RGI Framework™
The First Principles of Human-AI Collaboration
The universal methodology defining how humans work with generative AI across every industry and use case
Generative AI: The Transformer Breakthrough
Modern AI is powered by Foundation Models using transformer technology. This breakthrough enables something unprecedented: humans can now communicate directly with advanced computing systems using natural language, not code.
Before transformers, only programmers could tell computers what to do. Now, you can describe what you need in plain language, and AI can write the code, create the content, and produce the results.
*The RGI Framework is scoped to productive human-AI collaboration in professional and enterprise contexts. We acknowledge that a small percentage of human-AI interaction involves personal expression, casual conversation, or social-emotional exchange that falls outside this productive scope.
The Three Pillars of Human-AI Collaboration
Retrieve
Ask • Find • Search
Query AI as your knowledge engine. Instantly access, summarize, and extract information from vast datasets, documents, and knowledge sources with natural language commands.
Generate
Write • Ideate • Create
Direct AI to produce content, solutions, and ideas at scale. From reports and communications to strategies and creative concepts, generate high-quality output on demand.
Interact
Instruct • Query • Collaborate • Act
Engage AI as your workflow partner and autonomous agent. From iterative refinement to complex orchestration, AI can collaborate and act independently within your systems.
Universal Scalability
Individual Professional
RGI empowers every professional to enhance their work through AI collaboration. Whether you’re an analyst, manager, consultant, or specialist, the framework provides clear guidance for AI integration into daily workflows.
“What can I retrieve, generate, and interact with to enhance my work?”
Enterprise Leadership
For organizations, RGI provides the strategic framework for building AI capabilities across systems and workflows. From productivity gains to autonomous process orchestration, RGI scales from individual tools to enterprise transformation.
“How do we architect AI capabilities across our systems and workflows?”
Precision Communication
Effective human-AI collaboration requires structured communication. Master the art of prompting to unlock professional-quality results from any AI system.
Goal + Context + Format + Reference + Refine = Professional Results
Clear objectives, relevant context, defined formats, supporting references, and iterative refinement transform basic AI interactions into powerful professional tools.
An Open Academic Challenge
We are in the early innings of the modern AI era. We are all pioneers exploring this transformative technology together. There are no deeply seasoned experts or widely accepted frameworks of knowledge that belong in a vault in the archives of human knowledge.
In this spirit of innovation and collaboration, we invite rigorous academic scrutiny of the RGI Framework. Can it be distilled into more basic elements? Are there modes of human-AI collaboration that fall outside these three pillars?
How 700 Million People Actually Use AI
The largest study of consumer AI usage ever conducted
OpenAI Economic Research Team & Harvard University
NBER Working Paper No. 34255 • Published September 2025
Unprecedented Scale & Growth
Fastest technology adoption in history
Work vs. Personal Usage
Personal Usage
Personal usage growing faster than work-related. Challenges assumption that AI is primarily a workplace productivity tool.
Work-Related Usage
Work usage more common among educated users in professional occupations. Writing dominates at 40% of all work messages.
Top 3 Use Cases
Practical Guidance
Customized advice, tutoring, product guidance, financial decisions, how-to instructions
Writing
Document editing, email drafts, communication assistance. 67% edit existing text vs. create new
Seeking Information
Research, fact-finding, current events, product comparisons, decision support
User Intent: Asking vs. Doing vs. Expressing
Asking
Seeking information, advice, decision support
Growing fastest • Rated higher quality
Doing
Task completion, output generation, executing work
Dominates work usage at 56%
Expressing
Personal reflection, casual conversation
Social-emotional interaction
Key Findings for Professionals
Decision Support Over Automation
81% of work messages involve information gathering/interpretation and decision-making/problem-solving. AI functions as decision support tool, not task replacement.
Writing Dominates Professional Use
40% of work-related messages involve writing. Two-thirds request editing existing text rather than creating new content.
