August 25, 2025 Unlocking Manufacturing Value with AI Agents

Most manufacturers understand the promise of agentic AI. Here is where to invest, and how to scale across their organization.

By: Tom Cozzolino, Chief strategy officer at GFT

Most manufacturers today understand that AI will transform operations. In fact, 80% of manufacturers believe AI will be essential to growing or maintaining their business by 2030.

But as we know, “adopting AI” can mean a lot of different things. It can cover everything from generative tools that write content, to predictive systems that forecast demand or disruptions, to agentic AI, which is starting to gain traction.  Over the past few years, a lot of manufacturers have started using generative AI in small ways like creating training videos or drafting shift summaries for new team leads. It’s helpful, but given the massive shifts we’ve seen in the past year, sorely limited.

Now, the next front of efficiency will be driven by agentic AI. Unlike generative AI, which passively waits for the next request, agentic AI is always on. It can monitor what’s happening, make decisions, and take action automatically, based on a set of reasoning rules. That makes it a powerful tool for boosting efficiency and unlocking complex use cases.

Even though most manufacturers understand the promise of agentic AI, they are getting stuck on where to start, where to invest, and how to scale across their organization. Here, we will go through those topics as agentic AI becomes an industry standard.

Create a strong data foundation, even if you have to build it as you go

AI agents are only as powerful as the data and systems they can access. For most manufacturers, though, data is often stored across multiple different disparate systems that are not inherently interconnected. For example, a manufacturer can have thousands of machines across a single factory, each with its own data that isn’t connected to other machines, let alone connected to the wider organization.

This lack of integration poses a serious challenge. For instance, if an AI agent is programmed to alert a technician when a machine exceeds a performance threshold, but can’t access the larger data context the technician uses to take action, it simply can’t do its job. The value of the agent breaks down at that point of disconnection.

One obvious solution would be to bring all organizational data together on a centralized cloud-based platform, where data access is permissioned across the organization. And while this might seem like a daunting and time consuming task, viable new patterns like data mesh and zero-copy data access can provide manufacturers with the scalable tools they can use to unify and migrate data incrementally. “Stitching” data together in this manner can prove out use cases and drive immediate value.

Additionally, as data harmonization gains traction, AI agents can perform at the speed and scale required to drive tangible efficiencies and justify further investments.

Manufacturers can start this process with agentic AI pilot projects to flush out data gaps and build “good enough” bridges to fuel early projects.

Build small successes that roll up to big ideas

We are truly in an exciting time for AI agents, with endless possibilities and exciting big ideas that haven’t yet been built. In order to successfully move forward with game-changing innovations, manufacturers may want to decompose them into smaller, more achievable wins first. These experiments need to be rigorously planned and instrumented to deliver immediate results while they fit into the larger, more strategic picture. And along the way, teams can quickly understand what works and build internal learnings as they further scale AI.

Many manufacturers today are looking to apply agentic AI to quality control. AI-enabled processes can detect anomalies in product parts by reasoning over images captured with high-resolution cameras, then use robotic arms to physically remove the defective parts off the assembly line. This reduces the amount of down-stream defects that get shipped from the manufacturer, ultimately increasing consistency and customer satisfaction.

These insights can directly feed into Supply chain and procurement. If defect rates are high, for example, the supplier may suspect that they will miss a delivery deadline for a particular part or assembly. Building out an AI agent to automatically scan for alternative vendors could be a great start. And, over time, other agents could aggregate context to evaluate reliability and present options to a procurement leader for their approval. Once approved, the agent can place the new order, helping maintain output without disruption.

As these capabilities further mature, we see a direct line leveraging agents in scenario planning. For example, if a natural disaster disrupts a key raw material source, the agent can model multiple responses. This can include the potential of switching suppliers, absorbing increased costs, or delaying production. It then can show the downstream impact of each scenario on revenue and customer satisfaction. By aggregating multiple scenarios and their outcomes, leadership would be equipped to make well-informed decisions under pressure.

Manufacturers can start on this today by identifying some “big ideas” that can be decomposed into smaller, intermediate use cases with standalone, measurable results. Then start building those cases and linking them with ever-evolving Agent approaches.

Beware of Agent Sprawl

While agentic AI is new to the manufacturing industry, agentic adoption is rooted in the same core principles as overall technology oversight: clarity, accountability, risk mitigation, and value realization. However, we have also seen that many organizations still lack even basic governance for their existing technologies, let alone emerging systems like AI agents.

Effective governance starts with understanding which technology approaches are approved for use, defining their intended scope, and identifying both subject matter experts and the business processes they affect. This visibility enables leadership to assess where AI agents (and other technologies) can add value, what risks they may introduce, and how their intended outcomes will be captured and measured.

Achieving this requires cross-functional collaboration. In manufacturing for instance, shift leaders can provide real-time feedback to technology teams on how agents perform in daily workflows. And of course, change management must also play a key role to support training and set expectations for sustainable adoption.

Robust governance gives executives the confidence that their AI strategy is not only improving operational efficiency, but also safeguarding against risk. In the event of a failure (i.e., an outage or a cyberattack), governance structures help organizations pinpoint affected systems and respond quickly.

Ultimately, this level of oversight is what transforms AI agent initiatives from isolated pilots into enterprise-wide deployments that generate measurable, long-term value and continued funding.

Organizations should form a governance task force specifically focused on agentic AI, if only to catalog activities to reduce duplication of efforts.

We are entering a future where AI agents will start interacting with each other, not just within the organization but across business units. And while we are in the early stages, leaders will be companies that start rolling out their AI agent projects now, with “good enough” data, a “stacked” set of objectives, and a lightweight governance cadence.

About the Author:
Thomas Cozzolino is the Chief Strategy Officer, USA at GFT, a global digital transformation company.

Subscribe to Industry Today