The Compilation Illusion: Why AI Ladder Code Fools the Compiler
When an AI system generates control logic that compiles perfectly but executes catastrophically wrong, what went wrong? This episode examines a real case where two safety-critical requirements conflict in execution order—both are technically correct in isolation, but their interaction creates a dangerous logic failure in PLC systems. We explore published research on LLM-generated industrial control code and ask: why does a clean compile guarantee nothing, and what must we verify before deploying AI-written ladder logic? 00:00 - The Transmitter Fault Scenario 02:45 - How Two Valid Requirements Conflict 05:30 - Recognizing the Logic Error Pattern 08:15 - The Compiler is Blind to Behavior 11:00 - Survey of AI Control Logic Research 14:30 - Why Testing Can't Wait for Production 17:45 - Lessons for Industrial Automation --- Sources & further reading: • All fetched and read on 2026-09-22. • [peer-reviewed, ICSE-SEIP 2024] Fakih et al., *LLM4PLC: Harnessing Large Language Models for • Verifiable Programming of PLCs in Industrial Control Systems*, arXiv:2401.05443 • · https://arxiv.org/html/2401.05443 — doi:10.1145/3639477.3639743.: https://arxiv.org/abs/2401.05443 • [arXiv v2; journal version cited as IEEE TSE 2026, not opened] Liu et al., Agents4PLC • arXiv:2410.14209 — · https://arxiv.org/html/2410.14209 ·: https://arxiv.org/abs/2410.14209 • code: https://github.com/Luoji-zju/Agents4PLC_release • [preprint, not peer-reviewed] Tu et al., *SemaPLC: A Project-Grounded, Verification-Gated Agent • Harness for PLC Code Generation*, arXiv:2608.18565 — ·: https://arxiv.org/abs/2608.18565 • · repo https://github.com/midea-ai/SemaPLC.: https://arxiv.org/html/2608.18565 • [ICSE-SEIP 2026] Koziolek et al., Spec2Control, arXiv:2510.04519 • · https://arxiv.org/html/2510.04519 · track listing: https://arxiv.org/abs/2510.04519 • (seen in: https://conf.researchr.org/track/icse-2026/icse-2026-software-engineering-in-practice • search, not opened). • [preprint, status unknown] Zhang and de Sousa, *Exploring LLM Support for Generating IEC 61131-3 • Graphic Language Programs*, arXiv:2410.15200 — ·: https://arxiv.org/abs/2410.15200 • [preprint, status unknown] Koziolek, Ashiwal, Bandyopadhyay, K R, *Automated Control Logic Test • Case Generation using Large Language Models*, arXiv:2405.01874: https://arxiv.org/abs/2405.01874 • ·: https://arxiv.org/html/2405.01874 • [ICIT 2026 per arXiv comment] Kersting, Rummel, Benndorf, Vendor-Aware Industrial Agents • arXiv:2511.09122: https://arxiv.org/abs/2511.09122 • [preprint, abstract only] Ogundare, Ekpo, Wiggins, *Ladder Logic Translation using Large Language • Models in Industrial Automation*, arXiv:2605.31458: https://arxiv.org/abs/2605.31458 • [peer-reviewed] Jia and Harman, An Analysis and Survey of the Development of Mutation Testing • IEEE TSE 37(5):649–678, 2011, doi:10.1109/TSE.2010.62 — authors' copy • [bibliographic record only] DeMillo, Lipton, Sayward, Hints on Test Data Selection, IEEE • Computer 11(4):34–41, 1978, doi:10.1109/C-M.1978.218136 • [internal] data/series/ai-benchmarking/ep-04-scoring-is-the-hard-part.md • BrewSys/docs/BUILD-PLAN.md; BrewSys/docs/RED-FLAGS.md — . • [pointer only, no number above comes from it] BrewSys feasibility brief This podcast episode was fully generated by AI — research, script, voices, and production. Built with Claude, Piper TTS, and automated pipeline tooling.