By Greg Breidenbach, AI Transformation Lead & Senior Product Manager at Poka
Key takeaways:
- AI advances in manufacturing have now gone beyond the likes of generative tools. We now have autonomous digital agent co-workers that assess, understand, and enhance everyday practices on the factory floor.
- From troubleshooting to anomaly detection, agentic AI developments help optimize uptime and empower frontline workers.
- Despite the rise in AI agents, it is ultimately the human element in manufacturing that the plant relies on. It takes a human-in-the-loop model to build a healthy work culture and ensure the connection between agentic productivity and the human eye on the manufacturing line are aligned for operational success.
From riding the hype cycle to establishing working case studies, the manufacturing industry has seen plenty of AI developments in the last two years. GenAI enhancements have entered the manufacturing landscape. We now have AI-powered, easily digestible work documents that come in the form of digitally interactive, accessible instructions, intelligent multi-lingual translations that help eliminate language barriers, and teams of workers that can operate equally towards a joint operational goal.
But generative AI is only the start of the AI manufacturing marathon. Digitally-aware manufacturers who are seeking a competitive edge in modern manufacturing are already looking to the next leg. The next autonomous contender is here, and agentic AI is ready to take the baton.
The AI agent frontier is ready to transform operations in the plant
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.
It’s clear that agentic AI has potential in the manufacturing industry. 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.”
A digital helping hand, all hands on the job
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 four to five years. Workers need help in their everyday operations.
Setting human-machine boundaries to synergize connected working
The most benefit to those workers on the factory floor is having agentic AI operating 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.
It’s a relay, not a race
Here’s how we see a clear evolutionary path ahead in the relationship between AI agents and workers on the manufacturing shop floor:
- Stage 1 (present): autonomous aid
Conversational agents are used for content and data support, enforcing rule-based triggers, and performing background tasks. - Stage 2 (near future): eyes on the prize
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, and then asks for user approval. - Stage 3 (going forward): productivity and deviation diagnostics
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.
Agentic AI use cases to autonomously optimize operations
1. The troubleshooting agent supervisor
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 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. Threshold detection before defects arise
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. This may be worth investigating before it escalates.”
3. The prognostic approach to manufacturing performance
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.
The new alliance on factory floors
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?
Beyond the detailed capabilities of agentic AI is the significance of synergized collaboration from the human hand to machine productivity and the autonomous agents. The route to success is laid out from current generative AI aids, to goal-driven agents of the near future, and the productive diagnostic technology that will pave the way for human-machines collaborating on the factory floor of the future.
The manufacturers that will succeed in the next and current age of modern manufacturing will use the right digital tools to close the loop between humans and machines. The finish line is far from near, but the run to a more connected manufacturing workforce continues.
Greg Breidenbach drives digital transformation across manufacturing, telecommunications, and financial services. At Poka, he leads AI product strategy, developing practical, factory-ready AI solutions that improve knowledge access, training, and frontline decision-making. His core strengths include AI/LLM implementation, product strategy, solution architecture, and cross-functional leadership in B2B SaaS environments.
