AI is moving fast. Agentic AI is the next evolution in manufacturing, moving beyond generative AI to create autonomous digital co-workers that perceive their environment, analyze and take action — which gives them vital skills such as troubleshooting issues or providing early warnings — all with minimal human intervention. These AI agents will augment — and importantly not replace — humans on the shop floor.
The success of agentic AI in manufacturing relies on a Human-in-the-Loop model to build trust and leverage worker expertise, forging a vital partnership between machine proactivity and human judgment on the factory floor.
The last two years have seen manufacturing begin its journey through the AI hype cycle and all eyes are now firmly set on use case delivery. Over the last 12 months it’s been generative AI applications, powered by Large Language Models (LMMs), actually making their way on to the factory floor to aid workers and deliver productivity gains.
We are beginning to see some serious use cases for generative AI in manufacturing that dramatically accelerate the creation and conversion of dense, lengthy documents into easily digestible formats, such as digital work instructions and engaging video-based guides, reducing deployment time and cost as well as providing smarter search capabilities. We’ve seen how GenAI can empower a diverse workforce by enabling intelligent multilingual transcription of content, breaking down language barriers for better comprehension, safety and quality.
But this technology is only just out of the starting gate. It won’t stand still, and neither will the most digitally-conscious manufacturers. We’re already looking at the next iteration of AI, the Agentic era.
The power of 'agentic AI' in a manufacturing environment
In the basic sense, AI agents are autonomous software systems that use artificial intelligence to perform tasks, reason and make decisions with minimal human intervention. They perceive their environment, plan actions, use tools to execute those actions and importantly, they can learn and adapt over time.
The World Economic Forum and Boston Consulting Group have published an extensive report urging manufacturers to embrace the next AI frontier: “AI agents amplify the manufacturing vision of real-time decision-making, near-autonomous systems and seamless human-machine collaboration. While manufacturing productivity has stagnated over the past decade in markets such as Germany and the United States, this transformative vision presents a significant opportunity to reignite productivity growth and redefine the competitive landscape of industrial operations.”
Current use cases for Agentic AI have the AI act more like a supportive “assistant," a semi-autonomous goal-directed “agent." In manufacturing, we will see much more, with Agentic AI acting as digital AI agents operating as standalone autonomous systems working with workers, giving them confidence and job satisfaction in their work. Moreover, these agents actually free up human worker time to perform their own tasks to a higher standard or focus on other value-add tasks.
Key to this is the symbiotic nature of the agent-human relationship, as the end-user actually helps the agent to do its job better. This kind of support is also a way to better engage and retain shop-floor workers in an industry where retention rates are dwindling. Anecdotally, we have heard customers seeing drops in average tenure over the last decade from 28 years to around 4-5 years. Workers need help in their everyday operations!
Human in the Loop essential: Establishing guardrails and trust while laying an evolutionary path for connected workers
AI is most beneficial to factory floor workers when it operates as a digital co-worker who augments, not replaces, their skills. Realistic short-term value will be to automate repetitive tasks in the background, help with complex tasks or execute simple tasks faster and more efficiently, surfacing insights workers don’t have time to find, and acting as a co-pilot to navigate complex processes.
But manufacturers must build guardrails for Agentic AI to stop it “going off the rails” and remain a team member. This is where a distinction must be made between autonomy vs. being autonomous — agents may be able to act, but users must maintain final approval. This “human-in-the-loop” approach is not just safety — it’s part of augmenting performance and trust.
The path to building a trusting relationship
Here’s how we see a clear evolutionary path ahead in the relationship between AI agents and workers on the manufacturing shop floor:
Phase 1: Today — assistants and automation. Conversational agents are used for content and data support, enforcing rule-based triggers and performing background tasks.
Phase 2: Near future — goal-directed autonomy. Reasoning agents emerge that interpret intent and coordinate sub-agents/tools. For example, when troubleshooting an issue an orchestrator agent routes the right tools, prepares a draft, then asks for user approval.
Phase 3: Looking ahead — proactivity and prognostics. Eventually, agents evolve from reactive to proactive. This is where agents have the potential to detect early signals of breakdowns, deviations or inefficiencies before they fully manifest. They can also perform prognostic analytics to not just predict what will happen, but recommend what actions to take. Over time the AI agents have the ability to learn and improve, bringing in continuous improvement to AI agent workflows.
Three AI Agent manufacturing use cases wherein AI provides the trusting and helping hand
1. “It looks like you need help with that” — troubleshooting with Orchestrator Agents. Take the example of an operator trying to troubleshoot an issue. The “supervisor” (orchestrator) agent interprets the intent (“Dave wants to troubleshoot”), then routes the request to a knowledge-base agent that searches for possible answers. If no solution is found, the orchestrator proposes the next best step: “Do you want me to log this as an issue in Poka for you?” The agent prepares a draft issue report using the context it already has (plant, equipment, etc.) and asks the user for final validation before creating it.
2. “I’ve just noticed this” — Line manager early warnings. Instead of waiting for a threshold breach, an AI Agent can proactively surface trends.
For example, it may notice recurring issues in forklift safety checklists. The agent then flags the pattern to the line manager: “I’m seeing more frequent problems on your forklift checks—maybe worth investigating before it escalates.”
3. “We need to pick up the pace” — Shift performance prediction. Using data such as checklists completed, workforce skills and current line conditions, agents could identify when an entire shift is at risk. For example: “You’re one hour into the shift, some required checklists are missing, and 30% of the team isn’t fully qualified on this machine. These conditions typically lead to lower OEE and quality issues.” Using prognostic capability, the agent could suggest corrective actions, such as moving a worker or prompting checklist completion.
Building trust with the power of two
Interestingly the manufacturing sector has an open mind about implementing AI. Recent IFS AI research “The Invisible Revolution” surveyed 460 senior executives from many of the largest firms and manufacturing organizations to gauge how AI is being embedded and operationalized across core business processes. It found trust in AI extends into critical functions, 78% trust AI in strategic decisions and 84% in automation and operational innovation—the highest among all industries surveyed. It’s on the industry to stay open-minded to imagine how use cases today can help prepare for tomorrow.
Yes, we are still in the earlier stages of Agentic AI adoption on the shop floor and there are many questions still to be answered such as: What’s the right balance between efficiency and human oversight? What cultural or trust barriers need to be addressed on the shop floor?
Partnerships are forged hand in hand
But it remains clear that the true power of Agentic AI lies in establishing a productive partnership between human expertise and machine proactivity. The evolutionary path is clear: from today’s helpful assistants to the goal-directed agents of the near future, and finally to the proactive, prognostic systems that will anticipate issues before they escalate.
The most successful manufacturers will be those who stay curious, open and engaged with this emerging technology. The future of manufacturing productivity rests not on replacing people, but on empowering them with intelligent, proactive partners.