AI IN DIGITAL MANUFACTURING & ENGINEERING

Most AI pilots don't fail on the model. They fail on the kingdom.

Process engineering, project controls, automation, operations, maintenance and IT each guard their own data, their own tools and their own truth. AI only creates value across those boundaries — and that's exactly where 25+ years of ET/OT/IT architecture work becomes the precondition, not an add-on.

27+
years across IT / OT / ET
50++
projects delivered
3
phases — Assist → Augment → Automate
9
kingdoms one change usually crosses
A plant is born in engineering, lives in operations, and answers to the enterprise. Now it also has to explain itself to an AI — and an AI is only as good as the design basis, the standards and the identifiers it's allowed to see.
The IT-Gladiator position on AI in the process industry

Assist, Augment, Automate — a maturity roadmap, not a shopping list

You don't reach phase three by buying a better model. You reach it by fixing what's listed on each card.

PHASE 01 — ASSIST

A very capable colleague

The engineer stays fully in control. AI retrieves, drafts, explains, checks. Every output passes through someone who is still accountable and still competent.

  • Sourced answers across specs, vendor docs, old project reports
  • Drafting URS sections, test procedures, deviation reports
  • First-pass P&ID review against a checklist
PHASE 02 — AUGMENT

A co-engineer in the workflow

AI produces artefacts and proposes decisions that a human reviews and approves — not initiates.

  • Equipment / line / instrument lists cross-checked against design basis
  • Estimate-to-complete forecasts from schedule + cost + actuals
  • Cross-world anomaly detection: historian, MES, CMMS, shift log
PHASE 03 — AUTOMATE

Bounded execution, exception supervision

AI executes defined tasks end-to-end within approved limits. Technically possible today in narrow cases; organizationally the hardest, because someone still signs.

  • Closed-loop energy optimization within operator limits
  • Condition-based work orders raised automatically in the CMMS
  • Self-updating digital twins that track as-built reality

One change. Nine kingdoms. One shared change ID — that doesn't exist.

A reactor needs a bigger heat exchanger. Here's what nine functions have to agree on before that change is actually done — and where, in most organizations, the trace breaks.

01 · PROCESS ENG
P&ID revised
02 · PROJECTS
Schedule impact
03 · COST CONTROL
Forecast by phone call
04 · PROCUREMENT
Vendor quote
05 · AUTOMATION
DCS updated early
06 · DOC CONTROL
Revision released
07 · QA
Assessed at IQ, late
08 · PMO
Quarterly rollup
09 · OPERATIONS
Spares, if remembered
4 of 9 handovers have no shared identifier Phase 1 AI can help at every node Phase 2 needs one change ID flowing through all nine — today it only exists in one person's inbox
MARIA

Every organization has a Maria. Twenty-two years in, she doesn't raise a change request — she calls Thomas in cost control and has a number by the afternoon. She knows which document controller to talk to, and which of the three P&ID revisions in the archive is actually as-built.

None of this is written down. From the outside, the process looks like it works. An AI agent can execute a defined workflow. It cannot walk over to Peter. When Maria's bridge is replaced by a tool, the tool follows the documented process nobody actually uses, and stalls at the first undocumented handover.

Maria retires in four years. Making her knowledge explicit — before automating anything — is the actual project.

One asset, three languages — now with an AI thread through each

The plant is born in engineering, lives in operations, and answers to the enterprise. We work all three layers, and ground the AI in each one honestly.

ET
Engineering Technology
How a plant is born

Integrated engineering & handover

Engineering data dies in silos when nobody designs the information model for the whole asset lifecycle upfront.
  • DEXPI · CFIHOS · ISO 15926
  • Asset Information Requirements
  • IEC 81346 · VDI 2770 · AAS
  • GxP e-signature workflows
WHERE AI GROUNDS ITSELF

P&ID digitization, DEXPI-based tag extraction, and design-basis Q&A — but only once standards licensing is resolved.

