Advanced Air Mobility (AAM) has accelerated from concept to demonstration as the Federal Aviation Administration’s (FAA’s) electric vertical takeoff and landing (eVTOL) Integration Pilot Program (eIPP) rolls out. The need to prepare for AAM flight demonstrations has increased pressure on state, local, tribal and territorial (SLTT) governments to be “AAM-ready.” They must make strategic decisions on airspace integration, vertiport siting, equity impacts and garnering public acceptance, while policy and technological landscapes continue to evolve. To do this, they must balance their own finite public resources against an ever-changing technological and federal policy backdrop. Doing it well requires synthesizing hundreds of pages of guidance, frameworks and lessons from early adopters. Leveraging decades of aviation and emerging technology expertise, CNA has designed a specific LLM-based AAM Planning Agent to assist state and municipal aviation agencies with AAM planning and recommendations.
The Limits of RAG: Why Existing Approaches Fail
Large language models (LLMs), such as Google’s Gemini and OpenAI’s ChatGPT, are enabling technologies with capabilities that have coincidentally emerged and become accessible alongside AAM development. In the right context and construct, LLMs may offer planners a tool that can understand the AAM domain, reason across complex guidance and case studies and help translate strategy and policy into clear, defensible communication tailored to a jurisdiction.
But generic chatbots lack domain knowledge. Tools like Gemini and ChatGPT can draft text quickly, but they are trained on broad data and not tuned to AAM planning. Retrieval augmented generation (RAG) is typically used to provide LLMs with curated documents to improve response quality and relevance in esoteric domains. Most RAG systems behave like a smarter “CTRL+F.” They pull a handful of relevant passages from a document library and prompt a model to draft an answer around them. This works well for narrow questions, but it breaks down when the task looks more like policy analysis than simple fact lookup.
What CNA Created: A Purpose-Built AAM Knowledge Base
CNA reviewed emerging AAM strategies, identified several consistent themes and engineered its AAM Planning Agent around these realities. First, leading jurisdictions treat AAM as a complement to existing transportation systems rather than a standalone novelty. Second, leading jurisdictions pursue phased, conservative implementation that allows for leveraging existing infrastructure before committing to major new investments. Third, they view AAM economic development as part of a broader pipeline that links incentives, regulatory posture, workforce and education. CNA’s AAM Planning Agent embraces these realities. It provides state and municipal decision-makers with a practical, evidence-based tool for navigating a complex, fast-moving landscape with confidence and clarity.
CNA leveraged its expertise and FAA experience to create the foundation, a domain-specific corpus. Rather than scraping the internet, the AAM Planning Agent is grounded in a specially collected body of resources that state and municipal decision-makers rely on, including:
- Federal guidance and concepts of operations.
- State, regional and municipal AAM frameworks and roadmaps.
- Study data, scenario analyses and lessons learned from early adopters.
- Researched best practices in governance, equity, safety, integration and community engagement.
The curation of resources matters for two main reasons: noise reduction and comparative reasoning. With regard to “noise,” the agent does not guess based on generic aviation content or random internet results. It works from the same types of credible and relevant materials that human analysts use. For purposes of comparative reasoning, because the corpus includes varied approaches, jurisdictional diversity and CNA’s own knowledge lake, the agent can help identify patterns. These include how different jurisdictions structure governance, what kinds of community engagement strategies have emerged and where policy direction converges or diverges.
Why CNA Built It This Way: Balancing Speed and Judgment
CNA provides independent, not-for-profit analysis for federal, state, and local organizations that make tough policy and strategy decisions. CNA uses a strategic approach to AAM planning to safeguard public resources and needs amid the uncertainties of emerging technologies, concepts and regulations. Three design choices followed from CNA’s approach:
Domain Specificity Over Generality
CNA tuned the system for AAM planning, not to answer any question about anything. That constraint allows for deeper, more reliable reasoning within the domain.
Support, Not Substitution
CNA explicitly designed the agent as an assistant. It helps accelerate analysis and communication, but it does not claim to guarantee compliance, replace public engagement or substitute for human judgment.
Transparency of Sources
Because CNA grounded the agent in a curated corpus, it can point back to the documents and sections that informed its outputs. This supports traceability and review.
How It Works: Hierarchical Agents, Not a Single Chatbot
Instead of a single retrieval step, CNA’s AAM Planning Agent uses an agentic retrieval architecture. This means that multiple specialized agents repeatedly query the foundational AAM library, reason over what they find and critique each other’s work before presenting a result. If traditional RAG is CTRL+F, then this is the equivalent of a small and coordinated research team.
As an example of how this may work, a municipal planner may need to draft a stakeholder engagement plan for a new operator coming to their jurisdiction. They might need an outline for who to engage and how, as well as some materials to present to concerned citizens. When prompted:
- A research agent focuses on retrieval. It issues multiple targeted queries to the document library, iteratively refining what it pulls based on the task at hand. It might pull results such as multiple state and municipal stakeholder engagement plans, CNA-derived best practices, and relevant guidance.
- An analysis agent then synthesizes those materials, compares strategies, highlights tradeoffs and checks alignment with known best practices.
- A review agent critiques the draft, checking for internal consistency, clarity and fidelity to the source materials. It triggers additional retrieval if needed.
- An orchestrator agent oversees the team to ensure progress towards the response.
This loop can run several times for a single task. This provides a response that has been through multiple cycles of “think, check, revise” across different roles. A well-reasoned and evidence-supported response results, providing the planner with a thoughtful stakeholder engagement plan and supporting materials. This process is admittedly slower than typical LLM-based query-response interactions. As such, for simpler tasks, CNA also has a traditional RAG agent that uses the curated file library available.
The ROI: What It Actually Does and Doesn’t Do
AAM planning is as much about communication as it is about technical design. Using the AAM Planning Agent, decision-makers and CNA analysts can generate:
- Executive decision briefs comparing multiple AAM implementation strategies, with tradeoffs and supporting evidence.
- Tailored briefing packages that summarize FAA guidance, early adopter experiences and recommended state actions.
- Legislative testimony drafts with cited references to federal guidance and state frameworks.
- Community engagement plans tailored to local concerns, grounded in documented best practices from other jurisdictions.
- Public FAQ documents addressing safety, noise, equity and environmental impacts, with traceable references.
The agent does not replace policy judgment, but it can accelerate the early drafting and comparison of options, surface patterns and reference points and give decision-makers a clearer view of the landscape. This enables analyst teams to focus on decisions rather than first drafts.
In addition to the AI, the CNA human analysts can also support AAM decision-makers with decades of aviation expertise with the FAA, including support to Unmanned Aircraft System (UAS) Traffic Management (UTM) ecosystem design, demonstration and implementation and Urban Air Mobility (UAM) and AAM concept and systems engineering design.
In practice, the most valuable use of the AAM Planning Agent is in co‑production. Planners and CNA analysts refine prompts, review draft outputs against local priorities, and iteratively adjust recommendations.
For information about CNA’s aviation expertise visit https://www.cna.org/centers-and-divisions/ipr/esm/aviation or email aviation@cna.org