Agentic production planning. Your factory already knows its plan. Nobody was listening.

Production planning today is a spreadsheet war. Sales wants more, the plant says it can't, and the planner sits in between guessing which of the two is lying.

Nobody asks the only witness that can't lie: the machines.

Your presses, ovens, mixers and moulds have been recording, every day, what your factory can really do: where it loses time, which machine is really the fastest on which item, what slows everything down. The answer to your production plan is already in your data. Your software has simply never read it.

Atherya reads it. Then it plans with it.

Scheduling decides when. Planning decides what, where and how much.

Scheduling puts orders in sequence. Planning answers the questions that come before that:

• Which machine should make which item?
• Where is the real bottleneck?
• How much hidden capacity do you already own?
• What will go wrong on tomorrow's shift?

Their planning answers with averages from the catalogue. Atherya answers with the behaviour of every single machine. It's still the same brain that built the signatures, and it's still the same loop: it sees, understands, decides, acts and checks.

1 · Who makes what, and where: from real times, not catalogue times

The same item can run on several presses. Which one should make it? Your ERP says: the one in the routing. Atherya says: the one that really does it best, and is strong enough to take it. The rules are ones a production manager would sign:

• Only machines that have already made it. Atherya doesn't know the tonnage, the mould or the fittings. So it never recommends a machine that has never produced that item.
• Speed per hour of running. Availability depends on the programme, not on the machine, so speed is measured per running hour, shift by shift.
• Faster only if it's really faster. The gap has to be material and statistically solid. No lucky averages.
• Volume moves only where there is room. Atherya looks for hours when the plant was open but that machine was idle.
• The machine that receives volume must be robust. A fragile one that happens to be faster gets flagged, not loaded.
• Heavy items on fragile machines get called out. If an important item sits on a machine that keeps failing, and a robust alternative has already made it, Atherya says so, even if the alternative isn't faster.

That's not optimisation on paper. That's what your best production manager would do, if they had time to look at every shift of every machine.

2 · The real bottleneck: measured, not assumed

Ask anyone in the plant where the bottleneck is and you'll get three answers. Atherya measures it.

Take the ovens. A post-curing oven works in loads: it heats up, holds temperature for hours, then empties. From temperature alone, Atherya rebuilds:

• every load;
• how saturated the oven really is;
• the deadliest number in the plant: how long the oven sits empty while the rest of the plant is working.

If the oven is the constraint, every one of those hours is lost by everyone.

It also measures the buffer between presses and ovens: how long a part waits between its last moulding and its turn in the oven. And it puts a value on one hour of oven time: how many parts pass through it. Suddenly "should we run the oven on Saturday?" isn't an opinion anymore.

3 · The chain: when upstream coughs, downstream catches a cold

A part crosses several families of machines: mixer, then presses, then ovens. Atherya measures how much the downstream slows down when the upstream goes quiet, and which family really saturates its hours.

Then it goes further, and stays honest. Does an unusual compound from the mixer show up later on the presses? The mixer doesn't record which compound it made, so a batch can't be followed to the press. A correlation alone would be easy to sell. Atherya tests the link in time.

It compares days only inside the same month, because summer heat moves both the mixer and the presses, and that would create a fake link. It checks significance by reshuffling the strange days within each month.

If the link is there, you'll know. If it isn't, nobody will sell it to you as if it were.

4 · Moulds, not just presses: who is really to blame

When a press slows down, everybody blames the press. Often it's the mould.

The same mould runs on several presses, and Atherya uses that to separate the blame:

• If an item slows down on every press it runs on, the mould is wearing.
• If a press slows down on every item it runs, it's the press.
• If an item runs on only one press, Atherya checks whether the press's other items slow down too.

Wear is measured over time and over use. A mould that stops too often across several presses is suspect: the failure that looks like the press belongs to the mould.

Your planning system treats a mould as a code in a bill of materials. Atherya treats it as what it is: a part of the factory that ages.

5 · Preheating: the first minutes of the shift, recovered

Every morning, and worse every Monday, your presses start cold and lose the first part of the shift getting up to speed. Everybody knows it. Nobody measures it.

Atherya measures it, and proves it first. For every restart after a long stop it looks at four things:

• how late the first cycle came;
• how cold the machine was compared with its own normal on that item;
• how long it took to settle;
• how many mouldings were lost on the ramp.

If the longer stops really give colder, later starts, the delay is heating, and that's proven, not assumed.

Then comes the plan: for each press, what time to switch it on, after the night and after the weekend, so that it's at speed when the shift begins. Then the check: the same measurements, before and after. The improvement is measured, not promised.

6 · Tomorrow's shift, on one page

Every evening the shift leader gets a short list of what to look at tomorrow, in order of how much it's worth. Every item comes with its proof, why it matters and what to do:

• Drift: a signal that has been out of its normal for days, with what happened the other times this same machine did it.
• Plant-wide cause: many machines out at once, pointing to a common cause such as heat or a compound.
• Rhythm: a machine slower than its own normal on the same item.
• Long changeover: today's mould change took far longer than usual.
• Late start: the machine opened late compared with its norm.
• Idle: a planned shift in which the machine didn't work.

And here's the slap. Atherya replays the list on every day of the past and checks whether the machines it flagged really stopped more than the others afterwards. Without that proof, a list is noise. Most "AI assistants" never even ask the question.

7 · Hidden capacity: the factory you already paid for

Before you buy a new machine, find out how much of the ones you have you're actually using.

Atherya reads every machine's stops from the gaps in its own rhythm and classifies them by itself: weekend, holidays, compound change, micro-stop, long stop. It measures efficiency of use, availability times performance, against the champion for each compound.

That isn't the OEE from the textbook, with a quality figure someone made up. It's the honest version, measurable from the data that really exists. Then it turns the hidden margin into parts you can make with the same factory, without buying anything. No questionnaires, no stopwatches.

8 · And then it acts

All of this feeds one plan:

• effective capacity per machine and per item;
• maintenance windows reserved inside production;
• machines at risk getting nothing new;
• the plan reorganising itself when a line slows down, with your sign-off and recorded execution proof.

The brain moves the work on its board and journals every move. People pin what must not change. Autonomy climbs the ladder: shadow, then proposals, then guardrails. Ratified decisions go back into your ERP and your MES through governed write-back, and Atherya rereads them to check they really happened.

And this is only the beginning. Next, the same brain pulls orders, materials, shifts and energy into the same loop. The plan will know which component is missing, who is on the floor tonight and when power is cheapest, and it will move everything together. No more planning department reconciling five systems by hand: one brain, one plan, one response.

Their roadmap is a slide. Ours is already running on the shop floor, and it climbs one measured rung at a time.

Their planning vs ours

The bottom line

Production planning isn't about forecasting harder. It's about listening to the factory you already have. Every machine has been telling you, every day, what it can do, where it loses time and what's about to go wrong.

They plan around your factory. Atherya plans from inside it.