Same machines, different speed: Seven lessons from AMNA 2026 day two
Ford, GM, Toyota, Autoliv, Magna and BorgWarner used day two of Automotive Manufacturing North America to cut through the hype on launch speed, AI and automation. Their verdict: technology only scales on strong fundamentals and skilled people.
North American manufacturers are under pressure from several directions at once. Product cycles are getting shorter, capital budgets are tight, plants are being retooled for new programmes, and skilled workers are hard to find. On top of that, Chinese carmakers are setting a new benchmark for launch speed and cost.
Opening day two of Automotive Manufacturing North America (AMNA) 2026 in Detroit, Christopher Ludwig, chief content officer at AMS, said that even though Chinese OEMs do not sell vehicles in the US, their influence is already being felt in competition, launch pace and change management. New technology only delivers at scale when the basics are in place: clear standards, reliable data, operational ownership and skilled people.
Part 1: Launching faster
1. China speed is a process problem, not a technology gap
Closing the launch-speed gap with Chinese OEMs is now a strategic question for North American plants. Speakers said the gap has little to do with equipment.
Walter LaPlante, controls and industrial connectivity supervisor at Ford, has visited Chinese plants as part of Ford’s benchmarking. He said Chinese and North American plants buy from the same small group of equipment suppliers, especially in stamping.
Alberto Ely Vazquez, senior process engineer at General Motors, agreed that China has no “alien technology”. What he saw on a joint North America–China programme was faster decision-making and very close working with suppliers, who work side by side with the carmaker during a launch. However, he said North American plants should find their own pace rather than copy another region’s speed at the cost of safety, quality or reliability.
“In China, their equipment is the same as ours. What’s different is the standard process that people follow.”
Standardisation was the main lever discussed. Ford uses templates for its electrical hardware and PLC software to cut the engineering needed for each launch, and a Common Data Model to standardise how machines report data.
Arun Prakash Mantena, global manager for advanced manufacturing engineering (stamping) at Lucid Motors, said Lucid is standardising equipment, controls architecture and processes between its Arizona plant and its new plant in Saudi Arabia.
For brownfield changeovers, where obsolete components and unknowns in existing lines are common, Ford sets aside six months to a year of engineering and lab testing before the change.
2. Virtual commissioning speeds launches, but physical testing is still needed
Virtual commissioning is now a standard tool for compressing launch time.
Murugan Boominathan, senior analyst for manufacturing solutions at Magna International, said that in the past teams waited for hardware, process and controls inputs one after another. Now stations, routings, failure modes and KPIs are defined virtually up front, so teams arrive at the plant to “validate them, not integrate them”.
But there are limits. Mantena said virtual commissioning shows a part can be built, not that it can be built at rate or at the right quality. Coil and gauge variation, links between equipment, and fault recovery still need physical validation, especially in stamping. Vazquez explained the value in simple terms:
“What is the best time to solve a problem? As soon as you find it.”
3. Breaking silos means traceability that flows upstream
Many launch delays start long before production, when designs reach the plant that are hard or impossible to build. The design-for-manufacturing panel looked at how to close that gap.
Adam Polcha, manufacturing engineering manager at Daimler Truck North America, said one change can involve 100 or so design engineers, and its full impact on the plant is often not understood. In commercial trucks, where almost every vehicle is different, plants can only validate a small share of configurations. He said traceability needs to run from the operator back to where a project’s scope is set.
“If you get somebody to sign off on something and say, yes, it’s fine, when it’s not fine… it’s still not fine.”
“We see that something very clearly cannot be produced within our facilities… And you ask yourself, how did we even get to this point?”
Jaime Moore, senior principal engineer at Toyota Motor North America R&D, warned against sign-offs that are only a formality. Moore pointed to China’s highly integrated industrial sites, where design, suppliers and production sit close together.
Moore’s advice: put more resident engineers in plants, use shared visual data so teams understand each other’s problems, and stop holding hour-long meetings to review open issues.
