The importance of documentation cannot be overstated. It fulfills a number of crucial functions to get consistently reliable results. Documents act as an external influencer of state. They make important information explicit, editable, auditable and reusable across otherwise separate interactions.

State documents can be broadly categorized as canonical, evolving, historic. Canonical docs are governing and don’t often change. They could include business plans, style guides, policies and frameworks. Evolving docs are works in progress that move toward a completion goal. These include specs, plans, designs and requirements. Historic documents are a catchall for all that happened and why, and are updated every time anything relevant changes. This breakdown can be debated, but the labels describe concepts useful to state.

The importance of documentation cannot be overstated.

Documentation has gone from the most tedious part of systems engineering to a cornerstone for AI. Since its earliest public releases, AI has been able to turn a reasonably well articulated concept into an acceptable quality document. It now takes little effort to produce high quality business and marketing plans, coding philosophies, specs, sales playbooks, as well as legal documents. AI has not only made it economically practical to fully document, it has virtually made it inexcusable not to.

What’s important isn’t that AI can crank out a high quality document. It is that feeding more relevant information into the state, in particular through a larger project context, improves and aligns output across your project. For example, creating and loading a style guide as a canonical document will produce output that aligns with the colors, fonts, and messaging. Coding windows will create button colors, placement and names that align with these guidelines. The business plan, style guide, incorporation docs could be considered canonical.

Evolving docs are where you take an unrefined concept and break it down into specifics. A good technique for a project that can be broken into many deeper levels, each with its own complexity, is to stack windows. Start with a high level project discussion in one window, move to a second window for spec development with the output from the first window, and then build against the spec in a third. As the document improves and gains detail, each new window state receives more concise direction than the previous.

While evolving documents are the source of working truth, historic documents record what happened and why. This is often the missing documentation step from traditional systems engineering, documenting how you arrived at the current reality. It is important to ensure AI can update these documents as you go, allowing your state access to the evolution that led to your current state.