Autonomous manufacturing is production in which machines, software and AI sense what is happening, decide what to do and act on it with minimal human input, while people set the rules and handle the exceptions. Autonomy comes in levels, and most plants today automate individual tasks; a factory that runs with no human interaction at all is still what Yokogawa calls “a highly idealized state.”
Even the most advanced plants are only part of the way there. About 10% of the solutions running at the World Economic Forum’s Lighthouse sites are closed-loop and autonomous (WEF, January 2026). The next loop to close is less on the machines than in the operations work between systems: supplier follow-up, replanning, expediting and shift handovers.
Get your free Manufacturing AI Profitability Plan
In this guide
What is autonomous manufacturing?
What are the levels of autonomy in manufacturing?
Is an autonomous factory the same as a lights-out factory?
How far have the most advanced plants got?
How far has the average plant got?
Why is operations work the next step toward autonomy?
What does an AI that operates the factory do day to day?
Where does Morsa fit in autonomous manufacturing?
What changed at J4S?
Which plants are ready for autonomous operations?
What is autonomous manufacturing?
Yokogawa Electric separates two ideas people mix up. Automation “performs a sequence of highly structured pre-programmed tasks requiring human oversight and intervention.” In industrial autonomy, “plant assets and operations have learning and adaptive capabilities that allow responses with minimal human interaction” (Yokogawa). A robot that repeats a weld is automation. A system that sees the weld drifting, finds the cause and corrects it within set limits is autonomy.
EY, writing in April 2026, calls the goal lights-out operations, the ability to run “facilities and processes with minimal human intervention,” and names three domains: autonomous planning, autonomous warehousing and dark factory production (EY). Each runs one loop: sense, decide, act, check. An autonomous manufacturing system is any part of the plant that runs that loop by itself, from one work cell to one planning cycle. A person decides where the loop stops. For the tools themselves, see how AI is used in manufacturing.
What are the levels of autonomy in manufacturing?
The clearest published scale is Yokogawa’s IA2IA maturity model, short for industrial automation to industrial autonomy.
Level (Yokogawa) | What runs on its own | What people still do |
|---|---|---|
0 to 1: Manual or semi-automated | Some production steps | Many operations, with paper instructions and records |
2: Automated | Most production processes | Safe operation, and anything outside normal operations |
3: Semi-autonomous | Selected components that learn and self-optimize | Orchestrate those components |
4: Autonomous orchestration | Most assets, synchronized under set conditions | Many tasks, since not all disciplines are integrated; take control when conditions are not met |
5: Autonomous operations | Everything, across disciplines and out to supply chain partners | Nothing. Yokogawa calls it “a highly idealized state” |
Autonomy spreads one function at a time: Yokogawa lists process control, planning and scheduling, supply chain management, field operations, maintenance and engineering as separate domains, so a plant can sit at level 4 in one and level 2 in another. And it comes in steps: “To jump directly to autonomous operations is very difficult to achieve.”
The World Economic Forum’s January 2026 Lighthouse report adds a second axis, the risk of a bad decision. Its matrix runs from “human only” for high-risk, ambiguous calls to “AI as an operator,” which “runs operations, human oversees.” For moderate risk and moderate complexity it names “conditional autonomy,” where AI “executes tasks, human reviews exceptions.” Most daily operations work sits in that box.
Is an autonomous factory the same as a lights-out factory?
No. Siemens defines a lights-out factory, “also known as a dark factory,” as one that “operates with minimal human activity, allowing it to function in the dark with zero human intervention onsite” (Siemens). Lights-out is about who is on site. Autonomy is about who decides. A plant can be highly autonomous with a full crew, and a dark line can be plain automation running a fixed program.
Siemens notes the limit: “The lights-out factory is easily implemented for simple mass production of a standard product on a fixed schedule.” For complex, high-variant work it proposes a “lights-sparse” floor: crewless areas that gain “the advantages of autonomous production” without a fully dark plant.
The average US plant is nowhere near dark. In the Census Bureau’s survey for the second quarter of 2026, plants averaged 66 operating hours a week, out of 168, and used an estimated 78.5% of their full production capacity.
How far have the most advanced plants got?
