AI Returns Agency to Organizations

The Shift

For most of the software era, the organization came second.

Not in principle. Every software project began with some version of a business need. But in practice, the shape of the available software exerted enormous influence over the shape of the work. An organization chose a platform, implemented its model, and learned how to operate within it. The more broadly useful the platform was, the more likely it was to come with assumptions about how customers, approvals, teams, records, and processes should work.

Those assumptions were not arbitrary. They were how software became affordable.

Writing and maintaining custom software was expensive. Packaged products spread that expense across many customers. Standard processes made support, training, upgrades, and integrations more manageable. For most organizations, it made sense to adopt a capable system and accept the compromises that came with it.

But the compromises accumulated.

AI is beginning to change the cost of making software fit. That does not mean every organization will suddenly build every system from scratch. It means that more of the software that shapes daily work can be explored, adapted, and created around the organization itself.

That is a return of agency.

The direction has been changing

It helps to name the old pattern plainly:

Software → organization

Software did not merely support the business. It often organized it.

The system defined the available categories. It made some paths easy and others expensive. It offered one way to represent a customer, a case, an order, or an approval. It placed boundaries around what could be customized, reported on, or connected to another part of the organization.

Of course people adapted creatively. They added configuration, custom fields, reports, integrations, spreadsheets, and exceptions. They learned the system’s language and translated their work into it. In many cases, that translation was worthwhile.

Still, the direction of adaptation was clear. The organization’s operating model was continually shaped by the practical limits of the software it could buy and maintain.

AI makes a different direction possible:

Organization → software

An organization can begin with the way it actually works—the concepts it uses, the decisions it makes, the relationships that matter, and the behavior it needs—and shape the software, agents, automations, and reporting that serve that work around its understanding.

The shift is subtle but consequential. It moves the question from “How should we configure this system?” toward “What systems would faithfully express the way we intend to operate—and how should they act, automate, or report on our behalf?”

Agency is more than customization

It would be easy to hear this as a new version of customization. Enterprise software has always offered configuration. Organizations have always commissioned bespoke applications. Low-code tools and internal development teams have long filled the gaps between what a package offers and what a business needs.

Those things matter, but they are not quite the change I mean.

Customization happens inside the boundaries of a product. It lets an organization tune a workflow, add a field, or extend an object model. It is valuable precisely because someone else has already made the larger set of decisions.

Agency begins one level earlier. It is the ability to participate in defining the system itself.

That includes questions that are usually settled before configuration begins:

• What are the core concepts in this part of the business?
• Which distinctions are important enough to preserve?
• What can happen, and under what conditions?
• Where does one team’s responsibility end and another’s begin?
• Which exceptions are genuinely exceptional, and which reveal a weak model?

When an organization can work through these questions and carry the answers into software, agents, automation, and reporting, it is not merely tailoring a tool. It is making its own operating model more explicit and more usable.

AI lowers the cost of trying to do that. A team can turn an emerging understanding into a prototype, inspect the result, find the assumptions it missed, and change course. It can build a workflow that reflects a particular process without first winning a large, irreversible implementation project.

The result may still use packaged systems. It may still rely on standard platforms and shared infrastructure. Agency is not a demand for purity or an argument that every organization should own every line of code. It is the growing ability to choose deliberately where the organization should adapt to a standard and where the software should adapt to the organization.

The knowledge that used to disappear

Every organization has operational knowledge that does not appear cleanly in a process diagram.

It lives in the way experienced people recognize a risky exception. It lives in the order in which work actually happens, rather than the order the official procedure says it happens. It lives in the distinction between two customer situations that look identical in a generic system but carry very different consequences for the people doing the work.

Some of that knowledge should be challenged. Workarounds can hide unnecessary complexity or reproduce a problem that no longer matters. But some of it represents hard-won learning about how the organization creates value, manages risk, or serves its customers.

Historically, an implementation often required organizations to decide which of those distinctions could survive the move into a new system. The rest might be represented in a note, carried in someone’s head, or lost. The decision was frequently reasonable: preserving every nuance was simply not worth the cost.

As the cost of expression falls, that calculation changes. More of the useful nuance can be made visible, tested, and represented in the tools people use. An organization can preserve its learning without pretending that all of its habits are sacred.

That is what makes this a strategic shift rather than a productivity story. It is not only about building faster. It is about giving an organization a greater ability to decide what parts of itself should endure as its systems change.

The responsibility that comes with choice

Agency is appealing because it sounds like freedom. But freedom from a fixed system is also freedom from a set of decisions someone else made for you.

Once a team can generate a system quickly, it has to decide what the system should mean. If a customer moves from one state to another, what is actually true? If an exception appears, who owns it? If two teams describe the same business concept differently, which definition should guide the software, an agent’s response, an automation, or a report?

AI can help make those questions visible. It can propose a model, generate examples, and show the consequences of a choice in a working prototype. It cannot resolve the underlying organizational question on its own.

In fact, the easier software becomes to produce, the easier it becomes to hide unresolved questions inside it. A system can look complete long before the people responsible for it share an understanding of what it is meant to do.

This is why agency requires more than access to capable tools. It requires judgment, and judgment requires shared understanding.

The organizations that benefit most from AI will not necessarily be the ones that generate the most software. They will be the ones that can use generated systems, agent behavior, automation, and analysis as part of a disciplined learning process: make an assumption visible, test it with the people affected by it, refine the model, and carry that learning forward.

A different role for technology

This changes how I think about the role of software in an organization.

The goal is not a perfectly bespoke system that captures every detail of the business. That would turn flexibility into another kind of constraint. The goal is to give the organization a living way to express the parts of its operating model that matter, while continuing to use standards where standards are useful.

Technology becomes less of a fixed environment that the organization inhabits and more of a medium through which the organization can continue to clarify and evolve how it works.

That is a much more demanding role for the organization. It asks people to articulate assumptions that were previously invisible. It asks them to negotiate differences rather than simply hand requirements to a technical team. It asks them to treat understanding as something that must be developed and maintained, not collected once at the beginning of a project.

But it is also a more promising role. It gives organizations a chance to build systems that carry forward their best learning rather than repeatedly forcing that learning to start over.

The next question is important because it keeps this argument honest: if SaaS and standardization created real value, what exactly has changed—and what has not?