Today's industrial landscape is generating vast amounts of data from SCADA systems, IoT sensors, and edge gateways. However, traditional rule-based, deterministic control systems lack the reasoning and dynamic adaptability needed to manage complex, multivariate processes; optimize multiple parameters in real time; respond to shifting supply-chain dynamics; and bridge the growing skills gap of an aging workforce. Across Hitachi Ventures, AENU, and b2venture, we see the emergence of industrial AI agents—autonomous, reasoning systems that span perception, planning, and action—as the critical next step to bridge this “agency gap” (see the Gartner visual) and unlock dramatic gains in efficiency, resilience, and sustainability.

Building on 2024's foundation of General Purpose Agents, the industry is now following a SaaS-inspired three-phase evolution. This evolution comprises vertical embedding in year one, cross-vertical plug-and-play in year two, and full platform ecosystems thereafter and paves the way for startups to establish defensible moats through data lock-in, developer ecosystems, and orchestration layers as AI agents unbundle the enterprise stack one vertical at a time. These systems combine autonomy, continuous learning, and agency, allowing them to self-manage tasks in unpredictable, real-time industrial settings without preprogrammed instructions. This makes them ideal for use in manufacturing, mobility, and energy infrastructure. Startups that architect for full-loop ownership—from sensing to decision-making to action—will be best positioned to create lasting value. Closed-loop agents, who own the process from start to finish rather than serving as components or enablers, unlock deeper integration, richer data feedback, and higher switching costs. While full-loop ownership is ideal, in practice, most agents today operate as subsystems—augmenting specific industrial tasks rather than orchestrating entire workflows. These agents are often embedded within broader systems, offering modular gains in areas like root cause analysis, control tuning, or anomaly detection. The path to full autonomy will likely be incremental, layered over time atop existing control infrastructure.

‍

Gartner predicts that agentic AI will be embedded in one-third of all enterprise applications by 2028, up from less than 1% in 2024. This growth is fueled by the following converging trends:

• ‍Favorable Economics: The economics of running AI agents have undercut the cost of specialized human labor.‍
• Increasing AI Maturity:
  
  • Frameworks such as LangGraph, ReAct (Reason + Act), and Reflexion have advanced to support enterprise- and industrial-grade agents.
  • The emergence of protocols like the Model Context Protocol (MCP) is poised to accelerate the adoption of AI agents by enabling more dynamic and intelligent data pipelines. MCP exemplifies the shift from static integrations to context-aware systems. It allows AI agents and LLMs to access and orchestrate relevant data autonomously and in real time. For instance, an MCP-enabled pipeline could enable an agent to request particular subsets of streaming data or automatically ingest new sources when responding to a query. These capabilities are crucial for scaling agent-based systems in enterprise and industrial settings.
  • This allows agents to support stateful reasoning, which involves remembering information from previous interactions to make decisions or solve problems. Additionally, it also enables agents to support multi-step planning, memory, retry logic, and real-time decision flows. These capabilities make agents suitable for deployment in complex industrial environments, replacing one-shot AI.‍
• Geopolitical pressures: Escalating trade tensions, regional conflicts, and maritime disruptions have created unprecedented supply chain volatility. Industrial AI agents deliver strategic value by continuously monitoring global developments, autonomously reconfiguring sourcing strategies, and implementing contingency plans in real-time—transforming supply chain resilience from crisis management into competitive advantage.‍
• Decarbonization mandates: In the EU alone, industry accounts for 25% of energy consumption, and there are binding targets to cut 2022 consumption levels by 11.7% until 2030. AI agents can proactively reduce emissions in high-intensity sectors, such as chemicals, metals, and heavy industry, by continuously optimizing processes for energy and material efficiency.‍
• Workforce constraints: Shrinking operator expertise and labor shortages make autonomous decision-making a necessity rather than a luxury.‍
• Market Education is well under way: Enterprises are no longer asking if they should use AI. Now, they're asking, "Which process should we delegate first?"‍
• High Frequency, Feedback-Rich Use Cases see early adoption: Initial adoption is often concentrated in high-frequency, feedback-rich workflows, such as production scheduling, setpoint optimization, and process tuning, where agents can quickly learn and demonstrate their value to earn the trust of operators.

‍

Industry-specific AI agents are emerging as a transformational force, surpassing general-purpose AI by integrating real-time sensing, adaptive learning, and autonomous control to manage complex industrial operations.

