#IIoT_World™ is the first Global #DigitalPublication focused 100% on Industrial AI, #IIoT, #ICS #security. #1 Industrial IoT #Influencer.
United States
Joined January 2017
- Tweets38.4K
- Following30.1K
- Followers53.3K
- Likes45K
Scaling an AI model from one factory to 20 exposes every shortcut taken at the single site: tag names drift, clock signals misalign across pipelines, and data ownership fractures across competing teams. Partner content with @InfluxDB. #influxdata_iiot
Tag-based licensing in traditional historians adds a separate cost barrier to scaling: AI models perform better with more connected sensors, but each additional sensor raises licensing cost, meaning the economics of scaling work in the wrong direction for AI.
No single standard covers AI in manufacturing. Gary Tillery of Skkynet says you need a stack: IEC 62443 for OT security, IEC 61508/61511 for safety boundaries, and ISO/IEC 42001 plus NIST AI RMF to prove the system is still the one you approved after model changes. Partner content with @skkynetinc. #skkynet_ai
Tillery's stack includes OPC UA (IEC 62541) as the practical base: secure, vendor-neutral communication with information models so AI reads machine state rather than raw tags, keeping tool calls inside the information model.
When an AI agent recommends a process change on a pharma production line, the recommendation itself is secondary. The defensible record of how the agent reached that recommendation is what regulators actually audit. Merck's team treats AI output as evidence. Partner content with Adlib.
ALCOA requires every pharma record to be Attributable, Legible, Contemporaneous, Original, and Accurate. At Merck, that standard applies to AI-generated decisions with the same force it applies to human ones, which changes how agents must be architected from the start.
Thorlux discovered something shocking during migration: one ingestion routine was executing 8,000 separate database queries to insert a single report. Took minutes. Rewrote it as one batched query. Same operation: now takes seconds.
The database migration revealed the real problem was in the application layer. Read the full case study in the first reply. Partner content with @TigerDatabase. #tigerdata_ai
AI agents carry no sense of reputation, career, or penalty for rationalized misbehavior. Gary Tillery of Skkynet says the gap is closed by making the agent a platform object whose identity outlives any model, with authority re-issued only against the current stack. Partner content with @skkynetinc. #skkynet_ai
Tillery describes a clear operating split. Domain teams, the humans, own the job, meaning the prompt, the tool allowlist, and the evaluation suite that defines whether the agent is still doing its job.
Factories carry 30% to 40% additional cost in spare lines, backup equipment stored for 12 to 18 months without use, and thousands of engineers on call, all because no system could reliably predict when a specific component would fail. The only defense was duplication. Partner content with @AI4ProdOutcomes. #infinite_ai
Prescriptive AI that identifies machine state precisely enough to predict remaining useful life can begin engineering plant redundancy out of plant design itself, reducing spare lines, reallocating on-call engineers, and reconsidering how much backup capacity a production line truly needs.