During a closed-door introduction with ReilAI Corporation, I began to believe agentic AI might actually be ready for operations. I am delighted I got a peek at a new approach to making it easier to understand what’s happening in complex, ever-changing manufacturing and supply chain operations. This young company was founded on the belief that agentic AI holds the key to understanding and fully leveraging diverse data to inform the actions of a system in motion.
In this call, they showed me (representing Tech-Clarity, Inc.) and a select few others their Governed Agentic Execution Layer (GAEL), intended to bridge between data infrastructure and execution systems. Most exciting, it does so without displacing people or systems already operating in the plant, company, and ecosystem.
The Execution Gap
One of the curious points that ReilAI founder Joanne Friedman points out is that manufacturers measure the return on investment or return on invested capital for a $50,000 piece of equipment, but not for the $5-6M in data infrastructure that keeps the entire facility running – or not.
Data infrastructure is crucial to execution success. They point out three areas where an execution time gap exists today; these are how long it takes to:
- get data to someone who needs it
- understand it and make a decision
- move from decision to action that gains true value
In previous research (A New Era of Continuous Improvement), I’ve pointed out that time matters; it is one thing we cannot replace or regain. As most of us know, AI can save quite a bit of time. It can also capture and leverage knowledge, which is crucial as much of our experienced workforce nears retirement. So, a platform that leverages a knowledge and context graphs as well as agentic AI could help close those gaps.
GAEL’s Architecture
A manufacturer adopting GAEL can keep their current workflows or change them up front. Agentic AI choreography enables agents to run in parallel or in swarms. This goes beyond prescriptive orchestration to enable adaptation as conditions change, which is essential for ongoing execution. The design focuses on extensibility, observability, and explainability.
At the foundation of this new platform, and the first word in its name, is governed. Governance for swarms of AI agents is crucial. Key principles of this design is observability, traceability, and flexibility. The starting use cases are manufacturing and supply chain, which are inherently fast-moving, multi-disciplinary, and challenging.
The platform is also designed for collaboration. I mentioned a knowledge graph for context. This architecture is not unique to GAEL, but essential for many current systems to be execution-ready. Graphs can create meaning and context from otherwise fragmented data. GAEL also has context graphs and judgement layers. So context might include user intent, perspective, experience, expertise, authority, and security. With agents for nearly any role or discipline, the data appears with this full, deep context, ready for action.
The Path to Trust in Agents
The question is how to ensure the agents will generate great results. Friedman points out that no executive will just trust AI agents to work autonomously. Agents must earn trust. So, ReilAI has developed a training pathway for the agents.
Given that the founders are deep manufacturing and supply chain experts, they have built agents for specific roles with some starting knowledge:
- Initially, agents are like well-educated apprentices.
- As they learn from daily interactions, additional data, and tasks, they become journeymen.
- Only once the people using these agents agree can they become masters and run autonomously.
The platform has built-in “governed trust paths” that make sense for manufacturers. These start with physical safety, move to compliance, and also include margin protection logic. Until all of these are satisfied, even a “Master” level agent cannot autonomously execute its commands.
Return on Data
If you know Joanne Friedman, you have likely seen her work on Return on Data (RoD). She and I were recently on a podcast together discussing this topic. Her equation for RoD is:
[Sunk Data Capital] X [Contextual Intelligence] = [EBITDA Expansion]
The returns can be in top-line or bottom-line; in cost or revenue enhancements. Most manufacturers store tremendous amounts of data and pay for infrastructure and cloud services. The question is, how much of that data is delivering value, and how much?
One aspect of that we’ve also written about is expanding beyond production to link multiple companies in an ecosystem. The ReilAI vision is that each industrial facility (plant, warehouse, distribution center) can leverage GAEL to become an intelligent smart node within a coherent system. They say, “As more partners integrate, predictive power and execution value compound exponentially.”
Our Take
Agentic AI may be ready for industrial environments. To date, news about challenges in scaling and trusting AI in industrial settings abounds. We hear that investments may or may not pay off. The team at ReilAI has me thinking that might be news of the past, not the future or even the present. If GAEL can deliver everything in their vision, it will be a powerful platform to enhance operational execution. Governed agents ranked by expertise, leveraging existing people and data, seems like a good path forward.
Thank you, Joanne Friedman and Matthew Funderburg, for inviting us to this fascinating early look at your innovation at ReilAI Corporation. We look forward to following your progress in the market.