6+ months saved · 10–20% cost ↓
OT
Operations Technology
How the plant lives

Data foundation & industrial applications

One site can carry millions of assets split across applications that were never asked to be joinable.
  • Unified Namespace · OPC UA · MQTT
  • MES · Batch · Electronic Records
  • OEE · CMMS · Energy Management
  • Zero Trust OT · read-only DMZ patterns
WHERE AI GROUNDS ITSELF

Cross-world root-cause analysis — joining historian, MES, shift log and CMMS through one asset context model. Read-only, audited, never a direct agent into the OT network.

5–15% baseline improvement
IT
Information Technology
Where shopfloor meets strategy

Integration, data & AI

Two distinct jobs: connecting ET and OT into the enterprise, and running an operating model that still passes an FDA audit.
  • ERP · MES · LIMS · PLM integration
  • Governed data lake-house, FAIR products
  • AI & agents · MCP-based access
  • DevSecOps · Terraform · GAMP 5
WHERE AI GROUNDS ITSELF

The orchestration layer and the data model live here — owned internally, because this is the plant's most sensitive artefact, not a vendor's training set.

Survive & thrive · FDA-defensible

Make, integrate, or rent — the same rule we've always used for people, applied to agents

No owner builds its own pressure-vessel calculation engine. No owner should hand its design basis to a vendor's training set either.

OWN & HOST INTERNALLY

Make

Your process know-how, cost benchmarks and incident history are the actual competitive asset. The data model and the orchestration layer stay yours.

design-basis assistant
project cost forecasting
internal standards Q&A
BADGE, SCOPE, LOG, OFF-BOARD

Integrate

A vendor agent brought into your environment like a contractor — defined access, no path for data to leave, governed like anyone else with a badge.

P&ID digitization agent
drawing consistency checks
tool-specific configuration copilots
SUPPLIER KEEPS IT CURRENT

Rent

For standards-heavy, fast-moving domains, the supplier's job is to stay current — exactly as with a specialist consultancy today.

regulatory intelligence
code compliance checks
contract & claims analysis

What we check before recommending any tool

The gap row of a real user-journey workshop is the readiness backlog — owned by the kingdoms that created it.

01

One process, end to end

Interview the Marias. Map what they actually do. Turn the phone book into a documented process before anything else.

02

One asset ID, one change ID

Mapped across P&ID, historian, MES, maintenance, ERP, procurement and quality. The single highest-leverage investment.

03

Aligned breakdown structures

WBS, cost breakdown structure and engineering deliverable structure map to each other by design, not by monthly reconciliation.

04

Open interfaces, not chat windows

An "AI feature" that only sees its own tool's data is a chat window, not an integration. We demand data handover in every procurement and EPC contract.

05

The standards question, resolved

Legal, procurement and standards management at the same table before phase 2 — deciding what's rented, what's licensed, what's off-limits.

06

The bridge-builders, protected

The people currently holding the process together become the owners of its documentation — not its casualties.

What makes the engagement different

01

Hands-on depth

27+ years from DCS engineering and Infrastructure-as-Code to enterprise architecture and board-level decisions. Direct engagement in architecture sessions and vendor negotiations — including the AI ones.

02

No vendor agenda

Independent counsel. No incentive to sell a product that isn't needed. Honest assessment of what's worth building — including "do not automate this yet."

03

Pay what you actually need

This isn't a full-time occupation, so there's no pressure to sell. Trust and long-term partnership — the kind of engagement where you can also go for a beer afterwards.

“An engineer who signs off on work grounded on second-hand knowledge should understand exactly what they're signing. That applies double once AI is drafting the work.”

— Wolfgang Purrer, Founder

START THE ARENA

An honest second opinion on your IT/OT/ET and AI strategy

Architecture sessions, target-picture workshops, or a straight readiness assessment before you spend money on phase 2.

Send Email