Honeywell’s AI tools ease the move from automation to autonomy

AI can sometimes seem too big to be useful locally. However, just like any huge problem or project, it can be broken down into approachable, solvable chucks that make success not only likely but inevitable.

For AI, these doable chunks are known as agents, and they can help process control engineers and other professionals overcome one of their latest and greatest challenges—digitalization, migrating from hardware to software, and shifting from traditional automation to more autonomous operations.

Luckily, knowing about traditional automation can also help users understand and apply AI and autonomy, too.  Ankit Singh, senior data and AI architecture director at Honeywell, reports his division’s focus on the life sciences places it at an intersection, where it can benefit from the company’s more than 100 years of experience in industrial, building and process automation.

AI tools already at work

“This perspective allows us to look at optimization on all levels, including biological manufacturing, building and facility environments, and regulated process control,” said Singh, who presented “Intelligence layer: leading in the age of autonomy” on the third day of Honeywell User Group Americas 2026 this week in Phoenix. “This intersection is also where AI is facing the most challenges, which is why Honeywell is seeking to lead as autonomous operations emerge.”

In fact, Honeywell’s AI portfolio already boasts an array of innovative agents and other tools that users can apply to monitor, manage and optimize their individual processes, including:

• TrackWise AI Compass for intelligent navigation. Its AI-guided trending topics and anomaly detection assistant reduce cycle times and training burdens.
• Corrective and Preventive Acton Advisor (CAPA) analyzes deviation patterns, suggests root causes, and drafts action plans.
• Records Processing Agent for documenting intelligence extracts, classifies and routes quality records into validated Salesforce objects.
• Complaints Intake Agent manages and triages complaints, assigns severity, links products, and initiates investigations.
• Quality Risk Management for proactive risk management-- emerging risk to the surface across batches, sites and suppliers.
• Root Causes Analysis (RCA) cross-references complaints, CAPA records, and batch data to accelerate investigations.
• Document Management System is a controlled-document AI tool that manages standard operating procedure (SOP) lifecycles, flags expiring documents, and enforces version control.

These solutions include all the essential functions users need to establish their own AI architecture and begin to run more autonomously. Once this is done, Singh added they can also use Honeywell’s Autonomous Agent Design program to build their own autonomous devices. Its four parts include:

• Regulatory Compliance by Design, which audit-trails every action to Salesforce Shield; uses good-practice (GxP)-validated guardrails in agent logic; and doesn’t allow actions outside defined regulatory scope.
• Consistent Outcome Specificity, where agents own a precise, measurable boundary; every exception has a defined escalation path; and AI is scoped, testable and predictable.
• Trusted Contextual Reasoning with agents reasoning over structured, quality-management systems (QMS) before acting; batch records, deviation history and regulatory thresholds are employed; and context-aware recommendations are used instead of pattern-matching.
• Human in the Loop with critical decisions requiring qualified approval by a person; mandatory checkpoints are required at defined risk thresholds; and agents provide advice, but humans decide on regulated actions.

“Honeywell has been focusing on accessing data for 100 years, so now our autonomous platform can connect to any type of data, and give it a place land in. This lake is formed by a partnership between Honeywell Data Fabric and Salesforce Data 360,” explained Singh. “This unified, multisystem, integrated layer let us contextualize data using both structural ontology and cognitive ontology and check the connections and dynamics between data sources and their locations. These are the tools needed to capture and standardize information, so AI and large-language models (LLM) can understand their semantic relationships and begin to help users and their processes benefit from them."

Singh reported that Honeywell and its users define ontology using well-known tools, such as Ontology Web Language (OWL) and Resource Description Framework (RDF), which are the foundational World Wide Web Consortium (W3C) standards used to construct knowledge graphs and power semantic web functions. Once incoming process data is standardized, agents can start to run on the Model Context Protocol (MCP) layer, and report to Honeywell Forge Edge & Intelligence, which includes digital twins, small language models (SLMs) on the edge, and cloud-based LLM and SLM AI applications.

“We just used these solutions to connect 100,000 assets for Nvidia, and helped them reduce nuisance alarms by 90%,” said Singh. “This is where Experion Cognition participates as part of Honeywell’s big library of agents and other tools for assisting large computing jobs. This is also where Salesforce 360 and Honeywell Data Fabric complement each other by linking data layers to complete those ontology tasks, which show these teaching agents what they need to know.