AI Is Reshaping Industrial Software. Who Wins?

AI-native startups are challenging established industrial software vendors, traditional application boundaries are shifting, and the manufacturing technology stack is becoming increasingly composable. What does this mean for the companies building industrial software and the manufacturers investing in it?

Will AI simply make established industrial software better, or will it fundamentally change which companies dominate the market?

That's the central question I explored with Michael Finocchiaro, founder of ThreadMoat Intelligence, in this episode of the AI in Manufacturing podcast.

Michael tracks an extensive ecosystem of emerging industrial technology companies, interviewing founders, analyzing startups, and following developments across engineering and manufacturing software. His work gives him a broad perspective on how AI is changing the competitive dynamics of an industry historically dominated by a relatively small number of established vendors.

Our conversation explored where AI-native startups have an advantage, which traditional software categories are most vulnerable to disruption, and how the underlying industrial technology architecture is evolving. We also discussed what these shifts mean for manufacturers making long-term technology decisions and what the industrial software landscape could look like over the next five years.

Why Are AI-Native Startups Challenging Established Industrial Software Vendors?

For decades, companies such as Siemens, Dassault Systèmes, PTC, and Rockwell Automation have helped define the industrial software landscape. Their products have become deeply embedded in how manufacturers design, engineer, and operate their facilities.

But many of these platforms were originally architected long before modern AI capabilities existed.

Michael argues that this creates an important competitive opening. In his words, startups win where the incumbents' architecture becomes the problem.

Established vendors have spent decades evolving platforms built around existing data models, application structures, and customer requirements. AI-native startups, meanwhile, have an opportunity to design their products around today's technology capabilities without carrying the same architectural legacy.

One of the most significant differences Michael observes is development and deployment speed. He points to emerging manufacturing software companies claiming to implement capabilities in days or weeks rather than months or years.

Some startups are even demonstrating MES capabilities developed with AI-assisted coding tools over a weekend. While such demonstrations are not equivalent to deploying a production-ready system, they illustrate how dramatically the economics and speed of software development may be changing.

However, Michael doesn't believe established vendors will simply disappear.

Incumbents retain meaningful advantages in regulated industries, established customer relationships, proprietary data, and decades of embedded manufacturing knowledge. These are not easily replicated by a new entrant, regardless of how quickly it can develop software.

The more interesting question is therefore not whether startups will replace incumbents altogether, but which parts of the industrial software market are most exposed to disruption.

Is AI Fundamentally Changing the Industrial Software Stack?

Manufacturing technology has traditionally been organized around relatively distinct application categories: PLM for product lifecycle management, MES for manufacturing execution, SCADA for supervisory control, ERP for enterprise processes, and historians for operational data.

These categories have shaped both how technology vendors develop products and how manufacturers organize their software architectures.

Michael sees these boundaries becoming less rigid.

On the engineering side, technologies that previously operated in relatively separate domains—design, simulation, manufacturing engineering, and production—are increasingly converging.

AI is accelerating this convergence by enabling new forms of automation, orchestration, and information exchange across traditionally disconnected workflows.

On the manufacturing side, new companies are introducing capabilities that sit alongside or extend existing industrial applications, including manufacturing data platforms, composable operations applications, and AI-driven intelligence layers.

But Michael makes an important distinction: the industry may not necessarily be adding entirely new layers to the stack.

Instead, manufacturers are gaining more options for implementing the capabilities associated with each layer.

This creates an opportunity to move away from rigid, tightly coupled software architectures toward systems that can be composed around specific operational and business requirements.

Technologies such as MQTT and OPC UA, alongside increasingly standardized approaches to industrial data exchange, are helping make this possible.

The result could be a market in which manufacturers have considerably more flexibility in how they assemble and evolve their technology environments.

Where Will AI Create the Greatest Disruption in Industrial Software?

When I asked Michael which parts of the industrial technology stack are most vulnerable, his answer was the intelligence and insight layer.

