Research
Discovers, evaluates, and prepares evidence.
The DCPF Origin Story
DCPF did not begin as an attempt to create a new AI architecture. It grew from a practical need to make sustained AI collaboration easier to organize, maintain, and reuse.
1. An Accessible Creative Starting Point
In October 2024, while adapting to a life-changing visual impairment and building an accessible path into voiceover and audiobook production, I began using AI to help create scripts for adult relaxation and meditation experiences.
The goal was practical: spend less time physically writing and more time recording and editing audio. AI produced useful starting points that I reshaped, expanded, and refined into finished work.
That process supported my first published audiobook in 2025. It also showed me that AI could become more useful when the work around it was organized carefully.
Accessibility was part of the problem from the beginning.
As a visually impaired AI user, long conversations could become difficult to navigate. Research, instructions, decisions, and useful discoveries could become buried inside expanding chat histories. I needed a way to preserve important material outside the conversation so I would not have to visually search through earlier interactions and reconstruct the project each time.
2. The Project Outgrew the Prompt
The next audiobook project became more ambitious: professionally narrated journeys through America’s national parks. Each park required dependable information about geography, geology, plants, wildlife, seasons, weather, and cultural history, along with consistent narration, pacing, transitions, and production standards.
At first, I handled the growing complexity by adding more information to prompts. The stories improved, but the process became harder to manage. Research was scattered across conversations. Useful improvements had to be found and repeated. Prompts became longer because the project itself was becoming larger.
The challenge was no longer writing better prompts. It was managing knowledge.
The prompt was being asked to hold research, standards, examples, workflow instructions, and project history at the same time. I had mistaken conversation for organization.
3. Moving Knowledge Outside the Conversation
The first major change was simple. I moved recurring information into separate maintained resources. Research, storytelling guidance, terminology, production requirements, quality criteria, and project instructions began to receive stable homes outside individual AI conversations.
That separation made information easier to review, update, reuse, and supply when needed. A corrected definition or improved production rule could strengthen future work from one maintained location rather than being rediscovered or copied from an old chat.
The AI was not becoming more intelligent. The system surrounding the AI was.
Once information had separate homes, the responsibilities behind those homes became easier to see. Some work discovered and evaluated information. Some preserved information the project had accepted. Some directed the immediate task. Other decisions established standards, boundaries, review expectations, and authority.
4. Four Responsibilities Become DCPF
Across hundreds of AI interactions, those recurring responsibilities developed into four coordinated frameworks:
Knowledge
Preserves human-reviewed Accepted Knowledge for reuse.
Runtime
Uses approved Knowledge and current instructions to perform the work needed now.
Governance
Guides the system through human-authorized standards, boundaries, validation expectations, and review.
Research discovers. Knowledge preserves. Runtime executes. Governance guides.
The four-part architecture did not appear fully formed. It emerged gradually as repeated project problems were separated and given clear responsibility. What began as a way to organize an accessible creative workflow became a portable architecture for sustained human-AI collaboration.
5. From Practical System to Experimental Framework
The broader approach became Distributed Cognitive Prompting, or DCP. My specific contribution is the Distributed Cognitive Prompting Framework for AI Collaboration, or DCPF, which integrates Governance, Research, Knowledge, and Runtime into one coordinated architecture.
DCPF does not replace prompt engineering. It gives prompts a more focused role inside a larger system.
Prompt engineering optimizes individual interactions. DCPF organizes collections of interactions into persistent architectural systems.
DCPF is now being released openly as an experimental framework so other people can learn it, test it in their own domains, challenge it, adapt their implementations, and help determine where the architecture is useful and where further refinement is needed.
It has demonstrated promise through its development work, practical applications, and training practicums, but it has not been broadly or independently validated across domains, organizations, AI systems, or production environments.