Most of the talk about AI in safety sticks to the same few things. Tools that predict risk. Cameras that spot hazards on site. Chatbots that help you write a risk assessment faster. All useful. All worth knowing about.
But the thing I think will really change safety work day to day gets far less attention. It's called agentic AI. It's worth taking a minute on what that means, because the name makes it sound harder than it is.
Here's the simple version. Most AI tools today wait for you. You ask, they answer. You type in a near miss, they help you file it. You upload an inspection, they point out anything odd. But you, the person, still have to decide what to do next and then go and do it.
Agentic AI is different. It doesn't just point something out and stop. It takes the next step for you. It spots a job that needs doing, sends it to the right person, sets a date, checks whether it got done, and chases it if it didn't. Log an accident and it looks for similar ones, finds the rule that applies, and drafts the report you're required to send, all before you've asked.
It's the difference between a tool that tells you what needs doing and one that starts doing it.
This matters in safety because so much of what slows the job down isn't the thinking, it's the admin around it. Chasing people to close jobs off. Typing the same thing into three different reports. Remembering which authority you have to tell, and by when. That's not where skilled safety people should be spending their day.
We're still early with this. Getting it right in safety takes real care, because the stakes are high and you don't want a machine making big calls without a person checking first. But the direction is clear.
The safety tools that get this right won't feel like software. They'll feel more like a reliable team member who never forgets anything and never lets a job slip.