by Jordan Fulghum, February 2026
For most of modern business software history, structure lived inside the product. Now it lives inside the business.
As someone who has spent their career at the center of the intersection of product, design, and engineering, my work has always felt central to the companies where I've spent time. Historically, in software, the product is defined by these teams. You build something simple that works for one customer, then you iterate and grow that platform, effectively building to the average, with perhaps some configuration or tier-gated features to capture more of the pricing curve.
For example, customer success was downstream of product. These were teams that helped organizations adapt their workflows to the product. Sales has mostly been downstream of product. They'll funnel feedback up to the product teams, but ultimately their job has been to introduce new ways of working to their customers. When companies adopted the product, they often reorganized parts of their operations to live inside it. Mostly, the product has been the center of gravity.
Lately, I've been noticing a change. As I talk to more and more AI-native B2B founders, I'm seeing the traditional product development cycle get inverted. They assume the customer's existing workflow is the thing to preserve. The job of the system is to wrap around that workflow and make it run better. The existing structure is something to understand, encode, and automate - not redesign.
That means that each sale/instance is unique to the customer. There may be shared infrastructure underneath (orchestration, integrations, prompts, monitoring), but most of the value seems to come from how closely the bespoke system reflects the organization it serves. Put differently, two customers on the same vendor may share some plumbing, but not much else.
That level of tailoring used to be uneconomic. Custom logic meant real cost and permanent complexity - those who have built feature-flags for big enterprise customers, or whitelabeled an existing shared platform, understand this technical debt. In the past, standardization was the fastest way to scale.
Two things happened recently that changed the economics. First, agents can now actually (mostly) replicate the existing workflows. It can be automated. Second, agentic coding systems such as Claude Code or Codex have cut the cost of custom development, and thus technical debt, by a factor of ten. This means you don't have to force everyone into the same shape.
I think this changes what it means to scale a software company.
Traditional SaaS scaled by reproducing the same structured environment across many customers. You invested heavily up front to design the right abstractions, then captured leverage through reuse. Marginal cost per customer approached zero because the product was already built.
These new systems don't scale that way. They scale by repeatedly learning how organizations operate and turning that understanding into working software. The reusable asset is not a fixed product surface. It's the capability to model real workflows quickly and encode them reliably over and over again.
In the classic SaaS model, product managers and designers defined the structure of the system. They chose the abstractions, the boundaries, and the shape of the environment customers would operate within. Scale came from getting those abstractions right.
When the center of gravity moves outward, more structure originates in the field. Product work shifts toward observing, mapping, and translating real, nuanced workflows into reliable logic. Patterns still emerge across deployments and get stabilized over time, but they are discovered rather than imposed.
I suspect the scaling curve will look different. Less front-loaded abstraction and more cumulative learning across many specific implementations. Over time, advantage may come from how quickly a company can understand a new organization and operationalize that understanding.
That starts to look like a hybrid model: part services-oriented dev shop, part traditional SaaS organization. This isn't a new idea - Palantir's forward deployed engineer model has operated this way for years. What's new is the economics. When agents can replicate workflows and agentic coding cuts the cost of custom development by 10x, you no longer need Palantir's scale to make this model work. I bet we start seeing it in much smaller companies, where deployment, learning, and product blur together.
For most of modern business software history, structure lived inside the product. Now it mostly lives inside the business, and software adapts around it.
Follow me on Twitter for more.