Connectivity, data acquisition, and visualization have become relatively mature capabilities. Many established and emerging vendors can already connect industrial equipment, collect operational data, and make that information available to applications.

The more significant opportunity is what happens after the data becomes available.

How quickly can software identify meaningful operational patterns? How can it connect those insights to engineering and production decisions? And how much of that process can be automated?

Michael sees AI creating opportunities for a new generation of applications that move beyond presenting operational information toward interpreting it and supporting decisions.

This also changes how vendors differentiate.

Rather than competing primarily on connectivity, dashboards, or the number of supported integrations, industrial software companies may increasingly compete on the intelligence they can deliver using operational context.

For manufacturers, this raises a broader architectural question: where should intelligence reside, and how should it interact with the existing systems responsible for executing industrial processes?

What Happens to MES When Intelligence Moves Outside Traditional Applications?

MES is one of the most interesting categories in this transformation.

Traditionally, manufacturing execution systems have played a central role in coordinating production activities, tracking work in progress, and maintaining records of what was produced, when, and where.

But as AI-driven applications and composable software architectures become more common, the role of MES is being reconsidered.

Michael's position is that MES survives, but it may no longer occupy the same central position in the industrial software architecture.

He distinguishes between the deterministic responsibilities of MES and the intelligence increasingly being developed around it.

Manufacturing execution requires predictable and reliable outcomes. Production orders, material movements, equipment operations, and manufacturing records cannot depend on probabilistic AI outputs without appropriate controls.

These requirements don't disappear simply because AI becomes more capable.

Instead, Michael envisions intelligence increasingly operating around traditional execution systems, with AI-driven applications providing additional context, analysis, and decision support while MES continues to perform its essential execution responsibilities.

He also describes competing approaches emerging across the market. Some vendors are extending MES into broader manufacturing operations management platforms, while others are unbundling capabilities into more specialized applications.

The underlying question is whether MES remains the application around which manufacturing operations are organized or becomes one of several foundational systems within a more composable architecture.

Could AI Agents Fundamentally Change Industrial Integration?

One of Michael's more provocative predictions is that AI agents could eventually take over much of the integration work traditionally performed by humans.

Industrial technology architectures have long been shaped by the challenge of integrating systems that use different protocols, data models, interfaces, and application structures.

Manufacturers have invested heavily in connecting these systems through middleware, integration platforms, custom interfaces, and increasingly standardized communication technologies.

Michael argues that as AI agents become more capable, the challenge could shift from building individual integrations toward making information understandable and usable across systems.

In his view, context and intent become more important.

Context helps an AI system understand what operational data represents, how equipment and processes relate to one another, and what a particular event means within a manufacturing environment.

Intent describes what an engineering design, production process, or operational decision is supposed to accomplish.

This doesn't mean that established integration requirements simply disappear. Reliable data exchange, security, interoperability, and deterministic execution will remain important.

Rather, it suggests a potential shift in how integration and orchestration are implemented, with AI agents taking on more of the work currently handled through manually developed application logic and integrations.

How Should Manufacturers Evaluate AI Software and Avoid Vendor Lock-In?

As established vendors and startups introduce new AI capabilities, manufacturing technology leaders face an increasingly complicated software market.

Almost every industrial software company now has an AI proposition. But the level of technical differentiation varies considerably.

Michael argues that one of the first questions buyers should ask is whether a product provides genuine, defensible capabilities or is primarily a thin interface around a general-purpose foundation model.

He identifies several potential sources of differentiation, including proprietary training data, deep workflow integration, industry-specific expertise, and regulatory readiness.

The bigger strategic concern is what happens when manufacturers build important operational processes and institutional knowledge inside a platform they may struggle to leave.

Vendor lock-in isn't simply a question of proprietary file formats.

It can also emerge through application dependencies, data models, integrations, and the commercial incentives of platform vendors.

Michael's advice is that manufacturers should maintain control of their data models and favor architectures built around accessible

‍