For decades, we have designed health and safety teams around people.
We recruit advisors, managers and specialists. We define reporting lines, write position descriptions and allocate responsibilities. We build systems around the assumption that work is undertaken by people, supported by software.
Artificial intelligence changes that assumption.
AI systems are increasingly capable of undertaking parts of the cognitive work that currently makes up a large percentage of our professional roles: analyzing information, monitoring patterns, retrieving organizational knowledge, preparing first drafts, checking data, developing prototypes, answering questions and coordinating activities across multiple systems. AI agents take this further by being given a goal and completing a sequence of tasks rather than simply responding to an individual prompt.
For health and safety leaders, the interesting question is therefore no longer simply, “What can we use AI for?”
A much better question is: how should we design a health and safety function when some of its capability can be provided by machines?
Rethinking the safety org chart
Imagine looking at your current health and safety organization chart differently.
Instead of seeing positions, look at the work.
A safety advisor might investigate incidents, prepare reports, analyze trends, facilitate risk assessments, answer worker questions, update procedures, prepare meeting papers and spend time in the field understanding how work is being performed.
Those activities do not have the same requirements.
Some depend heavily on relationships, contextual judgment and trust. Others involve retrieving, comparing, synthesizing or transforming information. Increasingly, those latter activities can be supported by AI.
In many organizations around the world, AI agents are monitoring incident and assurance data for emerging patterns, maintaining organizational knowledge, answering routine questions about requirements, and preparing first-pass investigation timelines from evidence collected by a human investigator. AI capabilities embedded within existing systems classify information, check document consistency or prepare draft reports for human review.
This augmented capability means deliberately deciding which capabilities should be human, which should be machine-supported, and where the two should work together.
That is fundamentally a work-design problem.
Start with the work, not the technology
There is a predictable danger here.
Organizations can see a capable technology and immediately start looking for things to automate.
Safety leaders should resist that sequence.
Before introducing AI, we need to understand the work it is intended to support. That means spending time with the people doing it, understanding the decisions they make, the information they use, the exceptions they encounter and the workarounds they have developed.
Human-centered design techniques such as contextual inquiry, journey mapping and personas become particularly valuable here. They allow us to see the gap between how a process appears in a management system and how it is experienced by the supervisor, contractor or worker trying to use it at 2 a.m. on a maintenance shutdown.
Automating a poorly designed process simply allows the organization to perform poorly designed work more efficiently.
Understand the work first. Reimagine it in a world where both humans and machines work together to achieve business objectives. Redesign the work. Then decide where technology belongs.
Build the foundations before the agents
AI also forces safety leaders to confront a problem that has existed for years: the quality and architecture of our information.
Many organizations have safety knowledge scattered across procedures, spreadsheets, SharePoint sites, incident systems, dashboards, emails and individual experience.
People have become remarkably good at navigating that fragmentation.
Machines are much less forgiving.
If we expect AI to analyze risk, support decisions or provide workers with reliable information, we need to know where authoritative information comes from, who owns it, how it is structured, how current it is and what happens when sources conflict.
The AI strategy therefore cannot be separated from the data strategy.
Before asking what agent to build, ask what information that agent would need to do its job well. Exactly like you should ask a human.
Create space for more human work
The prize here is the opportunity to reconsider what we want safety professionals to spend their time doing to create the most impact.
If machines can reduce the administrative burden associated with searching, synthesizing, documenting and reporting, people can spend more time understanding work, facilitating difficult conversations, testing assumptions, coaching leaders and workers, designing better controls and making sense of complex situations.
These are activities where context matters enormously.
A machine might identify that a particular category of incident is increasing. A human still needs to understand why.
A machine might generate a technically correct procedure. A human needs to determine whether the intended users can realistically follow it while wearing gloves, working at height or responding to an emerging problem.
A machine can provide information. People create trust.
The future safety professional may therefore become less of a document owner and more of a designer and orchestrator of a socio-technical system: deciding how people, technology, information and organizational processes should work together. An architect of better work.
Leadership in work design
This has implications well beyond the safety function.
Safety, operational and HR leaders will increasingly need to think about capability rather than simply headcount. Position descriptions may need to distinguish between work that people perform themselves and work they supervise, orchestrate, validate or undertake with AI. New competencies will emerge around AI literacy, verification, data stewardship and human-machine collaboration.
Governance matters too. Organizations will need clear boundaries around accountability, ethics, privacy, transparency and the decisions that should remain human.
But governance should not become an excuse for paralysis.
The organizations that navigate this transition well will be those willing to experiment responsibly, involve workers in design and learn where AI adds value before scaling it.
We have an opportunity to design health and safety functions very differently from those we inherited.
The objective is to remove work that prevents humans from doing the things we want and need humans to do.
The organizations that understand that distinction will be far better positioned for the age of AI.
About the Author
Cam Stevens is Founder and CEO of PKG Safety Innovation™, a health and safety innovation consultancy working with high-risk industries on AI adoption, digital transformation, and critical risk. A Chartered Fellow of the Australian Institute of Health and Safety with a background spanning physiotherapy, human factors, and AI ethics, Cam has led more than 200 technology deployments and educated over 30,000 people globally through the Safety Innovation Academy™.