Part 2: Scaling AI and data
4. AI pilots fail on ownership, not algorithms
Nitin Sethi, vice-president and head of IT at Autoliv Americas, opened the AI sessions with a warning. He cited a study finding that 88% of AI proofs of concept never reach wide deployment. He calls this the “demo graveyard”: projects with a great demo but no plant owner, a strong ROI on paper but no funding to scale, or a good model that does not fit how people work.
Autoliv has seen this first-hand. A computer-vision inspection project costing almost $1m reached only 30–40% accuracy at first. It improved only when quality and automation engineers, not IT, took over labelling false positives. Once accuracy improved, linking defects to machine process data allowed the system to warn maintenance teams before defects occurred.
In another case, an AI tool that explained OEE losses was used by just 20–30% of staff, because it sat outside the normal workflow.
“We are not in a business of selling technology to our people. We are responsible for achieving the right outcomes.”
Sethi set out five questions for any AI project: which plant KPI will it move, by how much and when; who owns it; does it fit the workflow; can it scale; and can you trust the data?
5. Fix the data before you build the twin
The panel on connected data and virtual factories agreed that digital twins and AI depend on consistent, trusted data.
Maria Beemer, global chief engineer for transformation and digitalisation at BorgWarner, said even basic metrics are not always consistent: cost of poor quality is calculated differently across regions, with one including expedited shipping and another not. BorgWarner now agrees key metrics with cross-functional groups before scaling solutions.
Rajeev Kalamdani, director of manufacturing and IIoT analytics at Ford, described an AI tool that finds line constraints using machine-state data. But he admitted that keeping digital models up to date through launch and ramp-up is still hard: “We’ve not been very successful at it.” Asked to fund one project, he chose predictive quality because it is “closest to the customer”.
To build trust, BorgWarner moves each AI use case through stages: reviewer, then coach, then drafter, and only in some cases fully autonomous. Beemer said:
“Just because you’re using AI, it does not absolve you from the responsibility of the decisions being made.”
Part 3: Automation and people
6. AI vision inspection is working; cobots and humanoids are not there yet
The robotics panel focused on automation delivering results today, not humanoid hype.
Parag Patre, senior manager for advanced robotics at Magna International, said AI vision inspection now matches human inspectors on false positives and false negatives in most cases.
Pete Grabowski, chief operating officer at SkillReal, said one customer went from 65 to 80 jobs per hour after inspection stopped being the bottleneck.
Gus Collins, senior manager for manufacturing digital transformation at JTEKT North America, called cobots the biggest disappointment, because they must slow down far below human speed when working near people. He also ended an AMR pilot after the operator said her walk across the plant was the highlight of her day.
“If it’s not solving the problem for the operator, it’s not gonna work. They’ll make sure that it won’t work.”
Patre said physical AI does not have to mean humanoids. Adding vision, force sensing and AI to existing robot arms could open up assembly tasks that are hard to automate today.
Muhammad Anas, digital transformation specialist at Gestamp North America, added that this is not possible without linking quality data to process data.
7. Skills and people are still the biggest challenge
Hiring is hard across the US, and plants in newer manufacturing regions face an extra challenge.
Lazola Tutani, production department manager at Continental in South Carolina, said new hires may arrive from retail jobs to run a half-million-dollar machine, which leads to process variation, scrap and slow onboarding.
Continental has simplified its equipment and training, and uses AR and VR headsets to train new hires.
It has also made its global knowledge database, covering quality incidents and maintenance documents, searchable with AI. Tutani said US plants need to get the basics right before adding AI:
“If you don’t understand how to transform a raw material into a value-add product, and the only thing you know is how to prompt, then you might stand a risk of losing control of your business.”
Raidan Lage, regional smart factory coordinator at ZF Group, supports 11 plants across the Americas. He said he wants shop-floor pushback on new tools, because silence often means people do not care, not that they agree. ZF’s cycle-time monitoring, which shows line leaders where a 150-station line is slowing down in real time, won an internal excellence award.
