Automate 2026 recap: What’s new in industrial AI

Automate 2026 took place last week at McCormick Place in Chicago, from June 22nd to the 25th, and showcased a range of robotics, controls, and industrial software across the show floor. FullStack was on-site, sitting in sessions and talking with teams about how AI is being used in real plants.

AI shows up in many familiar areas—agents, digital twins, vision systems, and robots—but most detailed discussions focus on how these systems behave in production: how they connect to equipment, what they depend on for data, and how operators stay involved when something fails or changes.

Industrial AI agents with Yaskawa Motoman

Sarah Andrzejewski from Yaskawa Motoman led two sessions on AI agents in industrial environments. Because Yaskawa ships the cobots and motion systems these agents run on, the sessions stayed close to plant-floor situations rather than abstract diagrams. Topics included how agents coordinate tasks between robots, how they hand off work without introducing new risks, and how they report state and errors in ways operators can use.

Once agents influence motion or routing, they are part of the control stack. They need clear states, defined fallback behavior, and interfaces that let operators understand and override decisions. Engineering work here covers both agent behavior and the surrounding control logic, so the combined system stays within plant constraints.

Digital twins and alignment with the floor

Digital twins were prominent at Automate, with Siemens, ABB, and others showing models of lines and plants. Conversations with Hugo Roquero and Gunter Krenz from Bosch Manufacturing Solutions focused on the gap between what a twin shows and what the floor does on a typical day.

Plants change regularly. Equipment is updated, schedules are adjusted, and processes are tuned. A twin that doesn’t reflect those changes becomes a view of how the plant used to work. Keeping a twin useful requires steady data integration—ingesting signals from controls and historians, normalizing them, and reconciling them against assumptions in the model so the twin stays aligned.

Software-defined automation from Bosch Rexroth

Bosch Rexroth’s ctrlX AUTOMATION and ctrlX OS highlight a move toward software-defined control systems. In discussions with Jon Frey and Holger Lutzler, control logic was presented more like deployable applications than fixed ladder logic. ctrlX OS exposes APIs and supports containerized workloads.

This approach makes it possible to apply familiar software practices to factory controls. Logic can be versioned, changes can be monitored, and interfaces can be secured at the API level. Safety and uptime remain central, but update and observability mechanisms follow patterns that teams already use in other production systems.

OT/IT security with Wibu-Systems and AMDT

Security appeared across the Automate agenda. Marcellus Buchheit of Wibu-Systems spoke about preparing industrial software for the EU Cyber Resilience Act, looking at licensing, IP protection, and secure updates as design topics that need attention early in development. Riley Groves from AMDT, in his “Not on Your Watch” session, focused on operational resilience when plants face active attacks.

Identity, access, logging, update paths, and recovery strategies are described as system-level concerns. They sit alongside control and data decisions and are considered when systems are designed, not only when they are audited.

Vision AI and physical AI at the edge

Teledyne FLIR and Allied Vision sessions in the advanced vision track looked at moving models from lab setups to live production lines. Alex Finkelstein and Matthew Hori described how vibration, changing light, and throughput requirements affect models that performed well in controlled tests. Thermal and infrared imaging are now widely used for inspection and maintenance, but they have to run on embedded hardware under line conditions.

Physical AI and demo-based programming appeared in talks from Marc Fuentes at Eclipse Automation, Chris Matthieu at Intel RealSense, and Josh Leath at Yaskawa. Robots are taught tasks through examples, voice, and vision instead of fully scripted programs. The engineering effort in these projects goes into collecting and structuring data, validating learned behavior, and managing how models move into production, in a way that is similar to other AI product lifecycles.

Open standards and plant data connectivity

Plant connectivity is another steady theme. Yuta Inoue from Nidec DTC talked about drives and motors that now provide data for predictive maintenance and control. Andrew Sollecito from AutomaTech described connecting SCADA, MES, ERP, and IIoT systems that were built at different times by different vendors.

AI efforts in manufacturing rely on those connections. Without well-designed interfaces, schemas, and data pipelines, the information AI systems need remains fragmented across equipment and platforms, and it is harder to train and deploy models that do useful work.

Robots doing more than demos

Automate’s robot booths showed AI in concrete tasks. Boston Dynamics’ Spot ran inspection routes on the Automate floor, coordinated by Orbit, its fleet management software. Atlas has begun shipping units into industrial and research environments. Mech-Mind Robotics demonstrated 3D vision systems that attach to existing arms and let them handle unfamiliar parts. Dobot highlighted Atom-W, where embodied AI watches parts on a line and routes them without each shape being explicitly scripted.

The hardware in these examples is well known. The newer work is in vision, coordination, and the software that ties those capabilities into everyday workflows.

Our takeaway from Automate 2026

Automate 2026 may be over, but the innovation shows no signs of slowing. AI is present across agents, twins, robots, and edge systems. The questions that come up most often are about systems: how these tools connect to the plant, how they use data, how they are observed, and how failure is handled.

FullStack’s teams work on those system questions in AI projects, and it was incredible getting to learn and share with the manufacturers addressing them on the ground. If you’re working through similar issues—moving from a demo to something that fits into your production stack—let’s get in touch to explore your opportunities.