Further than anyone, and still mostly short of closed-loop autonomy. On 22 June 2026 the World Economic Forum added 16 sites to its Global Lighthouse Network, which now counts 238 of the world’s most advanced manufacturing and supply chain operations. Kiva Allgood, Managing Director at the World Economic Forum: “The world’s leading manufacturers are no longer optimizing individual processes; they are reimagining entire operating systems.”
HOW FAR THE BEST PLANTS ARE
Closed-loop autonomy is still the exception
238
sites in the WEF Global Lighthouse Network after 16 new awards
WEF, 22 June 2026
~10%
of Lighthouse solutions are closed-loop, autonomous systems
WEF white paper, January 2026
5%
of top Lighthouse solutions in 2025 were enabled by AI agents; about 62% were analytical AI and machine learning
WEF white paper, January 2026
READ TOGETHER
A Lighthouse deploys more than 50 solutions on average, typically 5 to 10 of them flagship. Autonomy arrives one closed loop at a time.
WEF press release, 22 June 2026, and white paper, January 2026, Figures 18 and 19 and page 44.
The WEF’s January report is plain about the ceiling: “While fully autonomous agentic AI remains theoretical in the production context, Lighthouses are leading the way in managing hybrid human-agent workforces.” The loops that do close are narrow. At Haier’s site in Shanghai, a self-adaptive welding programming agent cut programming time from 16 hours to 1 hour.
How far has the average plant got?
Most plants are still connecting the foundations. Rockwell Automation’s July 2026 survey of 1,560 manufacturing decision makers in 17 countries found that 93% have an MES, but only 28% have deployed it enterprise-wide and 23% report full integration across ERP, PLM, quality and OT systems. The same respondents expect 42% of processes to be AI-supported within a year and 54% by 2030, yet 43% say they are not effectively using the data they collect. Anthony Murphy, vice president of product management at Rockwell Automation: “MES adoption is no longer the hurdle, but enterprise scale is.”
Machines are further along. The International Federation of Robotics reported on 24 September 2026 that a record 5 million industrial robots were operating worldwide in 2025, up 9%, after more than 600,000 installations that year. Jane Heffner, President of the IFR: “Industrial automation is progressing at high speed.” The machines are being automated faster than the decisions around them.
Why is operations work the next step toward autonomy?
Because that is where the loop is still open. The work between systems runs on people: a buyer chasing a supplier, a planner rebuilding the schedule after a shortage, a supervisor writing up the shift. The pressure on that work is rising. The Institute for Supply Management’s August 2026 report found supplier deliveries slower for the ninth month in a row and production materials taking 84 days on average. Increasing lead times came up in 46% of negative comments. In the same Census survey, 22.2% of plants running below capacity cited insufficient labor. In the National Association of Manufacturers’ third-quarter 2026 survey of 220 manufacturers, 54.93% named attracting and retaining a quality workforce among their biggest challenges.
Lighthouses show what closing that loop is worth. Midea’s site in Wuhu runs an exception-driven supply chain control tower across more than 10 systems and cut delivery lead time by 39% (WEF, January 2026). Schneider Electric’s El Paso site raised on-time delivery from 61% to 97% (WEF, 22 June 2026). Hitachi Vantara’s site in Norman, Oklahoma, applied agentic AI to demand forecasting and inventory and reports a 77% shorter order-to-shipment lead time and 50% less inventory.
What does an AI that operates the factory do day to day?
It runs the operations work between the ERP, the MES and the people, inside rules the plant sets. In the WEF’s terms, that is conditional autonomy: AI executes, people review the exceptions.
Work | What it reads | What it does |
|---|---|---|
Supplier follow-up | Open purchase orders, confirmed dates, supplier emails and messages | Confirms dates before they are due, treats silence as a slip, ties a late line to the orders it will stop |
Replanning | The schedule, work order status, material arrivals, machines and labor | Spots the job that will not finish, proposes a resequence within approved rules, tells the people who must move |
Expediting | Shortages against the bill of materials, customer due dates, dispatch plans | Ranks what to chase by the customer order at risk, then chases it |
Shift handovers | The shift’s messages, holds, stoppages and open jobs | Writes the handover from what happened and gives each open item an owner and a date |
Coordination | Promises made in chat, email and meetings, and the entries that prove them | Tracks each promise to its date, escalates when it goes quiet, closes it on proof |
A person approves anything outside the rules, such as a change to the production plan or a new commitment to a supplier. It works in the channels and systems the plant already runs, such as WhatsApp, Teams, email, SAP Business One, Business Central, Odoo or Excel, or whatever the plant runs on. Each problem in detail: AI for supplier delays, AI for schedule attainment and AI for on-time delivery.
