ISHN consulting editor Dave Johnson recently interviewed by email R. Mukund, CEO of Benchmark Gensuite, a company offering a variety of AI tools, about the application of AI Agents in the EHS field.

ISHN: AI agents that can work independently and perform defined EHS tasks, make decisions, and continuously learn will not replace "human in the loop oversight" but it does seem agents will reduce the number of EHS pros needed. Other EHS AI experts I've talked to agree with this scenario. How do you see it?

R. Mukund: Benchmark Gensuite sees AI Agents as working with people — not replacing them. This belief is why we have embedded AI solutions across our platform as part of existing workflows — not to replace what people do but enhance it. Agents can build capability and develop competency, so employees are able to complete work more thoroughly and consistently even when the work is not their principal competency. Genny AI Agents serve as a digital co-worker joining the EHS team to help boost productivity, provide consistency in methodology, and knowledge democratization effectively expanding the collective competency of the entire organization.

How does agentic intelligence produce a ROI?

Agentic intelligence produces ROI in many ways. Automating manual tactical work, augmenting decision making and processes, monitoring various records and activity for trend flagging and intelligent insights all enhance employee productivity and efficiency. Agents can help employees shift bandwidth from manual tasks to activities more useful to the enterprise such as operationalizing compliance programs.

For example, instead of employees focusing on maintaining compliance and avoiding negative business disruptions from being in non-compliance, their time can be better spent studying how to analyze an air permit. AI Agents ensure standardized approaches and methodologies are applied consistently across projects, reducing variability and improving overall quality of work.

For agentic intelligence to transform EHS, how much "deep domain knowledge" must be curated? Who does this work? Are agents only as effective as the domain knowledge they tap into?

Deep domain knowledge can be curated intentionally — but knowledgebases are the key to directing knowledge, they are not the limit of knowledge. Agent access to LLMs (Large Language Models) provides all the information an agent needs to find. A good knowledgebase directs and applies the information an agent can access through an LLM. Multiple knowledgebases can be connected to ensure business policy is applied, any previous related material is referenced or used as an example. A controlled knowledgebase paired with intelligently designed prompts ensures that all the information fetched by an AI agent is properly sorted, prioritized, and organized into outputs that are helpful to the user.

What guardrails are in place to prevent independently acting agents from making wrong decisions regarding task execution?

The human-touch is crucial to preventing AI from making wrong decisions or providing false information, but there are many automatic guardrails that can be put in place. Ensuring agents are connected to controlled knowledgebases with intelligently designed prompts provides direction of information and data. When an agent provides information, ensuring direct links to the outside source provides quick access to verification. Finally, ensuring messaging to users is clear on what verification is needed with a sign-off on any agent generated content or decision-making.