For centuries, process has been the organizing principle, the primitive of Enterprise. That era is about to be over.
From a business perspective, a process is “a repeatable sequence with a trigger, inputs, activities, decisions, and owner.” Its power has been repeatability: we optimize the steps from A to Z so that a person or a computer can deliver the same outcome again and again. Entire industries, jobs, and software have been built around capturing and managing that sequence.
Even in 2026, AI-native companies are being built to deploy agents over the top of process: replacing a person with a large language model while keeping the sequence largely intact. There are good reasons to do this in the near term. Existing processes encode knowledge, controls, and the way enterprises operate today.
But treating those processes as the destination is short-sighted. It asks the enterprise to transform twice: first, rebuild workflows around agents; then, rebuild again around goals, rules, and agents that determine how to achieve them. We are still asking a new means of execution to follow the same sequenced steps over and over to deliver an outcome.
The problem is that the work processes exist to serve is dynamic. As things change upstream or downstream in an Enterprise, processes need to be reengineered, re-architected, optimized, changed, or removed. For process people, there has always been more work to do.
This way of thinking led vendors and businesses alike to treat process as a first-class citizen: a static pipe through which work flows. But the pipe is not the business. It is one way of organizing work under a particular set of conditions.
As we move towards autonomous Enterprises, it is tempting to assume that process will remain central. After all, process helps us get the job done, and agents can now decide, triage, label, action, send, and draft. Why not have them follow the same steps as their human counterparts?
Because end-game business autonomy is not simply the automation of process. An agent should not inherit a workflow merely because a human needed one.
There are several ways to think about AI’s role in enterprise work:
- An AI assistant retrieves information for a person, who still executes the task or workflow.
- An agent prepares part of a workflow for a person to complete.
- An agent performs large parts of a business-defined workflow, with a person approving key steps.
- Agents execute a predefined process without direct human intervention.
- Agents resolve towards goals without a predefined sequence.
These are different models, not necessarily steps every enterprise must pass through. The final model changes what we give the agent. It needs three things:
- The goal: the outcome that represents the unit of work.
- The artifacts: what is produced as a byproduct or deliverable of resolving that goal, such as an invoice, a report, or a record creation.
- The rules: the governance, policy, and exceptions that govern the work.
Those three things depend on a stateful representation of the environment in which the agent operates: the Enterprise, evolving as the Enterprise changes.
Why should the sequence we designed for a human remain the sequence we prescribe to an agent?
DeepMind’s AlphaZero achieved superhuman chess performance through millions of games of self-play. Given only the rules of chess, it learned strategies through reinforcement learning and search, without training on human games. It subsequently defeated Stockfish, one of the (if not the) world’s strongest chess engines.
The sequence of moves was something it discovered.
This points towards a different model for enterprise work. Define the outcome, provide the current state of the business, establish the boundaries, and evaluate the result. Allow the agent to determine the actions needed to get there.
An invoice must be legitimate, approved, paid correctly, and recorded. Those requirements endure. The sequence of screens, spreadsheets, and handoffs historically used to satisfy them does not have to.
As the environment changes, the agent can adapt its approach while remaining bound by those requirements.
The sequence no longer needs to be defined in advance and managed as the work itself. Instead, each execution becomes a log of how the goal was achieved. The path can change from one run to the next as the agent and the business change.
The consequence is an enterprise that can change how it works without first waiting for someone to redesign a workflow. As tools, conditions, and capabilities change, agents can find new paths within the rules. Adaptation no longer has to wait for someone to rewrite the sequence.
That changes the foundations of enterprise software. The system must know what the business is trying to achieve, what is true now, what actions are permitted, and what counts as completion.
Process was initially an input to work. We now see it become an output.