Skip to main content
Loading research indexW

AI code passed a tougher test because it had to run on a real factory controller

INVESTOR TAKEAWAY

Programmable logic controllers (PLCs) run industrial plants, and large language models can already generate independent program organization units (POUs) for them. Whether such logic integrates into an existing PLC project and then runs correctly has been checked only in limited tests.

ORIGINAL PAPER

SemaPLC: A Project-Grounded, Verification-Gated Agent Harness for PLC Code Generation

WHAT WE KNOW

We present SemaPLC, a project-grounded and verification-gated agent harness assembled from conventional tools but governed by a strict completion rule. Rather than stopping when the model judges its own output adequate, SemaPLC declares a task complete only when logged external checks confirm it. Those checks cover the specification, the compilation, and the behavior on a live runtime.

WATCH NEXT

Review the primary source, validate the main result, and establish whether any listed-company transmission is direct.

EVIDENCE

What the evidence supports so far

Research signals

agent

reasoning

What remains unverified

The full methodology, effect size, and limitations still require analyst review.

Company impact remains unverified until a direct economic transmission is established.