THE MOVING FRONTIER · SEPTEMBER 2026 RETROSPECTIVE
Features made sense when applications owned the journey
For much of enterprise software history, companies differentiated themselves through modules and features. A richer dashboard, a better editor, another automation rule: each could be packaged, demonstrated, and sold.
Intelligent agents challenge that unit of value. A customer trying to close a financial period, onboard a supplier, or investigate a service incident does not care which application’s feature completes each step. The customer cares whether the job gets done correctly and can be defended later.
The workflow—not the individual feature—starts to look like the more natural unit of product design.
A workflow has a beginning and an end
Calling something a workflow does not make it one. A genuine workflow has an initiating event, necessary context, a sequence of decisions and actions, an accountable owner, and a completion condition.
Take a customer-support escalation. The work may require a ticketing system, internal documents, product telemetry, account history, a communication tool, and a final decision about remediation. A model that summarizes the ticket is useful. A system that responsibly connects those elements could change the economics of the whole process.
This is the distinction between adding intelligence to a feature and designing a system of work.
The new architectural questions
Once software crosses application boundaries, product teams inherit difficult choices. Where does the canonical state live? Which system grants permissions? What happens if the agent completes four steps and fails on the fifth? Can a person inspect the chain of actions and resume from an intermediate state?
These are not only engineering concerns. They determine whether the experience feels coherent and trustworthy. An agent that acts across five tools but leaves no clear owner of state is fragile by design.
The product brief should identify the system of record, the orchestrator, the review surface, and the recovery path. Those roles may be played by one platform or several, but the boundaries must be explicit.
Where value and pricing move
Seat-based pricing made intuitive sense when value came primarily from people using a software application. As software performs more of the work, buyers may increasingly evaluate the cost and quality of completed tasks.
That does not mean every AI product should charge per outcome. Outcomes are often hard to define, depend on factors outside the vendor’s control, and can create perverse incentives. A hybrid approach combining access, usage, and carefully measured value may be more practical.
The important shift is that economic value becomes easier to discuss at the workflow level: time reduced, errors avoided, cycle time improved, or capacity increased.
Designing the new boundary
The winners will not simply stitch every application together with a conversational interface. They will understand where coordination breaks down, what evidence must travel with the work, and which decisions deserve human ownership.
When an application becomes part of a larger workflow, its relevance depends on the value it contributes to the outcome. The feature does not disappear. It becomes a component of something bigger.
Editorial note: This is a retrospective essay written in November 2026. September 2026 describes the period explored, not its original publication date.
