How AI Unbundles the Enterprise System
Enterprise IT Systems
Enterprise IT systems are complex with many layers of protocols, technology, and interconnections. But in a simple view you can think about them having three logical components. Systems have a user interface, a database, and processing logic consisting of business rules, validations, processes, workflows, roles, and permissions.
AI allows organizations to decompose the system into smaller atomic parts. Each of these atoms can be put together in new ways—new "molecules" if we continue the analogy.
User Interface - You don't have to have a pre-built user interface. AI chats will evolve—it can be voice, it can be chat. It could create a customized one-person interface for their job. Each person will get their own user interface related to their job and to do their work. They won't necessarily need fields and screens. They could just paste in an email and it would correctly update the back-end system. The user interface becomes really powerful and customized, but people don't need to tab through fields or figure out how to navigate things. They just interact with the user facing agent um how they feel comfortable
Business Rules and Workflow - The evolution of business rules is really interesting. Agents read files of rules and context (typically in a text format called markdown) to make things happen. Enterprises will develop more rigorous ways of capturing business rules, common practices, exceptions, and other scenarios in markdown files. AI agents need context—facts, declarative statements, information—to make decisions and take action. For example, Anthropic developed a standard called SKILLS.md which is a starting point, but it is rudimentary and we need to continue to make that more rigorous for enterprises. AI agents need context (Facts, declarative statements, information) to make decisions and take action. Context is an enterprise asset that needs to be built, managed, versioned, and agreed upon in an organization.
Data - Even with modern architectures, most databases are associated with a "system". AI continues and accelerates the trend towards enterprise data stores that aren't tied to one back end system. Enterprises have a set of data stores that different AI agents update according to approved business rules including security and permissions. Enterprise infrastructure protects sensitive or proprietary data but also pulls in data or LLM models as needed.
What's Coming Next
Business Rule Testability - Organizations following DevSecOps already understand the benefit of automated testing. With AI, the testing drives code development automation, the AI continues coding until all tests are met (geeks can look up Ralph Wiggum!)
Confidence Level Operations - Agent decision-making in an enterprise context is based on confidence levels. How sure is the AI on it making the right decision? Outcomes at 98% or higher move forward, 90-97% might get spot checked, 85-89% are put in a manual review queue. Human intervention is required to approve the decision, but critically this step is to improve the business rules if a wrong decision was reached. The process improves and iterates the rules and context to continually align the Agent decision making to the organization strategy, values, and outcomes. By the way, this Operational Audit role is a new job function!
Execution Traceability - Organizational leaders need to be able to back track Agent made decisions to understand what actions were taken and why. Logging of AI becomes critical to business, legal, and regulatory functions
Levels of Agent Authority - Just like employees have roles and authorities, different AI agents will be configured with assignments, responsibilities, and authorities. Agents will pass assignments to specialty agents trained in certain roles, just like human or IT system workflows.
How It All Fits Together
So how does this all fit together? An end user (manager, executive, producer) has interactive exchanges with a personal orchestration agent, moving seamlessly across a laptop, tablet, or phone screen (or Meta glasses or Apple VisionPro!) depending on the task. The personal agent prioritizes and helps you focus on goals or time sensitive requests. It interfaces with enterprise agents to conduct the mission of the enterprise. Agents talking to agents, yes, but with humans in the loop adding judgement, nuance, and context.
I used AI to capture my thoughts while I was out walking. I did not use AI to write any of the content.
Originally published on LinkedIn, January 21, 2026