AI and industry: what AI will not do for you

Generative AI is moving fast in industry: assistants already generate PLC code and HMI screens, and platforms are standardising predictive maintenance. Does that mean industrial engineering is about to be automated? Our view is more nuanced: AI automates what happens on screen. Three areas resist it for the long term.

1. The physical world and the installed base

Plants run equipment designed to last 20 to 50 years. Machine safety remains hard-wired on deterministic hardware, and manufacturers prefer to bring AI in through separate data interfaces (OPC UA, MQTT) rather than inside PLCs. In robotics, Boston Consulting Group estimates that installation and re-engineering account for about 75% of the total cost of a deployment: the work happens on site, not in a code editor.

2. Signed accountability

Cyber Resilience Act, NIS2, Machinery Regulation: each requires documented evidence and an accountable party. An assistant can draft a risk analysis; it does not sign a declaration of conformity. In aviation, EASA currently caps AI components at the lowest criticality levels, with no date announced for higher ones.

3. The hardware

On-target testing, debugging at the boundary between software and electronics, proprietary protocols: these tasks need a bench, an oscilloscope and someone in front of the machine. Coding assistants speed up embedded C, C++ or Rust; validation on hardware stays physical.

What AI will do anyway

• Write PLC code and HMIs: useful, but commissioning and on-site validation remain human.
• Analyse production data: provided the data is collected and reliable, which requires clean integration.
• Draft first versions of documentation: an accelerator for technical files, not a substitute for their validation.

What this means for a manufacturer

Investing in industrial AI starts with foundations AI does not lay on its own: a segmented, monitored OT network, data connected without modifying PLCs, products compliant with the new obligations. That is the logic of our three services: OT cybersecurity, CRA compliance and industrial data integration.

Sources: BCG, “How Physical AI Is Reshaping Robotics Today” (2026); Microsoft Research; EASA AI Concept Paper Issue 03 (via Swiss Aerospace Ventures); OEM Magazine; Foxmere.