AI: Shaping the future of autonomous operations

Agentic AI shift at scale

The speakers describe how AI has progressed from question-answering to agentic systems that can plan, check, and act. These systems often run multi-step workflows with little or no humans-in-the-loop.

AI orchestration systems (Open-source momentum)

A key example mentioned is Openclaw, described as an orchestration system for agentic AI. The speaker claims it grew extremely fast—reaching widely adopted usage on GitHub within roughly six months.

Siemens vision: chatbot → shop-floor co-worker

Siemens frames “agentic AI” not as a chatbot, but as a co-worker on the shop floor, aimed at greater production autonomy.

Definition: “autonomous operations” in manufacturing

Autonomous operations are described as:

• An AI-centric system with agents
• Increasing independence and reducing manual human intervention
• Built on top of existing industrial systems

Foundation stack required

The talks emphasize layering autonomous capabilities on standard manufacturing systems, including:

• SCADA
• MES
• IoT
• PLM

These systems provide the initial data/control foundation needed before agents can operate effectively.

Industrial knowledge graph as the integration layer

A major architectural point is that MES/IoT/PLM data is often siloed. An industrial knowledge graph is used to:

• Connect and contextualize data across silos
• Provide relationships that agents can use to reason and coordinate actions

Agentic AI on top of contextualized data

Once agents can access MES/IoT/PLM capabilities and are guided by knowledge graph relationships, the speakers argue this enables movement toward closer-to-autonomous production, with human involvement reduced step-wise over time.

Gap between vertical AI and end-to-end optimization

One speaker notes that many manufacturers use AI in vertical use cases (e.g., planning, maintenance, product solutioning) but lack horizontal/cross-solution integration to optimize end-to-end workflows.

The core barrier is framed as:

Scalable application and data readiness, not missing AI technology

Analogy: autonomous driving

Value can be created without full autonomy. The speakers compare the approach to early stages of autonomous driving—deploying AI where it improves cost/efficiency, even if full autonomy is not yet achieved.

---

Real-world Siemens examples (shop-floor use cases)

Erlangen electronic factory + Nvidia partnership

Siemens is investigating physical AI and agentic AI for shop-floor operations. Factories are treated as internal “customers” to prioritize urgent, practical use cases.

Warehouse/logistics robotics (“wave three”)

Two robots use VLAM (visual language action models) for flexible grasping. The approach is described as avoiding the need to train for every step; instead it identifies what to grasp (e.g., picking and placing into boxes).

Autonomous production planning (rescheduling under disruption)

The system targets high-variability production (example scale: ~1,000 products and ~24,000 materials). When parts are missing (e.g., a delayed truck), it supports rescheduling and adapting to changing material flow.

Orchestration support for non-end-to-end Siemens customers (Accenture angle)

When customers’ systems are not fully Siemens end-to-end, Accenture is positioned as helping orchestrate orchestration systems across heterogeneous environments.

Pop-up/temporary production agents (sports/Pepsi/Gatorade example)

Agents help determine which product variants are needed for different event contexts. The example frames this as adapting for semi-finals/finals where demand requirements are uncertain.

---

Concrete example workflow (machine-level “agent” debugging)

1) Injection molding quality variation detection

An agent detects quality variation caused by size differences in produced plastic parts.

2) Root cause analysis via contextualized industrial knowledge graph

Another agent analyzes contextual industrial data and finds:

• The quality index correlates with material viscosity
• Viscosity correlates with the age of the “plasterolate” (an autogenerated term referring to stored material/plastic compound)

3) Action recommendation

The agent recommends introducing a FIFO consumption process so the system uses the newest material first, reducing viscosity fluctuation.

4) Human-facing output and future automation

• The recommendation is returned to the machine operator.
• The speaker claims humans could handle this in daily standups, but expects it to increasingly be handled by agentic systems.

---

Organizational and transformation analysis (Accenture perspective)

The discussion frames a “three-body problem” across:

1. Technology, people, processes
2. AI affects core decision-making, not just an IT layer
3. You can’t simply overlay AI on fragmented processes/data—organizations must adapt processes to be AI-ready
4. AI systems are non-deterministic, so teams need time to trust, adopt, and use them correctly in structured industrial environments

Key adoption dynamic

Once AI creates value in daily work, organizations are more likely to adopt it and progress together.

---

Misconceptions / pitfalls (explicit “traps”)

The talk lists three common misconceptions about AI and autonomous production:

1. “AI is a magic box” — false; it requires solid foundations and ongoing digital transformation.
2. “Once deployed, it runs forever” — false; requires monitoring for model drift and potentially re-evaluating models, especially for self-improving systems.
3. “Throw more data in and it gets better” — false; depends on data quality, especially context/structure, as well as human–AI collaboration and guardrails.

---

Calls to action

• The speakers encourage organizations to act now and begin the AI journey.
• AI readiness is framed as something organizations should start immediately.

---

Main speakers/sources (as mentioned)

• Accenture (speaker referenced as Vlad, later Ronny)
• Siemens (speakers/mentions include Roland Bush and Jensen Wang, tied to the Erlangen/Nvidia announcement/cooperation)
• Nvidia (partner referenced for physical/agentic AI on the shop floor)
• Openclaw (orchestration system discussed as a key example)