Inside the rise of manufacturing ‘co-intelligence’ in real factory operations

AI in manufacturing is shifting from single, centralized systems to networks of specialized agents that work alongside people and existing technology. This “co-intelligence” model is proving its value on the factory floor, delivering faster decisions, fewer bottlenecks, and measurable gains without ripping out established infrastructure.

Artificial intelligence has made rapid progress in manufacturing, but its evolution has been uneven. Generative AI, or Gen-AI, so effective in office environments, has delivered far less in the factory. The structured predictability of the shop floor demands something different.

See also: IFS debuts package of ‘digital workers’ in next iteration of agentic AI industrial software

Complex equipment, variable processes, and a global workforce with differing languages and skills are not well-served by systems trained purely to predict the next word. Manufacturing intelligence must be grounded in the real-time realities of production, not abstract text generation.

This is where agentic AI changes the game. Rather than a single monolithic system issuing instructions, manufacturing can now deploy multiple AI agents, each with a defined role, domain knowledge, and decision autonomy.

These agents can work alone or in coordination, responding to events, querying data, and even taking pre-emptive action without waiting for a human prompt. The result is not an autonomous factory in the science-fiction sense, but a collaborative, mixed team of humans and digital specialists, each playing to their strengths.

The concept is known as manufacturing co-Intelligence, a model where AI agents complement rather than replace existing systems and personnel. In this approach, intelligence is composable and situational.

Agents integrate into the operational environment as colleagues would, working alongside MES, SCADA, ERP, and even decades-old databases without demanding wholesale infrastructure changes. The impact can be transformative, not through futuristic visions, but through precise, targeted improvements in operational efficiency.

From deterministic rules to collaborative agents

Traditional deterministic systems excel at repetitive, well-defined tasks, but quickly falter when reality deviates from the script. Language barriers, inconsistent terminology, undocumented methods, and machine-specific quirks are common across global production networks. Attempting to hard-code solutions for such variability is slow, expensive, and fragile.

See also: HighByte releases Intelligence Hub 4.3, expanding agentic AI on factory floors

Agentic AI takes a different approach. Each agent is designed to act much like a skilled human user of the system. It can log into MES modules, interrogate databases, pull data from sensors, and issue instructions to equipment, all in milliseconds.

Where a human might take half an hour to switch between three different systems, an AI agent can do so almost instantly, correlating the information in real time.

Crucially, the agent does not need a fully re-engineered IT stack to function. It can sit on top of existing systems, using APIs or standard interfaces to connect, regardless of whether the underlying technology is a modern database or a 20-year-old spreadsheet.

This allows manufacturers to bypass the high-risk, high-cost rip-and-replace projects that often stall digital transformation. Instead, they can add targeted intelligence exactly where it is needed.

Lessons from the shop floor

Early deployments suggest that this approach can deliver substantial value, with first experiences indicating savings of around $1.18 million per plant annually. For any manufacturer facing the combined pressures of rising costs, skills shortages, and production demands, that level of impact is hard to ignore.

The benefits are further illustrated through an early deployment with SICK, a leading provider of sensor-based industrial automation solutions.

As Niels Syassen, executive board member for technology and digitalization at SICK, explained: “By combining artificial intelligence with our sensor technology, we offer our customers significant added value. This means more productivity, elevated quality and more safety in their production.

Podcast: Do manufacturers dream of 'digital workers'?

He continued: “With this experience, we know what to expect and what to demand. That is why we were quite skeptical about whether AI agents could truly meet the needs of our own operations. Working with Bosch, it quickly became clear that the agent delivers with the production. For us, this is a great example of how industrial AI will enable manufacturing co-intelligence.”

The turning point came during a demonstration. A service technician, encountering a real production problem, tested the AI agent in front of colleagues. The agent produced an answer, which the technician, a recognized expert, declared incorrect.

An hour later, he returned to admit the AI had been right all along, having identified a deeper fault by drawing on data from another plant. This single moment did more to build trust than any slide deck could have achieved.