Where does Morsa fit in autonomous manufacturing?
Morsa is the AI that operates the factory for you: it runs the daily operations work of mid-market and enterprise manufacturers, from supplier follow-up to replanning, on top of the systems they already use. Morsa acts, and people approve anything outside the rules the plant sets. On Yokogawa’s map it works in planning and scheduling and in supply chain management, not in process control.
Morsa runs alongside the ERP and the MES and replaces neither. They stay the systems of record, as laid out in Morsa vs MES and ERP. It is not a dashboard or a chatbot that waits to be asked. It does not make a plant lights-out. Machine control stays with the plant’s automation.
A pilot on one live problem comes first. It goes live in three steps: Morsa connects to the ERP and the channels people already use, maps owners, rules and approval limits with the plant, then acts inside those rules. The floor installs no new app. The pilot is free when no implementation work is needed; otherwise there is a minimal implementation cost, refunded if the pilot shows no value. After it, Morsa’s fee is a share of the value created, agreed once the numbers are in. There is no public price list; how pricing works has the detail.
Morsa runs in its own cloud, in your private cloud, or fully on-premise, including the AI models and databases. It is also now onboarding its first wholesale distributors, covered in AI for distributors; the proof below comes from a manufacturing plant.
Get your free Manufacturing AI Profitability Plan
What changed at J4S?
J4S is a 120-person glass manufacturing plant that runs Morsa. It went live in two days, through Connect, Configure and Live, with no new software, no migration and no training. In the first four weeks, on-time completion of operational commitments went from about 30% to about 75%.
Anil Kohli, Production Head at J4S: “Our people don’t have to learn any new software. People just message the way they always have. Morsa coordinates all the messages in the background.” The J4S story has the full account.
Which plants are ready for autonomous operations?
Autonomy in operations pays where the signals already exist and nobody connects them.
Your plant | Fit | Why |
|---|---|---|
An ERP in place, coordination in chat and email | Strong | The signals exist; nobody connects them |
Suppliers often late, expediting is routine | Strong | Follow-up is the work an operating AI runs |
No ERP or stock records at all | Start elsewhere | Put a system of record in first; see the best ERP for manufacturing |
Output limited only by machine capacity | Weak | The fix is a machine, not coordination |
Pick the loop that costs the most while it stays open, pilot it, and count the result in your own numbers. To find that loop in your plant, Get your free Manufacturing AI Profitability Plan.
Sources
Yokogawa Electric, “What is IA2IA?” yokogawa.com
EY, Ehap Sabri and Cate Mork, “Why autonomous manufacturing matters,” 20 April 2026. ey.com
World Economic Forum, “Global Lighthouse Network: Rewiring Operations for Resilience and Impact at Scale,” white paper, January 2026. weforum.org
World Economic Forum, “New Global Lighthouse Sites Demonstrate How AI Is Rewiring Manufacturing and Supply Chains,” press release, 22 June 2026. weforum.org
Siemens, “Lights-out factory.” siemens.com
US Census Bureau, Quarterly Survey of Plant Capacity Utilization, 2026 Quarter 2, posted 14 September 2026. census.gov
Rockwell Automation, “93% of Manufacturers Have MES, But Only 23% Have Fully Integrated It,” 14 July 2026 (vendor research). rockwellautomation.com
International Federation of Robotics, “Five Million Robots now Operate in Factories Globally,” 24 September 2026. ifr.org
Institute for Supply Management, August 2026 Manufacturing PMI Report, via PR Newswire, 1 September 2026. prnewswire.com
National Association of Manufacturers, “2026 Third Quarter Manufacturers’ Outlook Survey,” 14 September 2026. nam.org
Hitachi, Norman, Oklahoma site named a Lighthouse, press release, 15 July 2026 (company-reported results). hitachi.com
Morsa, J4S customer story (Morsa customer data). /customers/j4s
Changelog
24 September 2026: first published. Definition, Yokogawa’s levels, the WEF oversight matrix, lights-out compared, 2026 adoption data, the operations work AI runs, and the J4S result.
