Can factory workers become the developers of their own AI?
A factory operator confronted with a fault on a packaging machine may have to search through a binder of instructions, even though the same person can find a video explaining how to repair something at home in seconds. That contradiction helped inspire Poka, a connected worker platform originally built to make practical manufacturing knowledge easier to share. Its latest ambition goes considerably further, allowing the people responsible for running production to help create the AI applications they need.
Antoine Bisson, co-founder and chief executive of Poka, an IFS company, traced the company’s origins to a conversation with fellow founder Alexandre (Alex) Leclerc about the way employees were trained on factory equipment. Speaking at IFS Unleashed in Orlando, Bisson explained how that initial emphasis on accessible work instructions has evolved into a platform for building intelligent agents around the problems manufacturers encounter every day. Poka, acquired by IFS in 2023, launched its no-code AI Studio at the event, targeting plant managers, continuous improvement teams and other operations specialists rather than conventional software developers.
The proposition raises a practical question for manufacturers. Employees may understand why a production line repeatedly stops or how an experienced colleague resolves a quality problem, but turning that understanding into software has traditionally required technical specialists. Poka believes AI Studio can narrow that gap, although Bisson is clear that making tools easier to use does not eliminate every difficulty involved in deploying them.
Starting with what the factory knows
Poka’s original focus was on capturing the experience of people who understand how equipment works, then making it available to colleagues through videos, instructions and training materials. That accumulated information now provides the foundation for agents intended to answer questions and assist with operational tasks. A worker investigating an equipment fault, for example, could describe the symptoms rather than search manually through several documents.
“Troubleshooting is one of the first applications,” Bisson says. “An operator can describe what is happening on the shop floor, and the assistant can involve an agent specifically designed to investigate that problem. It searches the knowledge database, which contains the work instructions, videos and other information the company has collected over the years.”
The approach depends on manufacturers having usable knowledge to search, a problem that Bisson acknowledges has historically slowed digital projects. Important information may exist in documents scattered across departments or reside with employees who have spent decades operating particular machines. For businesses starting without a structured knowledge base, collecting and organizing those materials can be a substantial undertaking before an intelligent assistant has anything reliable to draw upon.
Poka is now using AI to reduce that initial effort by importing existing digital material, including PDFs, Word documents and work instructions, alongside recordings of experienced employees. The intention is to bring established knowledge into the system without requiring a company to recreate every training resource from the beginning.
“When we first launched Poka, getting everything into the platform was a tedious task, particularly collecting recordings and information,” he adds. “Now customers can import their existing PDFs, work instructions and Word documents, and AI helps build that knowledge base. It removes a considerable barrier to getting started.”
Letting operations teams design the agents
AI Studio builds on that information by offering manufacturers a no-code environment for developing agents that reflect their own procedures. The launch includes ready-made templates as well as the ability to configure new agents, with Poka presenting the approach as suitable for employees who know the work but have limited programming experience. Bisson described applications ranging from troubleshooting to workforce qualification, with different agents handling defined tasks rather than expecting one general-purpose assistant to manage everything.
“If somebody cannot come in for a shift, we can have a qualification agent that looks at the schedule and identifies the most qualified person to replace them,” he explains. “We are also mapping agents across the processes used for continuous improvement, so they become part of how manufacturers standardize and improve work, rather than being isolated applications.”
The launch also includes agents that review outstanding production issues, escalate them to the appropriate people or identify abnormal values entered on forms. Bisson sees considerable variation between factories, which is why the company is allowing users to adapt predefined agents instead of requiring every customer to adopt identical workflows. His emphasis remains on supporting operators and supervisors in their existing responsibilities, rather than handing production decisions to autonomous systems.
“We have always said that the goal of AI and agents on the shop floor is to augment workers, not replace them,” he continues. “We provide customers with a set of predefined agents, but every manufacturer is different. AI Studio allows them to customize those agents to the way they work.”
Making an agent easy to configure does not, however, mean every member of a production team will immediately be able to build one. Bisson identifies technical confidence as a persistent barrier and says Poka offers templates and engineering assistance to help manufacturers develop their initial applications. Decisions about how much authority those agents receive, including when employees must approve an action, remain with the customer. He also made no claim that Poka itself determines whether a company’s IT department must sign off on an application before it goes live.
“Technology is still the biggest barrier,” he notes. “Our customers are extremely good at building production lines and manufacturing products, but technology is not necessarily their strength. We have tried to make the tools as easy to use as possible, and we have engineers who can help customers build their use cases.”
Connecting factory knowledge to the wider business
A second part of the announcement involves Poka’s Model Context Protocol, or MCP, server, which makes information held in the connected worker platform accessible to supported enterprise AI assistants. Manufacturers using tools such as Microsoft Copilot can potentially bring frontline work instructions, employee skills information and operational records into queries involving other business systems. This is distinct from using AI Studio to configure an agent within Poka, although both developments are intended to make factory information more useful beyond its original application.
Bisson describes a customer that used early access to the MCP connection to create an application for moving forms out of an outdated software system and into Poka. By exposing Poka’s application programming interface through MCP, the company is also making it easier for developers and less technical users of AI-assisted coding tools to build integrations using natural-language instructions. These capabilities still require appropriate decisions about access, data quality and how an application is deployed.
The potential becomes broader when manufacturers can compare production information with records of employee qualifications and training. Bisson suggested that a company could investigate whether particular skills are associated with stronger production runs and then use that analysis to inform training priorities. This was a proposed use case rather than a reported customer result, but it illustrates how the existing connected worker platform could contribute to decisions elsewhere in the business.
Early testing of AI Studio and its MCP server has involved manufacturers including Mars and Nestlé, although Bisson was unable to share specific operational results because they remain confidential. He said customers had made greater use of the integration capabilities than originally anticipated, reflecting growing demand for ways to connect AI tools across different enterprise systems. Initial agent prototypes can be developed within days, with deployment potentially taking weeks, although measurable improvements from the early-access programme have not yet been made public.
The suitability of these applications will also depend on the type of manufacturing operation involved. Different production environments have different requirements for managing employee knowledge, work instructions and operational procedures. Bisson acknowledges that Poka’s capabilities are better established in some sectors than others.
“We are especially effective in process-driven manufacturing,” he says. “Food and beverage, consumer packaged goods and packaging are areas where we have a great deal of experience. We are expanding into other sectors, including oil and gas, but process-driven operations remain our strength.”
Whether no-code tools make factory employees effective developers will ultimately depend on what they can build, how safely those applications operate and whether they reduce the work involved in resolving production problems. Poka’s approach gives operations teams a more direct role in shaping the software they use, without suggesting that engineering expertise or governance can simply disappear. For Bisson, the next stage is to expand the range of agents available while making them easier for manufacturers to develop and connect to the systems already running their businesses.