Trust is essential because introducing AI onto the shop floor is not simply a technical deployment. It is a cultural shift.

Operators need to see that the system can make their work easier, faster, and more reliable—not threaten their role.

Decision-makers need clear evidence of impact on metrics such as overall equipment effectiveness (OEE) and downtime reduction. Successful adoption hinges on convincing both groups that the AI is a capable teammate, not a black-box replacement.

Integrating AI into messy reality

One of the most significant strengths of agentic AI is its ability to thrive in brownfield environments. Plants vary enormously in their digital maturity. Some have sophisticated monitoring and analytics systems; others still rely on manual logging or spreadsheets. Many are a mixture of both, often with different systems running side by side.

Rather than requiring every site to meet a specific technical baseline, manufacturing co-Intelligence allows AI agents to connect to whatever is available. A wide set of connectors enables integration with diverse data sources, from modern APIs to legacy file formats.

See also: Executives advocate for reshaping of workforce following job cuts from AI

Low-code tools mean that integration does not require specialist programmers; production engineers can connect a new data source themselves. This reduces deployment bottlenecks and makes scaling easier.

Data quality remains a challenge in manufacturing, but it is not the showstopper many fear. AI agents can often bridge gaps by recognizing patterns and inconsistencies, where deterministic systems would simply fail.

That said, structured, semantically consistent data still delivers better results. Bosch has observed up to a 60% improvement in AI output quality when data is semantically structured. In multi-site operations, consistent semantics across plants can significantly accelerate troubleshooting and standardise best practices.

The real integration challenges are often organizational rather than technical: securing access rights, navigating compliance policies, and unpicking undocumented processes. Addressing these requires flexibility, co-creation with the customer, and an acceptance that every site will have its own operational quirks.

Rethinking roles and decision-making

The introduction of AI agents changes not only the speed of decision-making but also where those decisions are made. Problems that once required escalation to a central expert can now be resolved directly by the operator at the machine, supported by targeted AI guidance.

This decentralization shortens response times and frees human experts to focus on strategic or complex issues rather than routine troubleshooting.

See also: Taming the data beast is the first step toward smart operations that cannot be skipped

In many cases, this redistribution of responsibility improves morale as well as performance. Operators gain confidence from being able to solve issues independently.

Training becomes faster and more effective when new staff can consult an AI agent for step-by-step guidance drawn from current and historical data. Compliance improves when logs and reports are generated automatically in the correct format, reducing the risk of human error.

The shift also alters the skill profile required. In the past, adding new functionality to manufacturing systems often meant finding scarce software developers.

With agentic AI, production engineers can configure workflows themselves in a no-code environment, reducing dependency on IT backlogs.

See also: Human intelligence plus AI and how supply chains are changing with this collaboration

This is particularly valuable in addressing the manufacturing skills gap, where experienced experts are retiring faster than replacements can be trained. Capturing their knowledge in digital form ensures it remains accessible; the digital colleague can continue to support operations long after the human has left.

Looking ahead

Looking ahead, the most likely outcome is not a fully autonomous factory, but hybrid teams of humans and AI agents working in tandem. This raises its own challenges, particularly in designing intuitive, reliable interfaces for high-pressure environments. Solving these interface problems will be critical to embedding AI agents into daily manufacturing workflows.

See also: Talking to your data: Agentic AI’s utility in process manufacturing

Manufacturing AI is evolving quickly, but its success will not be measured by how much human input it removes. The real value lies in enabling human-led decisions to be made faster, more accurately, and with better use of scarce expertise.

AI agents are not the end-point; they are the connective tissue that allows people, machines, and systems to operate as a single, intelligent whole.

For manufacturing leaders, the lesson is clear. The technology is ready to prove its worth, but the window for competitive advantage is narrow. Start small, run pilots in targeted use cases, demonstrate value, and then scale quickly.

Those who wait for the perfect moment may find their competitors have already built their own teams, not just of people, but of AI colleagues working alongside them.