Adopt FAIR Data principles for Agentic AI

Importance of adopting FAIR Data principles for Successful Agentic AI Implementation.

Series: AWS Data and AI (15 articles)

1
Scalable AWS ETL Data pipeline
…
15
Adopt FAIR Data principles for Agentic AI This article

Naveena Ravi

Published Oct 15, 2026

Last modified Oct 17, 2026

Based on recent research conducted my MIT, AI has access to only 45% of enterprise data across globe. Of this less than 30% data is shared by companies with no proper data foundations termed as "Data Laggards" The companies that has shared around 70% of their data are termed as "Data Leaders". Data Leaders have strong foundation and believe their AI agents can make trustable decisions. (Ref: Scaling AI agents with trustworthy data | MIT Technology Review)

In this article I am sharing the details of my technical session as speaker for AWS UG Women-in-tech meetup in Chennai.

Coincidently, I was able to deliver this session after my AWS New Voice training. Hence, I was able to apply my learnings. I got badge after successful completion of this program.

Following the New voice program below was my introduction slide from recent MIT research publication.

Following the introduction slide, I introduced myself. After introduction to the session, I introduced the AI transformation, so that audience can appreciate the difference between Traditional AI, Generative AI and Generative AI.

Traditional AI - Using AI to provide a schedule for your trip.

Generative AI - Using AI to generate a schedule based on our preference for the trip.

Agentic AI - Using AI to book tickets, accommodations based on our preference.

Components of Agentic AI can be analogous to human beings - LLM to human brain, AI Agents to our hands, MCP to human nervous system and RAG to books we read. However, unlike humans, AI Agents cannot stop and ask its queries. Hence AI agents must have access to right data at all times.

FAIR Data principles help us build a Data foundation that provides findable, accessible, interoperable and reusable data for AI Agents and humans.

Taking an analogy of building world maps,

Findable- An Ai application must be able to identify each map with a definite name (METADATA) of the country name like INDIA map, Sri Lanka map, etc.

Accessible - The AI application for creating world map must be able to access maps of each country.

Interoperable - AI application must be able to build the world map with collections of all the maps representing each country.

Reusable - The world map created by AI application must be reusable for plotting multiple dimensions like political, geological, population, etc. We must not be required to create a new map for each application of world map.

Below illustration shows how a raw data is transformed in typical Enterprise applications.

Let us relate this to an analogy from Banking Applications:

Raw data is received from multiple ATMs, branch office, merchant outlets, etc.
Taxonomy/Ontology is created to identify specify banking operations and terminologies around it like wealth management, personal loans, corporate loans, exchange operations, deposits, etc
Build modular Database schema for each application like loans, deposits, branch maintenance, etc.
Data transformation is done using medallion architecture to convert raw data to business-ready (GOLD) data.
This is data is appropriately catalogued to be available for AI Agents and Humans.

Amazon Sagemaker helps us to manage Open Lakehouse and access data using AI Governance. This can be accessed by AI agents of the Enterprise Agentic Applications.

Amazon unified studio provides us with Open Lakehouse which can connect with (Metadata) Data catalogs in open table format. Ensure the data is accessible with AI Governance and available to all AI applications.

Get started with Amazon S3 Tables in Amazon SageMaker Unified Studio - Amazon SageMaker