Siemens links an AI-designed gripper demo to a 3D printing cell it’s building in Erlangen
Siemens and AGIBOT built an AI agent that designs humanoid-robot grippers from a task description and prepares them for 3D printing. The design side is the demo. Whether tailored parts pay off depends on the Erlangen production cell being built with EOS and AM Solutions, where post-processing is designed in from the start.
This story was produced through MarketScale. See how Industrial IoT teams put it to work with AI Visibility (GEO).
Key takeaways
An AI agent that designs a part is only half a system: Siemens is tying its gripper workflow to a production cell it’s building in Erlangen with post-processing designed in from the start, to see whether tailored parts can scale.
Ready-made pilot scorecard for AI in additive manufacturing: number of design iterations, simulation turnaround and time from concept to build preparation, with Siemens' AM lead citing design work compressed from weeks to a day.
Before buying into agentic engineering, ask which software functions the agent can call (Siemens cites MCP integration) and what human oversight exists to catch costly mistakes.
Free workspace
Turn your Industrial IoT expertise into content.
Record interviews, organize footage, and write with AI on a free trial of the MarketScale platform for qualifying companies. No demo required, no credit card.
Describe the object a robot has to pick up and the task it has to do, and an AI agent designs the gripper. That is the demonstration Siemens announced on November 16, 2026, built with AGIBOT around a gripper for a humanoid robot and headed for Formnext 2026 in Frankfurt.
The gripper gets the attention. The bigger claim in Siemens' press release is about plumbing: the company says this is the first time it has connected agentic, prompt-based engineering to an adaptive production environment, so an individual product requirement can travel from design to manufacturing along one digital thread. The production half is a cell Siemens is creating with EOS and AM Solutions at its Electronics Factory in Erlangen. It is still being built.
For anyone running an additive manufacturing program, that split matters. Software that designs a custom part faster helps only if the plant can print and finish custom parts without a fresh setup every time. Siemens appears to know this, which is why the cell carries as much weight in the announcement as the agent does.
Five steps from a description to a print file
Siemens describes the workflow as an orchestrator. Once the user has described the object and the task, the agent works through the engineering chain in order:
- Selects a suitable gripper type for the object and task
- Adapts the geometry in Siemens Designcenter
- Simulates the adapted design
- Validates the robot interface and the handling
- Prepares the data for additive manufacturing
The connective tissue is MCP integration, which gives the agent access to functions inside Siemens' engineering software so it can call them in sequence. Siemens pitches the benefit as access: people who aren't specialists in every tool could run a complex engineering workflow without operating each tool and each step by hand. Karsten Heuser, Siemens' vice president for additive manufacturing, laid out the logic in the release. Designing variants by hand is slow, agentic workflows can orchestrate the work from a set of requirements, and 3D printing is the natural way to turn each one-off design into a physical part.
Look closely at step one. An agent that selects a gripper type before adapting it suggests a workflow that starts from known designs and tailors them, rather than one that invents geometry from nothing. The release doesn't say which, and for a buyer that distinction shapes how far the approach stretches beyond grippers.
Grippers are a well-chosen first case either way. According to Siemens, each new handling task has meant reworking the gripper step by step, and the output here is a part whose fit can be checked virtually against the robot before anything is printed. Cheap to test in software, expensive to get wrong on the cell: that is a good profile for a first agentic workflow.
The closed loop Siemens' AM lead said was missing
In early October, Jaideep Bangal, Siemens' additive manufacturing and simulation industry lead, gave A3 contributing writer Roy Sarkar a frank read on AI in AM. The applications are real, he said: chatbot-style design assistants, AI-accelerated simulation, and anomaly detection that mines design, simulation and manufacturing data for defects. But they sit in silos, with no single closed loop connecting them yet.
Read against that interview, the November announcement is Siemens' attempt to wire the first stretch of that loop, with design adaptation, simulation, validation and build preparation handed from one step to the next by an agent. Bangal also described the payoff in terms a plant already tracks. AI, he said, is helping AM and design experts compress work from weeks to a day, measured through the number of design iterations, simulation turnaround and time from concept to build preparation.
For a manufacturing engineering manager scoping a pilot, those three measures make a ready scorecard. They also set the bar for judging the demo: an agent that produces a validated, print-ready gripper matters only if it moves concept-to-build-preparation time against the manual baseline the team already has.
