The next battleground for industrial AI is not large language models — it's industrial agents.

Many enterprises are now facing a puzzle:
Why are large language models so powerful, yet they haven't delivered the expected impact on the factory floor?

The reason is simple.
What industry truly needs is not a chatbot.
It is an AI that understands the factory, connects to systems, and executes tasks.
In other words:

The next battleground for industrial AI is not about who has a stronger LLM, but who can turn AI into agents that actually get work done.

01Past: Systems Recorded Information, People Made Decisions

What problem did industrial software solve over the past few decades?
It moved people's experience and processes into systems.
MES manages production, PLM manages R&D, EAM manages equipment.
But these systems share one common feature:
they record information, manage processes, and support decision-making.
Yet ultimately, it is still people who make the final decisions.
When equipment alarms, engineers need to analyse it.
When quality anomalies occur, experts need to judge.
When processes need optimisation, experienced staff must make adjustments.
So many enterprises find that: a lot has been done in digitalisation, but true intelligence has not yet materialised.

02Future: Agents Proactively Complete Tasks

What changes do industrial agents bring?
They don't simply answer questions.
Rather, they actively complete a series of tasks around a goal.
For example, when equipment has an anomaly:
In the past: system alarm → human analysis → check documents → find cause → issue maintenance work order.
In the future: agents can automatically read equipment data, combine historical failure cases, query maintenance knowledge bases, analyse possible causes, generate treatment recommendations, and even trigger follow-up processes.

This is where industrial AI truly delivers value.

03Three Core Challenges for Deploying Industrial Agents

However, deploying industrial agents is not easy.
Because industry is not like the internet.
Factories do not have massive amounts of standardised data.
Rather, there are: equipment manuals, process documents, maintenance records, quality reports, engineering experience.
If these scattered industrial knowledge assets are not organised, AI cannot truly understand the enterprise.

So industrial AI deployment must address three core challenges:
First, connect industrial data.
Let AI know what is happening in the factory.
Second, accumulate industrial knowledge.
Let AI know why certain actions are taken.
Third, connect business processes.
Let AI not only analyse but also execute.

04Platforms: Enabling Enterprises to Build Agents Rapidly

This is why Siemens is investing in an industrial AI agent development platform.
The logic behind it is not to create another chatbot.
It is to help industrial enterprises build their own industrial agents faster.
This platform provides a set of development capabilities tailored for industrial scenarios.
For example: enterprises can quickly create agents by describing desired functionality in natural language based on business needs.
No need to develop from the ground up.
At the same time, they can embed their own industrial knowledge into the platform:
including equipment manuals, process standards, maintenance cases, and technical documents – forming their own industrial knowledge base.

Beyond that, it also supports workflow configuration.
This means: agents not only answer questions – they can also connect to existing enterprise systems and tools.
For example: call data interfaces, trigger business processes, and execute specific tasks.
Truly bringing AI into production operations.

05The Development Model of Industrial Software Is Changing

For industrial software vendors and system integrators, such platforms also bring new dynamics.
In the past, developing an industrial application involved: requirements analysis, software development, project implementation – a cycle that could take months or even longer.
In the future, it may become: define scenarios, configure agents, connect data and knowledge, and rapidly validate applications.
The development model of industrial software may change accordingly.

06Future Competition: Not About Who Is Smarter, but Who Understands Industry Better

Of course, industrial AI also faces a critical challenge:
In industrial settings, the most important attribute is not "smartness" – it is "reliability."
An AI giving an incorrect answer may have only minor impact.
But an industrial agent making a wrong decision could affect production and safety.
So the future competition in industrial AI is not just about model capabilities.
More importantly:
who has deeper industry understanding, who masters more industrial knowledge, and who can truly connect to the production floor.

In the future, every manufacturing enterprise may have its own industrial agents.
They may not replace engineers.
But they will become the most important digital assistants alongside engineers.
From software helping people work,
to software working alongside people to get work done.

This may be the era when industrial AI truly begins to transform manufacturing.