AutomationBy Johanna Neuron
Robots on the Shop Floor, AI Agents in the Office: Why Automation Only Pays Off as a Whole
Robots that load machines and stack pallets have arrived in German factories. Companies like RobCo from Munich now rent them out. But the bigger lever is not on the shop floor, it is in the office: according to an analysis by the McKinsey Global Institute, 82% of the automation potential in German manufacturing lies in AI agents for office processes and only 18% in robotics. Companies that think both together, shop floor and office as one continuous workflow, can raise their productivity substantially. A worked example below arrives at roughly 35%.
The starting point: 0.3% instead of 1.6%
Germany has a productivity problem, and it is measurable. A study by the German Economic Institute (IW) for the Stiftung Familienunternehmen, reported by Handelsblatt in August 2026, finds that productivity grew by around 2% a year in the 1980s and 1990s and by about 1% in the 2010s. Over the past six years the average was 0.3%. To offset demographic change, keeping today's prosperity with fewer people in work, growth would have to rise to 1.6% a year. Five times the current pace.
At the same time, the people who could do that work are missing. According to the DIHK Skilled Labour Report 2025/2026, 36% of companies cannot fill at least some of their open positions, and among small and mid-sized companies the figure is above 40%. The cause is not the business cycle; in a good year the number would be higher. It is the cohorts retiring and the ones not coming after them.
For a mid-sized company this means: over the next ten years it will have to deliver at least the same output with fewer people, while competitors in the US and Asia deploy automation at scale for exactly that reason. The question is no longer whether to automate, but where to start and how far to think.
What Physical AI can do today: the RobCo example
For a long time, robotics was something for mass production. An industrial robot needed a safety cell, months of integration and a programmer for every change. That is changing, and a German company has come to symbolise it.
RobCo from Munich builds modular robots for typical production tasks: loading and unloading machines, palletising, moving material, finishing surfaces. The hardware is assembled from standard modules; the software adapts the robot to each workflow. According to Startup Valley, the company raised 100 million US dollars in February 2026 from investors including Lightspeed, Lingotto and Sequoia. BMW, T-Systems and the connector manufacturer Rosenberger are named as customers. IT-Boltwise reports more than 1,000 systems in the field and a valuation above one billion dollars. The next step is called "Alfie", a two-armed robot meant to perceive its surroundings and act on its own.
Three things about this matter for mid-sized companies:
- Rent instead of buy. RobCo offers its robots as "Robotics as a Service" with no upfront investment. The barrier that used to sit at several hundred thousand euros is gone.
- Adapt instead of reprogram. New variants are taught through software rather than weeks of rebuilding. That makes robotics interesting for small batches too.
- The tasks are ordinary. This is not science fiction. It is pallets, boxes and machine doors, exactly the work companies can no longer find people for.
At aiworx we do not build robots and we have no partnership with RobCo. We use the example because it shows how quickly physical automation is becoming accessible. And because it makes a gap visible.
The blind spot: 82% of the potential is in the office
In May 2026 the McKinsey Global Institute examined where AI agents and robots can take over work in Europe. For Germany the analysis produces three numbers:
- 59% of working hours are technically automatable with today's technology, the highest share in Europe.
- By 2030 German companies could unlock up to 486 billion US dollars of productivity potential, more than the UK (375 billion) and France (238 billion).
- In manufacturing, the largest sector at around 112 billion dollars, 82% of that potential lies in AI agents for office processes and 18% in robotics.
Consistent with that: 35% of employees in Germany work in roles that consist largely of desk work, such as accounting, administration and clerical processing, according to the study. That is the highest share in Europe.
These figures describe what is technically possible, not what happens tomorrow. But they overturn a common picture: when people talk about automation, most think of the shop floor. The larger part of the work an AI system can take over happens before and after, at the desk.
What an AI agent handles in the office
An AI agent is a digital colleague that handles a workflow from start to finish. It reads, understands, pulls data from your systems and gets it done. Three workflows from a manufacturing company:
| Workflow | Today | With an AI agent |
|---|---|---|
| Order intake | Order arrives as a PDF by email, gets typed in, items are looked up, delivery date is checked | Agent reads the order, creates it in the ERP, checks availability, proposes a date |
| Delivery note and invoice | Delivery note compared with order, invoice created, sent, payment monitored | Agent reconciles, creates the invoice, sends it, follows up on late payment |
| Complaint | Email lands in the inbox, gets forwarded, queries back and forth, credit note after days | Agent identifies the issue, pulls order and delivery note, prepares credit note or reply, a person approves |
How this looks in detail for invoices is on our page on AI invoice processing; for delivery notes and forms, see document processing.
