map plots every component of your value chain on two axes. The vertical axis is visibility to the user: your product experience sits at the top, GPUs sit at the bottom. The horizontal axis is evolution: every component drifts, left to right, through four stages — genesis (novel, uncertain, hand-built), custom-built (understood but bespoke), product (rentable, differentiated vendors), and commodity (standardized, utility-priced, boring).

The one rule of the evolution axis is that movement is one-directional. Components don't return to genesis. Electricity, compute, payments — everything that was once a heroic in-house engineering effort became a line item on an invoice. The strategic question a map forces you to ask about every box is not "can we build this?" but "where is this on the axis, and how fast is it moving right?"

What makes the map worth drawing for an AI stack specifically is the speed. Traditional infrastructure took decades to commoditize. Databases took roughly twenty years to go from research artifact to rentable utility. The layers of the AI stack are making the same journey in eighteen to thirty-six months. Any component you decide to hand-build today should be evaluated against where the axis will have carried its commercial equivalents by the time your version reaches production quality — not against where they are now.

Where the AI Stack Sits on the Axis Today

Draw the map for a typical AI product in mid-2026 and the layers sort themselves with surprising clarity.

Models are sprinting from product to commodity. Frontier capabilities still command premium pricing at the moment of release, but the premium window keeps shrinking. Epoch AI's tracking shows that the price to achieve a fixed benchmark score falls at rates up to several hundred-fold per year for capability levels that have been surpassed, and Gartner projects that inference on a trillion-parameter model will cost providers over 90% less by 2030 than it did in 2025.

Multiple providers now offer near-interchangeable APIs with low switching costs. Whatever capability you're paying for today will be a commodity — or free, via open weights — within a couple of years. Anything you build that assumes model access is scarce or expensive is built on melting ice.

Gateways and routers are already commodities. The "one API over many models" layer has fully evolved: managed services route across hundreds of models from dozens of providers, and open-source proxies give you the self-hosted version for the cost of a small VM. The economics are utility economics — single-digit-percentage fees or infrastructure costs, competing on reliability and compliance checkboxes rather than capability. If your platform team is building a bespoke model gateway in 2026, they are hand-crafting a commodity. The map says: stop.

Orchestration frameworks are still churning through genesis and custom-built. This is the layer where the axis position is most commonly misread. Agent frameworks feel like products — they have brands, docs, conference talks — but their churn rate betrays their real stage. The dominant framework of the 2023 prototyping wave went through breaking restructurings across three minor versions before committing to stability in a late-2025 rewrite, and meanwhile the dominant pattern shifted twice: from chaining, to graph-based workflow orchestration, to tool-calling-in-a-loop with filesystem and memory access. Many production teams have quietly migrated to vendor SDKs or raw API calls. Genesis-stage components punish deep integration; the correct posture is thin adoption with an exit path, not building your own and not marrying someone else's abstractions.

Loading…