Over the past few years, artificial intelligence has advanced at an unprecedented pace. Foundation models have expanded the capabilities of software far beyond what seemed possible only a short time ago, enabling intelligent assistants, content generation, task automation, and entirely new ways for people and machines to interact.
These breakthroughs are also transforming software engineering itself; accelerating development, increasing productivity, and changing how we build applications. Yet I believe the most significant transformation is still ahead.
The real shift is not about embedding AI assistants or agents into existing applications to automate individual tasks. It is about rethinking how software itself is designed, built, and operated.
For more than three decades, we have developed applications intended to be used by people. We designed user interfaces, APIs, databases, and business processes around a simple assumption: a human user would make the decisions and execute every business workflow.
That assumption is beginning to disappear.
The next generation of software consumers will not be people alone. They will be intelligent agents capable of understanding objectives, reasoning, planning, and autonomously executing complete business processes. When that happens, the backend is no longer simply a collection of services. It becomes the operating environment for the artificial intelligence that collaborates with humans to operate the business.
That is why I believe we are entering a new era of enterprise software architecture: AI Systems.
An AI System is not an application with a chatbot attached. It is an intelligent environment where business capabilities, APIs, MCP servers, enterprise knowledge, AI agents, and governance mechanisms operate together as a single coordinated system.
In this architecture, artificial intelligence is no longer an isolated feature. It becomes the execution layer that continuously orchestrates how software behaves.
This also forces us to rethink where knowledge resides, and what kinds of knowledge truly create value. General knowledge will increasingly be surpassed by domain-specific, contextual knowledge that reflects the unique expertise of each organization.
During the first wave of AI, the prevailing strategy seemed simple: send as much information as possible to a large cloud-hosted model. Organizations quickly realized, however, that their most valuable knowledge cannot simply flow freely through the cloud. Regulatory requirements, privacy policies, intellectual property, data residency, and mission-critical business processes often prevent that information from leaving the enterprise.
As a result, artificial intelligence cannot reside exclusively in the cloud either.
As AI Systems become embedded in banks, governments, factories, hospitals, critical infrastructure, connected devices, and enterprise environments, part of their intelligence must execute where data is generated and where decisions are made: at the edge.
For that reason, I believe the future belongs not to centralized AI, but to distributed AI architectures. More precisely, to distributed knowledge combined with distributed inference.
Each domain should retain ownership of its own knowledge under its own governance policies, while intelligence collaborates seamlessly across the cloud, the edge, and every execution environment.
The goal is not to centralize knowledge so AI can use it. The goal is to distribute intelligence so it can act wherever knowledge already exists.
This distinction will fundamentally reshape enterprise platforms.
Just as APIs transformed system integration two decades ago, the next generation of enterprise platforms will combine distributed intelligence across cloud and edge, enterprise knowledge, and intelligent agents into a single architecture capable of continuously evolving.
This vision is precisely what led us to create AVAP and Brunix. Not as platforms for integrating AI models, but as the technological foundation required to build and operate AI Systems.
Their purpose is to transform the traditional backend into an intelligent execution layer, where business capabilities can be consumed seamlessly by applications, APIs, MCP servers, AI agents, or whatever new software consumers emerge in the future.
I believe that over the coming years, the traditional boundaries between software development, operations, and data will gradually disappear. Systems will evolve continuously through collaboration between people and AI agents, while governance, observability, and enterprise knowledge become intrinsic parts of the architecture itself.
The conversation will also shift.
Instead of asking which model to use, organizations will ask how to build systems capable of learning, adapting, and continuously evolving without losing control of their knowledge or their business processes.
Knowledge graphs and inference workflows will become central architectural elements. The platforms that make these systems easy to design, govern, and evolve will define the next generation of enterprise software.
That, to me, is the true meaning of AI Systems.
They are not simply another category of software. They represent the next stage in the evolution of software engineering itself.
And I am convinced that technologies like AVAP and Brunix will help lead that transition.
