Wide Stack

The lateral integration of deep domain expertise with the capabilities of artificial intelligence.

For two decades, the highest praise for a developer was vertical: the full stack engineer who could move from database to browser without flinching. That praise belongs to a world where the hard part was the plumbing.

The hard part has moved. It now sits at the seam between what AI can do and what a specific domain actually needs. The people who can hold both — who are deep in artificial intelligence and deep in one domain — are doing a different kind of work. They deserve a different name.

wide·stack · developer

/waɪd stæk dɪˈveləpə/

Noun

A practitioner who holds substantial depth in artificial intelligence and in a specific domain — manufacturing, finance, healthcare, education, law — and works at the seam between them.
Not someone who builds the AI infrastructure itself, but someone who sees where it should be pointed, and translates between what is technically possible and what is actually worth doing.
A counterpoint to the full stack developer: breadth across two depths, not depth through stacked layers.

Fig 01 — Full Stack (vertical)

API

Logic

Database

Infra

depth ↓ through layers

Fig 02 — Wide Stack (horizontal)

AI Depthmodels · agents · evals

Domain Depththe field itself

Wide

depth × depth, side by side

The Manifesto — Five Principles

Two depths beat one stack.

Depth in AI alone produces tools without a target. Depth in a domain alone produces wishes without leverage. The wide stack developer holds both, and that is where the work is.

We translate, we do not transcribe.

We do not move requirements from one document into another. We sit between the model and the field, and decide what is technically possible and what is actually worth doing.

The domain is not a wrapper.

Manufacturing, finance, healthcare, education, law — these are not skins around a generic product. They are bodies of knowledge with their own logic. We learn them seriously, the way an engineer learns a system.

We do not have to build the model.

We do not need to invent the engine to drive somewhere worth going. Aiming the capability is its own discipline. Knowing where to point it is, in many domains, the entire problem.

Judgement is not optional.

The work is to decide what should exist. AI lowers the cost of producing things; it raises the cost of producing the wrong things. The wide stack developer carries the judgement that decides which is which.

Leave an email. You will hear from this manifesto, occasionally, when there is something worth saying.