Siemens and Chinese humanoid robot maker AGIBOT demonstrated an agentic AI workflow at Formnext 2026 in Frankfurt that designs a robot gripper from a single text description, then prepares it for 3D printing without an engineer touching CAD software. The demo ran at Siemens' booth in Hall 12.1, booth 119D, running November 17 to 21.
A user types out what object the gripper needs to handle and what task it must perform. From there, the AI agent selects a suitable gripper type, adapts the geometry inside Siemens' Designcenter software, runs a handling simulation, validates the fit against the robot's interface, and outputs a file ready for additive manufacturing, according to 3D Printing Industry. The test case the companies showed was a gripper built for one of AGIBOT's humanoid robots.
Collapsing a multi-week engineering job
Designing and validating a custom gripper for a new handling task has traditionally meant a mechanical engineer iterating through CAD, running simulations by hand, and coordinating separately with manufacturing. Siemens' pitch is that agentic AI collapses that sequence into one orchestrated workflow that a non-specialist can trigger with a sentence.
"Instead of manually designing variants in a time-consuming process, agentic workflows can orchestrate engineering tasks based on specific requirements," said Karsten Heuser, Siemens' vice president of additive manufacturing. The claim Siemens is making publicly for the first time is that this prompt-based engineering now connects directly into an adaptive production environment, so a single design request can travel from description to physical part along one continuous digital thread.
That production side is being built separately: Siemens is standing up an adaptive manufacturing cell at its Electronics Factory in Erlangen, Germany, with 3D-printing specialist EOS supplying the additive hardware and AM Solutions handling post-processing. Siemens contributes automation, software and the AI layer, with the goal of running design, printing and finishing as one integrated chain rather than three handoffs between vendors.
Why gripper design matters for humanoid robots
Grippers are one of the more fragmented parts of the robotics supply chain. Unlike robot arms or control software, end-of-arm tooling is often custom-built per task, which makes it slow and expensive to adapt a robot to a new job. AGIBOT and rivals including Unitree, Figure and UBTech are all racing to get humanoid robots handling a wider range of real-world objects, and gripper versatility is a recurring bottleneck.
Siemens is betting that generative, AI-driven gripper design paired with on-demand 3D printing shortens that bottleneck from weeks to days. If the Erlangen cell works as intended, a factory could in theory request a new gripper variant for an unfamiliar part and have it printed and fitted before a batch production run starts, rather than waiting on a specialist supplier.
Siemens has not disclosed pricing, lead times for the Erlangen cell, or a date when the workflow will be available to customers beyond the Formnext demonstration. The next milestone to watch is whether Siemens turns the showcase into a commercial service, and whether AGIBOT or other humanoid makers adopt AI-generated, 3D-printed tooling as a standard part of deploying robots into new tasks.