Key use cases we are seeing:

• ‍Manufacturing & Production Operations (i.e. AI-driven process control, production scheduling)
  
  • Nexus’ platform combines a cloud-backed Process Engine that easily integrates into existing controller architectures, bringing support for real-time Python and open-source alongside AI-native architectures. Its AI agents integrate with industrial controls to analyze data and provide automatic recommendations for continuous improvement.
  • Juna AI develops reinforcement learning (RL) driven industrial control agents that plug into existing process systems to optimize multiple targets simultaneously (e.g. output, quality, and energy).‍
• Asset Management & Predictive Maintenance (i.e. Monitoring equipment health, anomaly detection, triggering maintenance actions autonomously)
  
  • Augury, a unicorn startup and category leader in AI-driven machine health, is incorporating more "agentic" features that automatically recommend parts or adjustments.‍
• Supply Chain & Procurement Automation (i.e. Autonomous inventory planning, logistics optimization, procurement automation, contract generation)
  
  • Mandel.ai builds autonomous agents that act as AI-powered supply chain planners. Their system dynamically ingests enterprise data (e.g. ERP, procurement, logistics) and uses agentic planning to identify bottlenecks, optimize workflows, and generate proactive recommendations without human prompting.‍
• Energy Optimization & Agent-driven Process Controls (i.e. Reinforcement learning agents managing power usage, thermal stability, and energy efficiency; auto scheduling of new jobs in production)
  
  • Phaidra uses deep reinforcement learning to sit atop existing control systems and tune cooling setpoints in real time, cutting data center cooling energy by 20–40%.
  • Brick’s agent acts as a fully autonomous building energy manager, challenging legacy solutions from incumbents like Johnson Controls or Siemens that use simpler “co-pilot” control logicfile‍
• Enterprise Knowledge Management & Future of Work (i.e. Semantic search across systems, agentic task support, report automation)
  
  • Ramblr is a vision-enabled, multimodal co-pilot. Its agentic platform ingests real-time video, CAD, and text documentation to build scene graphs and contextualize physical processes. Then, it can make proactive suggestions, such as triggering security workflows when it "sees" a deviation, or reactive prompts, such as asking an operator to confirm an action, in context-rich environments.

‍

Investment thesis:

Future margin gains will not come from LLM ownership, but rather from companies that operationalize AI through the following five core capabilities:

Bottlenecks & Open Questions

• ‍Fragmented Data infrastructure: Only a fraction of manufacturers have robust data pipelines, unified namespaces, or scalable digital twins in place—gaps that delay agentic deployments. (McKinsey, 2024; McKinsey, 2025). Only 16% of companies have achieved their AI goals, largely due to lack of data governance, poor quality data, and fragmented systems (BCG, 2023). Before industrial agents can proliferate, we must break down data silos in legacy systems (e.g., SCADA), unify and normalize multiple data types (e.g., sensor, time series, logs), expand agent capabilities beyond basic automation, and address the fact that traditional IoT platforms have not solved data infrastructure problems, even as companies like Litmus and Hive NQ demonstrate early success.‍
• Integration and Middleware Gaps: Legacy systems and organizational silos continue to hinder end-to-end agent loops. Legacy systems, lack of interoperability, and organizational silos (e.g., unclear IIoT ownership) delay adoption (HiveMQ, 2025).‍
• Trust and regulation: As agents make higher-stakes decisions, frameworks for governance, explainability, and safe failover are still in their early stages.‍
• Agent Lifecycle Management (AgentOps): As agents transition from the experimental phase to the production phase in high-reliability industrial environments, we anticipate the emergence of a new operational layer: AgentOps.
  ‍
• This includes systems and practices for managing the full lifecycle of deployed agents, covering:
  
  • Observability and monitoring: Real-time tracking of agent actions, outputs, and anomalies.‍
  • Versioning and Rollbacks: Safe deployment, update management, and rollback functionality.
  • ‍Retraining and feedback: Incorporating new data or operator feedback to improve performance.‍
  • Governance and compliance: Logging, explainability, access controls, and auditability for regulated settings.

A robust AgentOps infrastructure is essential for scaling agent-based systems safely and sustainably. This is particularly important as agents begin making or supporting real-time operational decisions in manufacturing, energy, and logistics.

‍

Call to Action: We believe that industrial AI agents represent a paradigm shift: closing the loop between data abundance and decision scarcity, delivering autonomous optimization, and driving decarbonization in the hardest-to-abate sectors. If you’re building the next generation of agentic solutions for manufacturing, logistics, energy, or any complex industrial domain, we’d love to hear from you. Reach out to Hitachi Ventures, AENU, or b2venture—let’s explore how we can partner to shape the next generation of Industrial.

‍

If you are a founder building in this space, we’d love to connect. Feel free to reach out to us at: galina.sagan@hitachi-ventures.com, robert@aenu.com, anna.bosch@b2venture.vc

‍