TwinThread customers have been running fully closed-loop AI quality optimization in production for five years, with no human adjusting the set points. Andrew Waycott, President and Co-founder of TwinThread, described at Hannover Messe 2026 how the company’s industrial AI platform, running over 1 million digital twins and 2 million AI models in production, presented at Hannover Messe in partnership with AWS, progresses through four distinct AI layers: predictive, prescriptive, generative, and agentic. The agentic layer, where AI pushes optimizations directly to control systems, is where closed-loop autonomous manufacturing begins. The question for most manufacturers is how to get there safely.
How Does a Virtual Operation Center Work in Manufacturing?
A virtual operation center lets one subject matter expert monitor and optimize similar production lines across multiple facilities worldwide. The concept works like NASA mission control: if an expert understands a particular type of manufacturing, that person can be sitting at home or in a plant anywhere in the world and guide operations across every similar line the company runs.
The concept addresses a structural problem. Most manufacturers have many pieces of equipment that are very similar, whether multiple pumps and motors within a single plant or multiple facilities and production lines making the exact same product across the world. The expertise to run that equipment stays local, and learnings from one site rarely reach another.
TwinThread’s platform is designed to get that expert the information they need to monitor all the plants around the world that run similar processes. The technology does not replace the expert. It amplifies the expertise of the few experts a manufacturer has across every facility they operate.
What Happens When One Factory Teaches Every Other?
Fleet-wide AI optimization means insights gained from one production line or asset are pushed to all similar equipment across the organization. TwinThread calls this the “wisdom of the fleet”: what works at one facility becomes available everywhere.
Most manufacturers have multiple facilities or production lines making the exact same product. The insights gained from one are most likely applicable across the rest of the fleet of lines that are similar. The question is: what can we learn from one that can help us optimize the other?
TwinThread’s platform takes the virtual operation center concept and pushes insights and expertise across the organization. When a model identifies an optimization at one site, that learning reaches every similar asset in the fleet. For manufacturers with multiple facilities running similar processes, the fleet improves as each site contributes learnings that benefit every other.
How Do Manufacturers Set Guardrails for Autonomous AI?
Autonomous AI in manufacturing is safe when it operates within the same constraints applied to human operators, and TwinThread customers have proven this model in production for five years. The guardrails framework is this: every control system already has safety boundaries that prevent an operator from entering a dangerous set point. If a human operator is trusted to make a process adjustment within those boundaries, an AI model operating within the same boundaries carries the same level of risk.
Andrew Waycott described the logic at Hannover Messe 2026: “If an operator can make that change and you trust it, why can’t an ML model or an AI model also make that same change that you apply the same guardrails to, effectively?” The AI communicates to the control system in a similar manner to how a human communicates to the control system, and the same guardrails apply.
The practical difference is speed and consistency. An operator makes adjustments once an hour, waits for results, then decides the next move. Closed-loop AI makes those adjustments continuously, within the approved tolerances, knowing that the right adjustments are happening at the right time to deliver the right results.
The prerequisite is confidence in the AI’s recommendations. TwinThread’s approach builds that confidence incrementally: start with predictive insights the operator reviews, move to prescriptive recommendations the operator acts on, then progress to agentic control where the AI acts within the guardrails the manufacturer defined.
What Are the Four AI Layers in Industrial Manufacturing?
Industrial AI in manufacturing operates across four layers: predictive, prescriptive, generative, and agentic. TwinThread’s platform spans all four, and each layer adds a specific capability on the shop floor.
| Layer | Function | Decision Authority |
| Predictive | Forecasts what is coming down the line in the next hour | Human reviews the forecast |
| Prescriptive | Tells operators what they should do right now to optimize the process | Human acts on the recommendation |
| Generative | Exposes insights in natural language, helps solve problems faster | Human explores and asks questions |
| Agentic | Pushes fact-based optimizations directly to control systems | AI acts within defined guardrails |
At Hannover Messe 2026, every booth had the words “industrial AI” on display, but the question is what that actually means. For TwinThread, it means leveraging all four layers together. The predictive layer forecasts. The prescriptive layer recommends action. The generative layer helps operators and engineers explore insights in natural language and solve problems faster. The agentic layer closes the loop entirely, pushing fact-based optimizations to the control system without waiting for a human to intervene.
The key to closing the loop is delivering fact-based insight with enough confidence to push it to the people that matter, and even to the control systems that act on it. TwinThread has customers running fully closed-loop autonomous quality optimization for five years already. This is not a futuristic concept. It is production reality.
Frequently Asked Questions
1. How is AI being used in manufacturing operations?
Industrial AI in manufacturing covers four distinct layers: predictive AI forecasts equipment or process issues before they occur, prescriptive AI recommends specific adjustments operators should make, generative AI surfaces insights in natural language so teams can investigate root causes, and agentic AI pushes optimizations directly to control systems in a closed-loop model. TwinThread’s platform runs over 1 million digital twins and 2 million AI models in production across these four layers, and some customers have operated fully autonomous quality optimization for five years.
2. What is a virtual operation center in manufacturing?
A virtual operation center is a centralized monitoring model where one subject matter expert oversees similar production lines across multiple facilities worldwide. TwinThread compares the concept to NASA mission control: the expert does not need to be physically present at each plant but receives fleet-level visibility into every similar asset through the platform. This amplifies the expertise of a small number of specialists across an entire global operation.
3. How do manufacturers set guardrails for autonomous AI?
Manufacturers set guardrails for autonomous AI by applying the same safety constraints that already exist for human operators. Every control system has limits that prevent dangerous set point changes. If a human operator is trusted to make adjustments within those boundaries, an AI model operating within the same boundaries carries equivalent risk. TwinThread’s approach builds confidence incrementally, moving from predictive insights to prescriptive recommendations to fully autonomous closed-loop control.
4. What does fleet-wide AI optimization mean for factories?
Fleet-wide AI optimization means that insights from one production line or asset are pushed to all similar equipment across the organization. When one site identifies an optimization, that learning reaches every similar line globally without each plant running independent analysis. TwinThread calls this the “wisdom of the fleet.” For manufacturers with multiple facilities running similar processes, the fleet improves as each site contributes learnings that benefit every other.
This article is based on a video interview with Andrew Waycott, President and Co-founder of TwinThread, and Lucian Fogoros of IIoT World, recorded at Hannover Messe 2026. AI tools were used to help summarize and organize the content. Reviewed and edited by the IIoT World editorial team.
Watch the full interview on YouTube.
Related from IIoT World
- Eliminating the “Hidden Factory”: The Economics of Agentic Quality Control — How agentic AI and explainable AI are reshaping quality control economics in manufacturing.
- Why Most Manufacturers Never Feel “Ready” for a Digital Twin — A practical guide to starting with digital twins using existing IIoT infrastructure.
- AI Manufacturing Day 2026 — Free virtual conference on autonomous AI, industrial architecture, and edge AI for predictive maintenance. May 12, 2026.
About the author