A case from practice: logistics and packaging
One of our customers is a logistics and packaging company that we do not name here. There, AI agents now handle three workflows: they read incoming quotes and hand the data over to the ERP system. They sort incoming mail, digital and scanned paper alike, and route it to the right people. And they pre-process customer support requests, so the team only approves or takes over the cases that really need experience. The core of the experience: the warehouse had long been tuned for throughput. The bottleneck sat in the office, where every shipment was handled several times before it was allowed to leave the building.
The worked example: why together adds up to more than the sum
Take a fictional metal processing company with 80 employees, 60 of them in production and 20 in the office (order processing, purchasing, scheduling, accounting). The figures are assumptions, guided by the studies above and by our projects. In the workflows we automate, effort drops by around 30% on average.
Step 1: Robots on the shop floor. Two stations for machine loading and palletising run in three shifts. Assumption: a quarter of production hours goes into such tasks, and 60% of that can be handed to robots. Result: 15% fewer working hours in production.
Step 2: AI agents in the office. Assumption: half of office hours is routine (capturing, reconciling, forwarding, chasing), and 70% of that can be handed to agents. Result: 35% fewer working hours in the office. That sits within what McKinsey describes as 59% technically automatable hours, and below what we see in specific workflows.
Step 3: Add it up. 60 × 15% + 20 × 35% = 9 + 7 = 16 full-time equivalents. That is 20% fewer working hours for the same output.
Step 4: The real effect. Now comes the part a single department never gets. When order, production, delivery note and invoice interlock without idle time, lead time falls. An order that used to take ten working days from intake to dispatch takes six or seven. With the same team, the company can accept more orders. Assumption: 8% more output.
Result: 8% more output with 20% fewer working hours. Productivity per working hour rises by 1.08 ÷ 0.80 = 1.35, roughly 35%.
A company that only buys the robot reaches 11% in this example. One that only automates the office reaches 9%. Only both together, plus the shorter lead time, produce the jump. The exact figure differs from company to company. The logic does not.
Why the combination delivers so much
Three things disappear when shop floor and office run as one workflow:
- Waiting time. An order no longer sits in an inbox for two days until someone has time to enter it. The machine gets its next job as soon as the previous one is done.
- Handovers. Every handover from person to person costs time and creates errors. An agent hands over to the robot, the robot reports back, the agent creates the delivery note.
- Queries. Most queries arise because information is missing. When the agent knows order, stock and delivery date, nobody has to pick up the phone.
This is also the difference from classic process automation. Where RPA replays rigid click sequences, an AI agent understands a free-text order or an angry complaint email. More on that in our comparison AI agent or RPA.
Where to start: the office first
If you want both, start in the office. For three reasons:
- It is faster. A pilot for one workflow takes us two to four weeks. A robot needs floor space, a safety concept and a connection to the machine.
- It is cheaper. A pilot starts at €2,500 net, operation from €250 a month. We break down what automation costs in What does AI automation cost?.
- It prepares the shop floor. A robot waiting for hand-typed orders is of little use. Clean order data is the precondition for physical automation to run through at all.
The typical path looks like this: in the free intro call we identify the workflow with the biggest lever, usually order intake or invoice processing. The pilot runs on real data. Then further workflows follow, and by then at the latest it pays to look at the shop floor. How we approach this in manufacturing companies is on our page AI in manufacturing.
The opportunity for German companies
The study figures can also be read the other way round: Germany has the largest automation potential in Europe because so much work here sits in processes that can be automated. Industry is strong, workflows are documented, the data exists. What is usually missing is only the start.
The skills shortage will not go away. But a company that hands its routine work to robots and agents needs fewer people for the same output, and the people it has do the work they were hired for.
Frequently asked questions
What is Physical AI? AI that acts in the physical world: robots that perceive their surroundings through cameras and sensors and adapt their movements, instead of just following fixed paths. Companies like RobCo use it so robots learn new tasks faster.
Does aiworx build robots too? No. We build AI agents for office processes and custom software. For the shop floor there are specialised providers. What we do is make sure the data and workflows around production run cleanly enough for physical automation to take hold.
Where does the 35% in the worked example come from? From assumptions that are disclosed above: 15% fewer hours in production, 35% fewer in the office, 8% more output from shorter lead times. In your company the values may be higher or lower. In the intro call we calculate with your numbers.
Is this only for large companies? No. Rental models for robots and pilots from €2,500 for AI agents put both within reach of companies from around 20 employees. What matters is not size but whether there are recurring workflows.
What happens to the employees? Nobody has been replaced in our projects so far. The hours freed up go into work that used to be left undone: customer contact, quotes, quality. And into positions that could not be filled anyway.
How do I start? With a workflow that occurs often and involves many manual steps. In the free intro call we find it in 30 minutes.
