Underpinning the solutions
Appendix
This section provides overarching information that will help you better understand the foundational work needed to complete the work in any of your selected solution areas.
Context
Personal
Guided personal context statement creation. This is performed for all key stakeholders in the organization.
Organizational
Organizational context gets it working on your actual problems. But organizational context isn't always optimized for agentic consumption. We have to identify it, collect it, and structure it.
Most organizations have enormous amounts of valuable context that exists in these three states:
- Documented
- Written down, accessible, and current — but needs to be restructured so AI can actually use it.
- e.g. Published certification framework (PDF)
- Tribal knowledge
- People know it, but it's not written down anywhere. Valuable context that lives only in someone's head.
- e.g. "How we actually connect farmers to brands", "How the ROC framework evolves over time", "How we answer questions from the field"
- Gap
- Nobody has this knowledge, or what exists is too outdated to be useful.
- e.g. Missing policies, no data classification guide
Security
The SECURE Framework
A checklist for any agent deployment:
- S Scope permissions narrowly. Minimum access for each tool. Read-only by default.
- E Explicit safety rules. Written constraints in a dedicated file. Reload every session.
- C Confirm destructive actions. Human approval for delete, send, publish. No bulk operations without review.
- U Unattended limits. Cap operations per autonomous session. Heartbeat intervals. Cost alerts.
- R Record everything. Append-only action logs. Session transcripts. Tool invocation traces.
- E Emergency stops. Out-of-band kill switch. Network isolation. Automatic halt on rule violations.
Why Agents Are Different from Chatbots
When you deploy an agent, you're exposing: missions, data access, action capabilities, your credentials, and unattended time. That's a fundamentally different risk profile from a chatbot.
Connections
Four Approaches
The AI has access to a menu of available tools and decides on its own which ones to use for each task — like giving a contractor a fully stocked toolbox. Flexible and powerful, but more expensive to run since the AI loads every option into its working memory each time.
The AI runs one specific vendor-provided program and reads the result. Less flexible than Smart Connect, but more precise — fewer choices means less room for the AI to reach for the wrong tool.
No AI judgment involved — a script calls a service the exact same way every time. Think of it like an automated form letter: reliable, inexpensive, and completely predictable.
We engineer a purpose-built connection to your internal systems from scratch. The most upfront investment, but also the most control — the right choice when nothing off-the-shelf fits.
GEO / AEO
Optimizing for AI Discovery
The New Search Landscape
Search is no longer just about ranking on a results page. A growing share of people get their answers directly from AI tools — ChatGPT, Perplexity, Claude, Google's AI Overviews — without ever clicking a link. If your content isn't structured for AI retrieval, it doesn't matter how well it ranks on Google.
GEO (Generative Engine Optimization) and AEO (Answer Engine Optimization) are two names for the same shift: designing content so that AI systems can find it, understand it, and accurately cite it in responses.
How It Differs from SEO
- Target
- SEO targets Google's crawlers and ranks pages by backlinks and keywords. GEO/AEO targets AI retrieval systems that summarize and cite content directly in responses.
- Success metric
- SEO measures page rank and clicks. GEO/AEO measures whether your content is cited — or better, whether it becomes the answer — when someone asks an AI a relevant question.
- Content format
- SEO rewards keyword density and link equity. GEO/AEO rewards clear structure, explicit entity relationships, and consistent terminology that AI can parse and attribute accurately.
- Do both still matter?
- Yes. Well-structured content performs well in both. But GEO/AEO demands more explicit context — who you are, what you certify, why it matters — that traditional SEO never required.
What AI Systems Look For
- Structured, scannable content
- Headers, definitions, and lists AI can parse cleanly. Dense paragraphs of prose are harder to retrieve accurately than well-labeled sections.
- e.g. A clearly labeled FAQ page outperforms a white paper with the same information buried in running text.
- Explicit entity relationships
- Clear statements about who ROA is, what the ROC certification means, how farmers and brands relate to the program. AI systems build answers from named relationships — vague prose produces vague citations.
- e.g. "ROA's Regenerative Organic Certified® (ROC) program certifies farms and products across three pillars: soil health, animal welfare, and farmer fairness."
- Authoritative, consistent terminology
- Using the same terms consistently across all ROA content trains AI models to associate those terms with ROA. Inconsistency splits authority across multiple definitions.
- e.g. Always "Regenerative Organic Certified" not sometimes "regenerative organic certification" or "ROC label".
- Freshness signals
- Recently updated pages rank higher in AI retrieval. Stale content gets deprioritized even if the information is still accurate.
Glossary
- Agent
- An AI system that can take actions in the world — reading files, calling APIs, sending messages — not just generate text. Unlike a chatbot, an agent has tools and can operate across multiple steps without a human in the loop for each one.
- Context Window
- The maximum amount of text an AI model can hold in its working memory at one time. Think of it as a whiteboard: everything the AI can currently see and reason about fits on that whiteboard. Older messages and documents scroll off the edge when it fills up.
- Embeddings
- A way of converting text into numbers that capture meaning, so that similar concepts end up mathematically close to each other. The technology behind semantic search — finding content that means the same thing even when the words differ.
- Fine-tuning
- Retraining an existing AI model on new examples to shift its behavior. Like an apprenticeship: you show the model many examples of what good looks like for your specific task, and it adjusts. More powerful than prompting alone, but requires labeled data and compute.
- Hallucination
- When an AI generates text that sounds confident and plausible but is factually wrong. A structural risk of large language models — they predict likely-sounding continuations, not verified facts. Good system design (like RAG) reduces but does not eliminate hallucination.
- Inference
- The act of running an AI model to get a response. Every time you send a message and get a reply, that's inference. It consumes compute and costs money; high-volume agentic systems need to account for inference costs at scale.
- LLM (Large Language Model)
- The underlying AI technology behind tools like Claude, GPT-4, and Gemini. Trained on vast amounts of text to predict and generate language. The engine inside most AI products today.
- MCP (Model Context Protocol)
- An open standard that lets AI models connect to external tools and data sources in a consistent way. Instead of building a custom integration for every service, MCP provides a shared protocol — similar to how USB standardized device connections.
- Prompt
- The instruction or input you give an AI model. A well-crafted prompt can dramatically change the quality of the output. Prompt engineering is the practice of designing these inputs systematically rather than by trial and error.
- RAG (Retrieval-Augmented Generation)
- A technique that lets an AI answer questions using your documents, not just its training data. Before generating a response, the system retrieves relevant passages from a knowledge base and feeds them into the model's context window. The result is answers grounded in your actual content.
- System Prompt
- A set of instructions given to an AI model at the start of every session, before the user says anything. It defines the model's role, constraints, tone, and rules. The system prompt is where you encode organizational policies and guardrails for an AI deployment.
- Token
- The unit AI models use to process text — roughly three-quarters of a word on average. Models are priced and limited by token counts, not word counts. A 100,000-token context window holds approximately 75,000 words.
- Tool Use (Function Calling)
- The ability of an AI model to call external functions — search, database queries, API requests — and incorporate the results into its response. What separates a basic chatbot from an agent capable of taking real-world actions.