My original outline for this series included information on loops, graphs, agents, harnesses and other ways AI is being used. The more I wrote, the more apparent it became that state is the environment in which all of these things operate.

Chat has a state; agents have states; loops use and update state; graphs may have several states. Each one is influenced by state. In all of these cases, not only is state present, but control of state is crucial.

Take an AI game for a child. The state needs to understand the game, but it also needs awareness of age appropriateness and safety, as well as an understanding of what can and cannot be displayed. These are all elements of the game’s state.

None of these methods or tools need a definition of state to work, but the concept provides a way to interact with them more accurately and predictably. When purposefully applied across a stack or a company’s AI initiatives, it can provide a framework for designing safe and consistent implementations.

In addition to its practical uses as a business and production concept, states also help explain some of the contextual inaccuracies and other anomalies that users experience. As state develops and evolves through interactions, the AI can appear enthusiastic, skeptical, highly disciplined or even playful.

During these interactions, the AI can seem to express human emotions and qualities. These expressions, words and ideas are selected based on state. If a state has been shaped by positive, upbeat and optimistic interactions, the AI will often naturally use positive human expressions within the state.

There were two motivations for writing this series. The first was to present a concept that I found to be helpful and productive while working with AI. The second was to move the conversation about “why” AI does certain things forward with a framework that appears to provide insight into some behaviors.

A better understanding of why particular results appear also creates a foundation for thinking about broader and more complex AI systems. As companies add AI to development, marketing, customer service, internal knowledge, automation and other functions, each implementation brings its own state. Some information should be common across them. Some should be isolated. Some should be constantly updated, while some information should remain fixed and authoritative.

At scale, state becomes more than a way to improve an individual response. It becomes a question of how an organization keeps its AI implementations operating from a consistent, accurate and intentional understanding of reality.

At scale, state becomes more than a way to improve an individual response. It becomes a question of how an organization keeps its AI implementations operating from a consistent, accurate and intentional understanding of reality.