Most IR teams experimenting with AI right now are using it to summarize. Run the earnings script through a model, get three key takeaways back, move on.
It's fine. It's also close to worthless. Summarizing is the cheapest thing a model does, and your own earnings script is the document you least need summarized. You wrote it.
The work that actually pays is monitoring. An agent watching the tape and the register — a block crossing, size that doesn't fit the usual pattern, a peer announcement, a shift in who's reading your site — that tells you what happened and why it matters to you specifically. A competitor misses on a supply chain issue: so what? The so what is that you don't have that exposure, and here are two points for your next investor meeting. That's the difference between reporting what happened and being a real partner to your C-suite.
The goal is narrow and worth stating plainly. When your CFO asks what moved the stock this morning, or an investor asks how you're positioned against the sector, you should already know. Not by the afternoon. Already. Being the best-informed person in the room about your own stock is what an IRO is actually paid for, and it's the thing this technology is genuinely good at protecting.
The gap isn't between teams with AI and teams without. It's between tools you have to remember to open and systems that interrupt you when it counts.
There is a real security question underneath this, and it's why a lot of IR functions have stalled. Material non-public information does not belong in a general-purpose chatbot, and anyone who tells you otherwise hasn't thought about it hard enough. But "we can't put this in ChatGPT" has quietly become "we can't use AI," and those are not the same sentence.
What's actually changing is the shape of the job. The tools have moved past automation, which answers a question you thought to ask, toward agents that do the work — build the dashboard, pull the ownership picture, draft the Q&A. The IRO stops doing all of it by hand and starts directing the things that do. That's a different skill from the one most of us were hired for, and it's the one worth building now.
The teams that get this right don't start by asking what AI can do. They start by asking which decisions they are consistently making late, and work backwards from there.