Andreas Eschbach is the founder and CEO of eschbach, a software company that helps production teams work smarter and enhance collaboration.
Continuous improvement (CI) has always been the goal in process manufacturing. In practice, however, it has rarely been continuous.
Most plants have gotten good at troubleshooting, stabilizing a process fast enough to keep it running. Far fewer have gotten equally good at real problem-solving, which requires understanding why something went wrong and making changes that prevent it from happening in the future. True CI requires moving away from firefighting and temporary fixes to permanent, sustainable change grounded in an understanding of root causes and real-world conditions.
Closing that gap requires time, attention and access to the right information and expertise at the exact moment it’s needed. Agentic AI is positioned to change that with just-in-time coaching, decision support and active root cause analysis.
Why Continuous Improvement Hasn’t Been Continuous
The pressures facing process manufacturers are substantial and growing: tight margins, supply chain shocks, an aging workforce and, especially for pharma manufacturers, strict safety and regulatory requirements. In this environment, CI is a mandatory policy for manufacturers wanting to remain competitive.
Yet CI initiatives have stalled or failed to deliver real results in many facilities. None of this is for lack of established methodology. Structured problem-solving frameworks such as 8D and the Five Whys have existed for decades, and most plants already have them in place. What’s been missing is the ability to apply them consistently, under real conditions, without waiting on a scarce expert.
In response to these pressures, many process manufacturers are already implementing some forms of AI in their systems and workflows. These include smart search tools that surface relevant history in seconds instead of hours, conversational assistants that answer plain-language questions about plant data and predictive maintenance models that flag failing equipment before it fails.
While many of these AI tools have had a measurable impact on key performance indicators (KPIs) such as safety, quality and productivity, they’ve fallen short of moving the needle on CI because they’re still reactive. They’re built to answer when someone asks the right question, not to notice a problem and reason through it unprompted at 2 a.m. under real production pressure with no expert in the room.
What’s needed is a system that doesn’t wait to be asked but can reason across plant data the way an experienced process expert would: continuously, on every shift.
The Shift To Agentic Reasoning
Agentic AI marks a decisive evolution from passive dashboards to active operational partners. Rather than waiting for prompts, agentic systems are goal-driven; they detect anomalies across telemetry and shift logs, formulate hypotheses, test them against historical records and guide floor teams through structured investigations in real time.
Instead of an engineer spending half a day cross-referencing temperature spikes across multiple reactors, an agent can surface the drift unprompted, cross-reference maintenance records, test possible causes and present a verified hypothesis to the line operator.
This transforms CI in three fundamental ways: It replaces passive monitoring with active investigation, bridges the critical night-shift expertise gap and captures shopfloor tribal knowledge so an organization solves a systemic issue once rather than relearning it indefinitely.
Overcoming The Roadblocks To Mainstream Adoption
If the operational upside is so pronounced, why hasn’t every plant implemented agentic CI? Across boardrooms and plant tours, operational leaders consistently raise three critical concerns:
1. Operator Trust And ‘Black Box’ Skepticism
Plant managers and front-line operators are rightly skeptical of software that claims to diagnose physical processes. If an algorithm hallucinates a cause in a chemical or pharmaceutical plant, the consequences can be catastrophic.
Providers must build systems that “show their work.” Adoption depends on auditable reasoning—every hypothesis must link directly to sensor trends, maintenance logs and verified operating procedures so humans remain firmly in control.
2. OT/IT Data Silos And Legacy Infrastructure
Manufacturing data is notoriously messy, trapped in disparate SCADA networks, MES archives and handwritten logbooks. Leaders often fear that implementing agentic AI requires a multiyear, multimillion-dollar data infrastructure overhaul.
The industry must prioritize lightweight integration that can interpret unstructured operational data in place rather than demanding pristine enterprise data lakes before delivering value.
3. Workforce Buy-in And The Threat Of Replacement
Front-line teams often view AI initiatives as monitoring tools or precursors to workforce reductions. Successful adoption requires reframing the technology from replacement to enablement. When leadership frames agentic systems as “digital apprentices” that handle mundane log-digging and protect operators from repeat midnight breakdowns, skepticism turns into advocacy.
To make agentic systems a mainstream manufacturing standard, technology leaders must stop selling futuristic autonomy and start proving shift-level reliability. That means deploying targeted pilots focused on high-frequency, repeatable pain points, establishing clear human-in-the-loop governance and tying AI adoption directly to operator safety and stress reduction.
From Firefighting To CI
When guided investigation is accessible to every operator on every shift, plants stop settling for quick patches and commit to the disciplined work that eliminates recurring downtime. Senior process experts are liberated from routine triage to focus where their judgment is irreplaceable—plant redesigns, capital investments and complex operational strategy.
That’s how continuous improvement finally becomes continuous—equipping front-line teams to solve the challenge in front of them across every shift, everywhere they operate.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?