Back to Main
This Week in AI — 🤖 When will AI agents come to communications?
AI | Media Relations | Public Relations

This Week in AI — 🤖 When will AI agents come to communications?

April 22, 2025 | By: Dave Isaac

Each week we’ll gather headlines and tips to keep you current with how generative AI affects PR and the world at large. If you have ideas on how to improve the newsletter, let us know!

What You Should Know

 

How Close is Agentic AI for Communications?

AI agents have dominated headlines in 2025, with Google, Amazon, and OpenAI rolling out customizable agentic frameworks and a Deloitte survey finding that 26% of C‑suite leaders are exploring autonomous agents at scale. But how close are these systems to delivering value, especially for communicators?

A Carnegie Mellon simulation asked top models to handle everyday workplace tasks and found the best one completed under 25% of assignments. It misread colleague prompts, mishandled file formats, and left projects unfinished. While AI agents in some fields come up short today, that likely won’t be the case for long. Anthropic’s Chief Information Security Officer, Jason Clinton, told Axios that AI-powered virtual employees are only a year away as the company continues its work to make AI agents safer.

To date, applications for AI agents are mostly writing code and automating data analysis. Agentic AI would likely make a different impact in communications, like performing competitor research or analyzing the sentiment of media coverage. While our industry waits for a more relevant species of AI agent, there are lessons to be learned from the software engineers grappling with how agentic AI works best for them.

Communicators should pay particular attention to what kinds of tasks AI agents are assigned and which they perform well. That performance will help you pinpoint which chores can be handed off to AI and which demand the human touch. Armed with that insight, you’ll steer a balanced rollout that boosts efficiency without sacrificing creativity or empathy.

Elsewhere …

Tips and Tricks

🤔 Should o4-mini-high be your go-to model?

What’s happening: Last Wednesday, OpenAI released two new series of models: o3 and o4-mini. The new launch makes things even more complex, especially for those who don’t use AI all that often. If you’re a paying customer, there are eight models you can choose from (next Wednesday, GPT-4 will no longer be available and the list will shrink to seven).

Why it matters: It can be hard to tell the difference between the models, especially because many of them can access the Internet, accept file uploads, and generate images — all of which are core functions for communicators. However, OpenAI’s recent reasoning models have become more advanced, and their ability to show their work could be enough to make one of them your go-to model. 

Try this: When you prompt o4-mini-high, you’ll see a little drop-down option before the output that says “Thought for (insert length of time here) >.” If you click on that arrow, it will drop down and show the reasoning steps that went into generating the response. Seeing how the model arrived at its output can help trigger new ideas or explain why you might’ve got a confusing answer. There doesn’t appear to be a drastic difference in writing performance between 4o, OpenAI’s flagship model, and o4-mini-high, so the latter showing its work might make it your new favorite. While you cannot set a default model, ChatGPT appears to pre-select the model you used most recently. 

Quote of the Week

“We have theories about what kinds of capabilities these systems will have. That’s obviously what we try to build into the architectures. But at the end of the day, how it learns, what it picks up from the data, is part of the training of these systems. We don’t program that in. It learns like a human being would learn. So new capabilities or properties can emerge from that training situation.”

— Demis Hassabis, CEO of Google DeepMind, in a 60 Minutes segment on the future of AI

How Was This Newsletter?

😀 😐 🙁