As promised, Agentic AI has arrived. REWIND: What even is Agentic AI? A computer is just a complicated calculator. Automation can be programmed to do a thing under certain conditions. AI can execute tasks you give it, but not choose what tasks to do or how to do them. Agentic AI works like this: you give it a goal, and it decides the path to get there.
Let’s say you want leads generated, outreach crafted, and follow-ups scheduled. An agentic AI doesn’t wait for you to prompt it to find the leads, craft those messages and add it to your calendar. It builds its own checklist, checks it twice, and does the thing. This is the next phase of autonomy in machines, and yes, it should make you both excited and a little nervous.

What makes this different from the ChatGPT you’re used to is initiative. Traditional GPT is like an assistant who sits patiently, waiting to be asked clever questions. Agentic AI gets up and makes coffee, schedules your meetings, replies to emails you forgot about, and even flags problems before you notice them. (Okay, not really the coffee.) It’s proactive and under the best conditions, smart and capable. Discernment becomes a problem. “Wait, I was going to respond to that email, and that is NOT what I would have said.”

Imagine an AI that knows your candidates, your clients, and your market, and which thinks about how to get them in front of each other without you dragging it there. Instead of running Boolean strings, you just say, “Find me 99 more like Carla, and email them something crafted to match their personality.” Boom. Done. You didn’t even have to click.

Currently, tools like ChatGPT Operator, Zapier AI agents, and Cognosys are available as Agentic options. The challenge here is none of these (in their current capabilities) can autonomously and reliably browse professional networking sites, interpret profiles in the nuanced way a human recruiter does, and build a highly targeted list of 100 candidates.

Discernment becomes an issue. Who is judging these candidates? As does repetition. You know if you ask Chatgpt to do a task 100 times, the first 10 are pretty darn good, after that though… There’s also restrictions, like Chatgpt can’t get on LinkedIn. Also, the number of steps involved require a level of thought which AI is not at yet. That’s because of Hierarchical Planning. If we think about orchestrating a meeting between a candidate and a client, that “decomposes” (in AI speak) into lots of tasks: Find people, engage repeatedly, lose some, interview, lose more, schedule, etc….

Often, the decomposed tasks have dependencies and a natural hierarchy. An Artificial General Intelligence would need to plan the execution of these sub-tasks in a logical order. We’re not quite there yet. Next week, I’ll share my thoughts on what ChatGPT CAN do for a recruiter in its current level of capability.