MES · ERP · Artificial intelligence in manufacturing
The truth about MES, ERP and their “artificial intelligence”
Over the last two years, every management system has become intelligent. Same interface, same tables, same modules. Plus one label: AI.
An assistant answering natural-language questions. A ‘predictive’ dashboard. A button summarising the week's data. A presentation full of agents, copilots and autonomous factories.
Ask one simple question. After the ‘AI’ update, does your management system truly know how your machines are doing? No. It did not know before, and it does not know now.
What MES and ERP really are
An ERP holds the company's numbers: orders, materials, costs and deliveries. A MES records what happens on the shop floor: who made what, when and on which machine. They are serious, useful and often indispensable tools.
But they are records. They hold the plan and note what happened. They do not understand machines.
The cycle time used for planning comes from the catalogue. Capacity comes from the nameplate. To the management system, a press slowing for weeks runs at the same speed as on installation day. An oven sitting empty for half its time while the plant works is ‘available’.
The system does not lie. It simply does not know.
MES and ERP record important facts. They do not automatically learn the physical behaviour of machines.
The ‘AI’ they are selling you
Artificial intelligence has now been added to those records. In most cases, this is what it means.
- 01A new way to question old data. Ask ‘how much did line three produce last month?’ instead of opening a report. Convenient, but the answer is unchanged and comes from the same tables. If the data does not know a machine is deteriorating, neither will the assistant.
- 02Generic models laid on top. An anomaly algorithm that does not know your products, compounds, restarts or seasons. It finds everything unusual, and everything in a factory is unusual. You get more alarms and no more answers.
- 03More modules to buy. Predictive maintenance is one module. Energy another. Advanced planning another. Each has its project, integration and consultants. Each knows only its own piece.
- 04Presentations about autonomous factories. Agents deciding and plants governing themselves. Ask which process the system truly owns today, how it is verified, and who answers when it is wrong. The answer is usually a demo.
Artificial intelligence is not a label
Artificial intelligence is not a label to stick on a product. Either the system knows your factory or it does not. No language model, however brilliant, can tell you the health of a machine nobody has learned to read.
The real problem is not inside the software. It is between systems.
A factory runs many systems today: one for maintenance, one for planning, one for energy, one for the shop floor. Each knows one piece.
Who connects them? People.
A machine slows down. Someone must notice, decide whether maintenance is needed, recalculate capacity, move orders, notify the customer and reorganise teams. All by hand, across systems that do not speak, with spreadsheets in between.
That coordination is the real cost in your factory. No ‘intelligent’ management system removes it. It merely makes it prettier to look at.
Each system sees one piece. Coordination remains with people.
Software holds the plan. Machines decide what is possible.
Every planning system on the market plans against nameplate capacity: nominal cycle multiplied by available hours. It cannot do otherwise because it does not know the machines.
But a press's real capacity today is its actual cycle on that compound, with its health today, its typical stops and its ongoing degradation. A plan built on nameplate capacity is built around a factory that does not exist.
The problem is not the planning algorithm. It is the data the algorithm plans on.
A sophisticated algorithm on nominal capacity optimises a factory that does not exist.
What Atherya does
Atherya is not a management system with AI on top. It is the opposite: factory knowledge from which decisions follow.
- 01It learns how machines truly behave. For every machine, product, season and restart: its normal, what it really produces, and when it leaves health. It learns from history the factory already holds, without questionnaires or months of configuration.
- 02It plans on effective capacity: observed rate, multiplied by real availability and health today. A stable chronic defect is not punished; active degradation is, because tomorrow the machine will produce less.
- 03It connects consequences. When a machine slows, Atherya knows what it touches: capacity, plan, deliveries and maintenance. One brain interprets the change instead of leaving people to reconstruct it across five systems.
- 04It measures without inventing. Missing data is not filled with a default. An indicator that cannot be calculated stays empty, with a written reason. A system inventing numbers cannot be trusted.
- 05It explains everything it says. Every alarm, capacity and proposal carries its evidence: which data, against which reference, and why.
- 06It cannot cause damage. Atherya writes only ratified decisions, with your signed permission, through predefined operations and after a dry run. It reads the system back to verify the result.
Real autonomy is earned, not announced
This is the greatest difference, and it is about honesty.
We too are building a system that takes on production management, always with your sign-off. But autonomy does not switch on with an update. It is earned one step at a time, through evidence.
First it observes. Atherya calculates its plan beside yours, touching nothing, and records where it would act differently and why.
Then it proposes. When the difference matters, a proposal reaches a person with all the evidence. The person decides. A rejected move is not proposed forever. A manually fixed choice is respected.
Only then does it act, with your sign-off and only within explicit limits. Never on critical orders. Never towards an at-risk machine. Only after the same proposal has proved itself repeatedly. Every decision is signed, so you always know what a person decided and what the system decided. Stepping down a level takes one click.
Anyone promising an autonomous factory tomorrow morning is selling a presentation. We show you the step we are on and the proof needed to reach the next.
Autonomy is earned through evidence, not activated with an AI label.
Why Atherya
Because we do not start from software. We start from the factory.
Management systems know what you planned. Maintenance systems know what sensors measure. Neither knows what your machines can truly do today. Atherya does, and that knowledge makes every later decision meaningful.
We do not sell features. Features are consequences: predictive maintenance, real capacity, the bottleneck and the plan. They all come from one thing — plant knowledge.
Atherya integrates with your existing ERP and MES if you want. Over time, it can also replace them. Your choice. Your call.
The truth in one line
A management system with AI is still a management system. It knows what you wrote, not what is happening.
Today's factories run software, and people hold the pieces together. We are building the brain that knows the machines, coordinates the consequences and takes on the work one verified step at a time, with your sign-off.
Today, factories run software. We are building the brain that will run the factory.
The Atherya team
Do not add another AI label.
Discover the brain that reads machines, connects consequences and works above the systems you already have.
Discover Atherya