Week 1: Does AI Make CI Obsolete?

Does AI Make CI Obsolete?

I’ve been asked this three times in the past month. Let me give you the honest answer.

After 20+ years, 40+ organisations, 17 years at Shell, and over $1 billion in documented CI savings — here’s what I think.

Both sides are wrong.

The CI practitioners who pretend AI doesn’t matter are delusional. I watched an AI model identify the top three drivers of a quality defect from 18 months of production data in four minutes. A similar project at Shell took my team six weeks.

Six weeks versus four minutes.

If your entire value as a Black Belt is “I can make a Pareto chart and run a regression” — that value is disappearing. A laptop with a monthly subscription can do it now. I’ve trained 300+ practitioners in my career. At least a third of them relied on tool proficiency as their differentiator. That differentiator is gone.

The AI consultants who’ve never walked a shopfloor are equally dangerous.

I visited a manufacturing site where a consulting firm had deployed a $500,000 AI anomaly detection system. Beautiful dashboards. Real-time alerts. Machine learning trained on two years of data.

It flagged 147 anomalies in month one. The operations team ignored every single one.

Why? The consulting firm had never done a process walk. Never spoken to operators. Never mapped the value stream. The AI flagged planned maintenance as anomalies. It flagged routine changeovers as catastrophic events. It flagged normal operating variation as problems.

By week six, every alert went to a folder nobody opened.

That’s what happens when you deploy AI without understanding the process first. Expensive noise.

The real answer is AND, not OR.

AI without CI is building on sand. You’ll automate waste instead of eliminating it.

CI without AI is leaving money on the table. If you’re spending three weeks on analysis that AI could compress to three days, you’re wasting everyone’s time.

Map the process AND use AI to analyse the data. Go to the Gemba AND use machine learning to spot patterns you’d miss. Talk to operators AND use AI to mine thousands of data points at scale.

When I deployed CI across Shell — 12,000 FTEs, $400M+ in savings — the tools changed every year. The thinking never changed. Understand the process. Involve the people. Use data. Sustain through discipline.

AI is a powerful new tool. But a tool without a thinker behind it is expensive noise.

What this means for you:

CI practitioners: Learn AI. Not to become a data scientist. To become a faster problem solver. The Black Belt of 2026 needs to know how to prompt a model and validate its output against process reality.

AI consultants: Learn CI. Go to the shopfloor. Talk to operators before deploying your model. If you’ve never watched the work happen for an hour, you have no business telling that operation how to improve.

Leaders: Don’t choose between them. Build CI capability that’s AI-literate. Deploy AI that’s process-grounded. Either/or will waste money on both.

The question isn’t whether AI makes CI obsolete.

The question is whether you’re willing to evolve — or whether you’ll be the one that becomes obsolete.

Are you a CI practitioner learning AI, or an AI consultant who’s never walked a shopfloor?