Trend Watch: Agentic AI

Or what we’ll call “The Recipe” vs. “The Chef”

You’d be hard-pressed to find an End of 2024 wrap-up article that didn’t point to Agentic AI as THE hot AI topic of 2025 (Our content was no exception). Add on some heavy-rotation Salesforce commercials with a bantering Matthew McConaughey and Woody Harrelson, and talk of AI Agents – or at least the promise of them – seems to be just about everywhere.

But what IS Agentic AI?

By definition, it’s Artificial Intelligence that can make decisions, evaluate options, and take actions without the need for continuous human intervention. This allows them to:

• Adapt dynamically to complex situations
• Execute multi-step tasks without much – if any – intervention
• Self-correct and improve through iterative learning

So does that mean AI Agents are completely autonomous, determining its own goals and stopping at nothing to achieve them? Are they sentient? Are they going to take over the world? No. Nope. And no, not quite yet, at least.

To understand what Agentic AI is, as compared to something created with langchain, LlamaIndex, or similar tools, really requires understanding what Agentic AI is NOT.

Tools like langchain are essentially workflow orchestrators for AI operations. Think of those tools creating a pipeline that:

• Connects different AI and data processing steps in a predetermined sequence
• Handles the routing of information between these steps
• Manages context and memory across interactions
• Provides structured ways to access and query data sources

Now contrast that with Agentic AI, which is more dynamic. AI Agents can:

• Determine their own sequence of actions based on the goal
• Identify when they need additional information and decide how to get it
• Revise their approach if initial attempts fail
• Make complex decisions based on multiple factors and uncertainty

To simplify it, you can think of workflow tools relying on a recipe to deliver an outcome. An AI Agent is more like a chef.

“The Recipe” vs. “The Chef”

When you’re cooking from a recipe, you have a defined set of ingredients, a series of linear steps, and, for most cooks, a predictable result.

A chef, on the other hand, is free to improvise. Change the cooking time? Swap in oil for butter? Add a little extra salt at the end to improve the flavor? The chef uses the ingredients and tools at their disposal to accomplish the goal – a finished dish.

Continuing our analogy a bit further, a chef doesn’t create a recipe out of thin air. They have a pantry, a fridge, some appliances and other tools, as well as an idea of what the desired outcome is of the kitchen endeavor. But they’re able to improvise using those components, making decisions and even changing course along the way.

And that’s pretty much how Agentic AI works (well, without having a sink of dirty dishes at the end). The agent is set up with a goal and is instructed on what tools and ingredients are at its disposal.

For example, if you wanted an agentic AI to gather information about market trends (the goal), you would need to:

• Configure access to specific market data providers
• Set up API connections to news services
• Provide authentication for premium data sources
• Define which sources are trusted for which types of data
• Establish how to validate and cross-reference information

But recipes are pretty handy! And not everyone has access to a chef! That is so very true, and is why Agentic AI certainly won’t fully replace process-oriented tools any time soon. There are costs and often steep learning curves associated with deploying agents. And for many situations, the recipe will work out for users just fine.

Will AI Agents Take Over the World?

There are few, if any, areas of AI that aren’t fraught with potential ethical landmines. Permissions to use source data, biases within the source data, transparency with user that you’re using Ai, AI being used for nefarious tasks, the environmental impact of larger and larger models and the electricity needed to power them. And on and on and on. Agentic AI certainly isn’t without its potentials potholes.

But there are some good reasons to breathe a bit easier.

1. AI isn’t “conscious.” It’s not self-aware, and it doesn’t have motivations outside of what it is programmed to do. If an AI system goes out of control, it didn’t do that on its own. It was programmed to.
2. It’s not truly creative. Although it might certainly seem like it at times, language models are only iterating on existing patterns. And it doesn’t understand the meaning of its creations.
3. It’s not ethically autonomous. It doesn’t independently decide that it’s going to act nefariously. Its behavior and the ethics of them are determined by its programming.
4. It’s likely that laws, guidelines, regulations, etc. will be put in place by various government entities to put some guardrails on the technology.
5. While AI’s energy demands are growing, the industry is making significant strides through more efficient hardware, optimized models, and increasing use of renewable energy in data centers.

While that doesn’t mean that AI Agents can’t be used toward malicious ends, that’s on their programmers. The agents themselves aren’t able to “turn evil.” And while that might be little consolation, especially given the state of the world today, individuals and organizations who adhere to a strong standard of ethics will be able to create safe and truly helpful automations.

The Near Term

Not every use case will need agents – in fact, it’s overkill for a lot of processes that current tooling can handle well. But if we had to predict some near term uses that take the best advantage of what agents could offer, they’d be:

• Patient Care Coordinator: Think of it as a super-organized health assistant that keeps track of your medical journey across different doctors, flags potential medication conflicts, and makes sure nothing falls through the cracks – with real doctors still making all the important calls.
• Hospitality Experience Manager: A digital concierge that knows what you like before you ask, spots potential hiccups in your stay, and helps staff deliver personalized service without missing a beat.
• Classroom Support Agent: A behind-the-scenes assistant that helps teachers spot learning gaps, suggests targeted help for struggling students, and handles the paperwork – freeing up teachers to do what they do best: teach.

Final Thoughts

The attention on agents isn’t going away. AI took center stage in a good number of Super Bowl ads. And OpenAI just launched a research preview of “Operator” – their first real foray into general purpose AI agents. As with any new step in this AI journey, ethics must be at the forefront. But with the right guardrails and execution, AI agents might just popping up before you know it.