The architecture matches how Siemens talks about agentic AI more broadly. On the company's Future Ready podcast, summarized by Conor Peick on the Siemens Digital Industries Software blog, Matthias Loskyll described agentic systems as taking a high-level goal, breaking it into subtasks and executing them across a multi-step workflow, with specialized agents doing the work, orchestration agents managing it and humans keeping oversight to catch costly mistakes. The gripper workflow is a compact, single-product version of that design.
Why the Erlangen cell is the harder half
Variety is cheap on a screen. On a shop floor it costs setups, handling and finishing. Siemens' release says as much: faster development of individual parts puts new demands on downstream production, which has to adapt to different components and process requirements while staying repeatable and scalable.
The Erlangen cell is the answer Siemens is building. EOS brings additive manufacturing expertise, AM Solutions brings post-processing, and Siemens supplies automation, software and AI, all designed to run as one process chain. David Soldan, who heads 3D post-processing technology at AM Solutions, argued in the release that industrializing AM doesn't stop at the printer, and that additive parts are produced efficiently and reproducibly at scale only once downstream steps are integrated into the chain. EOS CEO Marie Niehaus-Langer framed the partnership as a push for an independent European approach to manufacturing, aimed at competitiveness, resilience and sovereignty.
An AI agent can pick a gripper type, reshape it and validate it in software, but every one of those tailored parts still has to come off a printer and through finishing, which is why Siemens designed post-processing into the Erlangen cell from the start.
An AI agent can pick a gripper type, reshape it and validate it in software, but every one of those tailored parts still has to come off a printer and through finishing, which is why Siemens designed post-processing into the Erlangen cell from the start.
The production side has its own AI requirements, and they are less glamorous. Scott Woida, president of Midwest Engineering Systems, told A3 that AI in AM is only as good as the in-situ sensors feeding it, whether a pyrometer for temperature, an IR camera on the weld pool or audio monitoring of the process. On an exhaust manifold for a military vehicle, his team trained AI on a single process variable, weld pool size, and parts two through thirty-two came out far more consistent than the first.
A framework paper in Production and Operations Management makes the general case: scaling AI safely benefits from core operations disciplines such as process control, drift detection and continuous improvement. That is the territory the Erlangen cell has to cover. It's a different job from designing a gripper, and arguably the one that decides whether prompt-designed parts ever reach volume.
Shops this fits, and shops it doesn't
The prompt-to-part model fits best where a plant needs many small, task-specific variants, such as end-of-arm tooling and custom handling parts. Bangal told A3 that automotive and industrial users increasingly use AM for tooling, and that healthcare uses it for personalized products such as hip, spine and jaw implants.
It fits less well where AM's value sits in a few large parts. Woida's customers typically print components of half a meter to a meter or more for defense, aerospace and space, often replacing castings, and Bangal pointed to wire arc additive repair of large components on Navy ships and submarines as a fast-growing use. For those shops, part quality is the constraint. Bangal himself called reliable quality prediction, using layer-by-layer sensor data to flag porosity, cracks, layer mismatch or thermal stress, the biggest KPI AI can bring to AM. A faster design front end does little for them until that problem is solved.
Data is the other boundary. Siemens says integrated software and data chains are essential to join adaptive engineering and adaptive production. Loskyll noted that in brownfield plants, operating knowledge is scattered across PLCs, ERP systems, spreadsheets, paper documents and the heads of experienced technicians. A plant in that condition would likely need data work before an agent has much to orchestrate.
What to check at Formnext
Siemens is pairing the demo with an onboarding tool. Its AM I Navigator, now with Industrial AI added, is meant to help companies identify and prioritize AI applications across AM design and production, then define pilot projects and a roadmap for scaling. That sequencing suits a team that hasn't yet decided whether its first AI budget belongs in the design office or on the printer.
Ask which Designcenter functions the agent can call through MCP, and whether each call is logged. Siemens lists quality, traceability, security and human responsibility as requirements for industrial deployment, so an agent's design decisions should be as auditable as an engineer's.
The signal worth watching comes after the show. If Siemens, EOS and AM Solutions publish repeatability or throughput results from the Erlangen cell running mixed part designs, the digital thread story gets production evidence behind it. In Frankfurt, it's the agent on display.
Sources
Featured companies
Your experts belong here
Every story in MarketScale Industrial IoT starts with a company putting its controls engineers, plant-floor specialists, and integration partners on the record. Buyers are already reading this topic. The only question is whose experts they find.
Plant and controls buyers research deep before contact, and your engineers get to shape that research.
About the author
The MarketScale Newsroom reports on the companies, technologies, and trends shaping 16 B2B industries. It turns primary sources and expert commentary into clear, useful coverage for the people doing the work.