facebook

By Jag Gattu, Founder & CEO at UptimeAI

This post originally appeared on LinkedIn.

 

Most industrial software products now come with some flavor of ‘agents’ attached. The word has been applied to so many different products in the last two years that it no longer tells a buyer much about what they are getting. I dove into the overuse and misconceptions created by a world where everything is ‘agentic’ in my last post.  

The biggest difference I’ve seen between what we are building at UptimeAI and many of the agent claims on the market is the type of work the agent is fit for: Expert work versus assistant work. In the recently released 2026 ARC Advisory Group MarketMap: Industrial Copilots for Plant Operations, UptimeAI’s Reasoning Agents emerged as market leaders. And when I look at the full report and the definitions used to guide the rankings, it’s clear that our focus on expert agents played a large part in this.  

What it means to do expert work 

Expert work is iterative. It means contextualizing unstructured data and lived operating experience, applying reasoning and stating the assumptions behind it, and arriving at a data-backed decision you can act on with confidence. It means choosing among many possible paths, deciding how far to go down each one, and resolving the conflicts you find along the way. It applies assumptions, judgement, and reasoning. 

Optimize the maintenance plan across these 500 pieces of equipment. Find the root cause of an issue that spans process, mechanical, instrumentation. Build a HAZOP draft that captures all nodes and all possible failure modes and consequences, and continually update it as plant conditions change. 

The complex nature of expert work necessitates explainability. Ranked hypotheses, the causal chain and the evidence behind each step, and an interface to question it all were capabilities UptimeAI was praised for in the ARC report. Assistant agents speed up information retrieval and routine analysis. Expert agents take on complex, judgement-heavy decisions.  

How ARC defined the category 

The ARC definition of an Industrial Copilot is an adaptive, context-grounded, human-governed operational interface that synthesizes real-time telemetry, engineering context and business state into actionable, deterministic execution paths at the point of work, and that strictly requires human validation before closing any physical loop. 

It’s a demanding definition asking for grounding, determinism, and explainability—key indicators of expert work rather than information retrieval. 

How UptimeAI ranked 

Suppliers were scored across six criteria in two groups: Solution Capabilities, covering technical capabilities, functional capabilities and execution and support services; and Strategic Vision, covering product vision and strategy, corporate strategy and customer base. Our two strongest results tell a consistent story. 

  1. A top-scoring supplier for Product Vision & Strategy. This criterion covers long-term vision, innovation roadmap, and overall strategic direction. Of the six, that is the one I would pay closest attention to, because it is a judgement about direction in addition to the product as it stands today. 
  2. A top quartile supplier for Functional Capabilities. This covers the depth and breadth of what the product does, how configurable it is, and how well it supports the processes specific to an industry. This score represents UptimeAI’s ability to deliver solutions specific to the needs of our target markets today.  

Unpacking the scores 

Three elements underscored in the report differentiate the expert work UptimeAI is doing from a typical AI assistant. 

  1. Reasoning grounded in physics, not just statistics. UptimeAI’s reasoning engine draws on thermodynamics, mass and energy balances, and 200+ years of process SME experience, in contrast to generic statistical models. That grounding is key to providing a conclusion that a senior engineer is willing to stand behind. 
  2. Diagnostics that can be audited. Root cause findings carry explicit citations back to P&IDs and FMEAs, which the report credits with cutting investigation cycles by more than 10x. If an agent cannot show its evidence, there is no sound basis for acting on what it says. 
  3. Specialized agents, orchestrated. Rather than one general-purpose assistant, separate reasoning agents handle root cause analysis, maintenance optimization, HAZOP and process optimization, working across telemetry, P&IDs, operator logs and work orders. Different decisions draw on different expertise, and coordinating between them is the key to solving high-impact, cross-functional business problems. 

Where this goes next 

The report’s recommendation for UptimeAI is to keep investing in physics-based reasoning while broadening beyond diagnostics into remediation planning and execution support. It’s a direction that we are already heading, and it is the logical next step for expert agents: from working out what is wrong and what should be done, toward helping get it done, with human approval remaining at every point of physical consequence. 

If you are evaluating agents for your own plants, the question I would start with is: what kind of work is this agent going to take on, and who is doing it currently? Not all agents are created equal, and having an answer to that question can avoid the disappointment that comes from buying one type of agent but expecting another. Those who watch this market closely will see the gap between organizations that focus on productivity and those that focus on profitability through expertise continue to widen over time.  

 

For more analyst analysis, read the full ARC Advisory Group MarketMap for Industrial Copilots – Plant Operations, 2026. 

 

Stop monitoring.
Start deciding.

Discover how UptimeAI reasoning agents turn expertise into real-time, scalable advantage.

Contact Us