istock.com/ParradeeKietsirikul
The skills gap permeates throughout manufacturing. Whether it's the technical skills for running a CNC machine, understanding how to program robotic workcells, or retaining the specialized maintenance knowledge needed to keep today's modern plant floor up and running, there is an undeniable workforce deficiency that holds many manufacturers back.
Cybersecurity and artificial intelligence are not excluded from this ongoing need to find and retain qualified workers. Recently, Google pledged $10 million to the Manufacturing Institute to help train 40,000 U.S. manufacturing workers in AI. While this is a great start, Nishkam Batta, the Founder and CEO of GrayCyan, recently sat down to offer some perspective on the ways manufacturing needs to adjust it's approach to AI, and embrace the current and legacy dynamics of its workforce.
Jeff Reinke, editorial director: What is your response when people talk about AI replacing manufacturing workers?
Nishkam Batta, GrayCyan: AI can’t replace manufacturing workers. Their role is changing from manually processing information to supervising, validating, and acting on information faster. Many of the strongest results come from automating repetitive ERP and coordination tasks that pull skilled people away from higher value work.
In most manufacturing environments, AI creates value by reducing coordination overhead, manual handoffs and work that gets stuck between systems.
For example, GrayCyan helped Bottoms Up Beer reduce custom SKU and BOM setup from 25-30 minutes to about three seconds per item, saving more than 940 hours per work year (and it compounds every year) while reducing mis-shipments and production delays.
That’s the real shift: AI is not removing the need for people; it’s removing the low leverage work that should never have depended on them in the first place. The workforce issue is not just about job displacement. It is also about institutional knowledge displacement.
When experienced people leave and take undocumented operational knowledge with them, that becomes an AI problem as much as an HR problem.
JR: Why won’t workforce training alone be enough to make AI work on the plant floor?
NB: Training matters, but AI only works when the business is ready to use it inside real workflows. That’s why I built GrayCyan’s AI Maturity Model, to assess operational readiness rather than just technical readiness. In my experience, the issue is often not whether employees can learn prompts or tools, it’s whether the organization has the structure, clarity, and trust needed for AI to deliver consistent results.
A lot of manufacturers are still evaluating AI readiness by looking at data quality and cloud adoption. That does matter, but it still misses the operational layer where the real friction sits.
Training helps people use AI better, but it can’t compensate for a business that hasn’t defined how work, data, and decisions actually flow. If teams are still working across disconnected systems, unclear ownership, and undocumented processes, training alone isn't going to make AI work reliably.
JR: What about pilot programs? How can manufacturers educate their workforce to help turn pilots into real implementation?
NB: What turns AI pilots into real implementation comes down to two things that most vendors get wrong.
The first is operational readiness. Before a single line of AI is built, manufacturers need to understand where their real friction lives: documented processes, connected systems, and institutional knowledge that is captured instead of trapped in people's heads.
The second is integration over addition. Too often the pattern is the same: demo a polished product, promise ROI, win the contract, and deliver yet another system that employees are expected to adapt to. That change management burden is where implementations stall.
What works better is AI that lives inside the systems manufacturers already rely on; their ERP, their workflows and their existing tools, rather than something new they have to learn alongside everything else.
Get both right and pilots become implementations. Get either one wrong and you end up with another cautionary tale. That's the difference the AI Maturity Model makes in practice, not just identifying where AI belongs, but ensuring the operational conditions are in place for it to actually work when it gets there.
JR: Where does human oversight remain essential?
NB: Decision support, not decision replacement, represents AI’s highest value use case in manufacturing. That’s why there must always be a human in the loop whenever quality, compliance, safety, or customer commitments are at stake.
Trust begins with traceability. People need to see what the system did, why it did it, and how to override it if needed. AI can triage, summarize, route, and suggest, but people are still accountable for the outcomes. That oversight matters even more because so much operating knowledge is still undocumented and sits with experienced teams.
JR: What does Google’s announcement mean for the future of AI in manufacturing?
NB: Google's announcement signals that AI for manufacturing has moved from experimentation into workforce infrastructure, but investment in training only creates value when the operational foundation is there to use AI well.
The manufacturers I work with aren't asking whether AI matters anymore. They're asking where it delivers measurable ROI without disrupting what's already working. The ones that will win in this next phase are not the ones that train fastest. They're the ones that combine training with documented processes, connected systems, and captured institutional knowledge. That's what turns AI from a pilot into a competitive advantage.