Quality Favors Information-Seeking
“Asking” messages consistently rated higher quality than “Doing” messages. Good interactions 4x more common than bad by July 2025.
Massive Economic Value Created
Estimated $97+ billion annually in consumer surplus in US alone. Substantial productivity gains in knowledge-intensive roles.
Education Drives Adoption
Higher education correlates with work usage: 37% (< bachelor's) vs. 48% (graduate degree). Professional occupations show highest adoption.
Programming Is Minor Use Case
Only 4.2% of messages relate to computer programming. Far lower than expected, showing broad non-technical adoption.
Who’s Using AI
Gender Parity Reached
Female users by July 2025, up from ~20% in early 2023. Gender gap has closed.
Young Users Drive Adoption
Of adult messages from users under 26. Younger clients are AI-native.
Global Growth Accelerating
Faster growth in low-to-middle income countries vs. high-income. Becoming globally accessible.
Read the Full Research Paper
Access the complete NBER working paper with detailed methodology, findings, and implications. 64 pages of comprehensive analysis from OpenAI’s Economic Research Team and Harvard researchers.
Download Full Paper (PDF)Bottom Line
This research reveals AI adoption at unprecedented scale, with usage patterns that emphasize advisory and decision-support functions over task automation. Remarkable consistency across occupations shows the same work activities dominate regardless of job type—AI is functioning primarily as a decision support tool rather than replacement technology.
RGI Framework™ Finds Convergent Evidence
Largest AI Usage Study Independently Validates Core Architecture
OpenAI and Harvard researchers analyzed 1.5 million conversations from 700 million users and independently developed a taxonomy with substantial overlap to RGI’s three-operation structure—providing empirical support for patterns discovered through operational practice.
The OpenAI/Harvard Study (September 2025)
Convergent Discovery: Two Independent Approaches
Information-Seeking
Study: “Asking” behaviors
RGI: Retrieve operations
Output Generation
Study: “Doing” behaviors
RGI: Generate operations
Outside Scope
Study: “Expressing” (social)
RGI: Non-productive usage
Convergent Validity: Despite independent development and different methodologies—one empirical (1.5M conversations), one practice-based (operational discovery)—both approaches converged on tripartite architectures with information-seeking dominating output generation.
How Study Findings Align with RGI Operations
The study’s three largest use cases map to RGI’s compositional architecture, validating the framework’s emphasis on Retrieve-first workflows.
| Use Case (% of Usage) | RGI Pattern Alignment | Key Supporting Evidence |
|---|---|---|
|
Practical Guidance
29% of all usage
|
Retrieve Interact | Customized advice requires pulling knowledge then iterating through conversation. Study notes these are “highly customized and can be adapted based on follow-up”—matching Retrieve→Interact sequential pattern. |
|
Writing
24% of all usage
|
Retrieve Generate | 67% of Writing requests modify existing text. Study validates RGI’s compositional model: Retrieve context first, Generate output second. Pure generation-from-scratch is rare. |
|
Seeking Information
24% of all usage
|
Retrieve | Pure Retrieve behavior. Classic search and research use case. Foundation for all other productive workflows. 24% standalone usage validates Retrieve as independent operation. |
The study’s “Expressing” category (11%) captures social/emotional interaction outside RGI’s productive collaboration scope. When filtered to work-related messages (30% of total), Expressing drops to ~9%, with RGI operations covering the productive 91%.
The Decision Support Thesis
Both RGI Framework and the OpenAI/Harvard study converge on a critical finding: AI’s primary economic value comes from decision support (Retrieve + Interact) rather than task automation (Generate alone).
The study found that 81% of work messages involve information gathering/interpretation and decision-making/problem-solving. Asking behaviors (49%) dominate Doing behaviors (40%), grow faster, and receive higher quality ratings.
This empirical evidence validates RGI’s architectural principle: Retrieve serves as the foundation, Generate follows, and Interact orchestrates productive workflows.
Powered by AVA