By Daniel Okafor

Hours, sometimes days: that is how long it still takes most automotive plants to turn a detected defect into a corrective action. Miguel Saldamando, CEO writing in Mexico Business News, argues that in 2026 "agentic AI" collapses that gap to minutes — and that for Mexico's Tier 1 suppliers, the timing is not incidental.

Unlike the analytical AI most plants already run, agentic AI doesn't stop at presenting conclusions. Saldamando defines it plainly: it "observes, decides, executes, and learns, operating within boundaries defined by the organization but without waiting for a human to press the button."

That is a different animal from a machine-vision system that rejects a bad part and moves on.

What changes on the shop floor?

Today's plants are, in Saldamando's phrase, rich in data and poor in reaction time. Eddy current, ultrasound, machine vision and X-ray inspection generate real-time information. SPC charts update continuously. Traceability ties every part to its process parameters. Yet the response chain remains stubbornly human: operator notices alarm, technician investigates, quality engineer analyzes, decision lands hours or days later.

An agentic system shortens that chain. Saldamando sketches the workflow: when reject rates on a characteristic climb, the agent correlates inspection results with variables like temperature, tool wear or cycle time, forms a root-cause hypothesis, adjusts parameters within pre-approved limits, verifies the fix and documents everything. If it exceeds its authority, it escalates with a full diagnostic already prepared.

In his words: "That is the difference between a system that informs and a system that acts."

The building blocks, he notes, already sit in most modern facilities — networked inspection equipment, PLCs with bidirectional industrial protocols, historians full of process data. What matured over the last two years is the orchestration layer: the agent connecting perception to decision to action, driven by advances in large language models and multi-agent architectures.

Takeaway: this is an integration story, not a rip-and-replace story.

Why Mexico, why now?

Saldamando, who works with Tier 1 suppliers and manufacturing plants, frames 2026 as consequential for three reasons:

• Mexican automotive production enters the year with strong volumes
• OEMs are tightening expectations on part-level traceability
• Plants face a persistent shortage of experienced quality technicians and certified NDT inspectors

He's careful on the labor question. Agentic AI "does not eliminate the need for that talent; it multiplies it." A specialist once watching one line can now supervise a fleet of agents, stepping in only where human judgment genuinely matters. Given that finding and retaining skilled inspectors is "one of the most common complaints I hear from plant managers," he calls this a practical answer, not a futuristic luxury.

Takeaway: the technology is arriving at the same moment the labor gap it addresses is widest.

What are the guardrails?

Saldamando applies the same discipline to agentic systems as to any inspection technology — arguably stricter. His conditions:

• Explicit boundaries: which parameters an agent may adjust, within which limits, and which decisions always require human approval
• Every action logged and auditable, because customers and auditors will ask why the system acted
• Validation against historical scenarios and edge cases before touching a live process
• Workforce training for supervision, not operation — "a genuinely different skill"

He also flags the cultural hurdle: organizations still treat AI as a recommendation engine. Trust must be earned gradually. He expects successful implementations to start with low-risk autonomy — documentation, scheduling, alert triage — before expanding to process adjustments.

"Autonomy is not granted; it is accumulated, one verified decision at a time."

What to watch next

The claim to verify against production data: whether early adopters actually report containment times falling from hours to minutes, as Saldamando's containment scenario implies — quarantining suspect lots, notifying the affected customer program and issuing sorting instructions autonomously. Also worth tracking: documentation work. If agents start drafting 8D reports and PPAP updates, quality engineers shift from compiling to approving, and that's a measurable headcount-hour change.

Saldamando's verdict on the endpoint is blunt: "the plants that respond in seconds instead of days will be the ones that define the next standard of quality."

That's a supplier-tier claim, not a confirmed capacity figure. Treat it as an intention announced, and watch the first auditable case studies.