Summary
- Agentic AI can act on its own, making decisions and taking actions without direct prompts.
- Big names like OpenAI and Meta are working to introduce agentic AI across platforms.
- The benefits of agentic AI include convenience and automation, but there are serious privacy concerns and a loss of personal choice.
AI chatbots have taken the world by storm, but the next big thing is agentic AI. The goal is to create AI that can do everything on your behalf, from booking appointments to buying your groceries. As convenient as it sounds, however, is agentic AI something I really want?
What Is Agentic AI?
There are a lot of terms flung around relating to artificial intelligence. One of the big buzzwords at the moment is agentic AI. Agentic may be a pretty ugly word, but it does the job of giving a good indication of what this means: AI that can act as an agent.
Agentic AI is capable of making decisions and taking actions without direct prompts. In other words, it has its own agency. This is a big difference from the AI chatbots that we’ve come to know and love, which on the whole can only do anything when given a prompt.
Agentic AI, however, would be able to do things such as checking your schedule, looking up the traffic conditions, and telling you when you need to leave to your meeting on time. All without you having to tell it to do anything.
A Lot of Big Names Are Introducing Agentic AI
It’s easy to see that this would be appealing. While AI chatbots are impressive, they’re still a long way from the humanoid robot butlers that science fiction has been promising us for so long. Agentic AI would be one step closer to making that a reality.
That’s why a lot of big names are working to bring Agentic AI to their platforms. OpenAI, for example, has added scheduled tasks to ChatGPT, which is the first time that the AI chatbot has been able to perform actions without having a direct prompt. The new Operator feature is taking things even further by giving you a way to get ChatGPT to find and book hotels or perform other tasks that can be achieved using a browser.
Meta is also working on agentic AI, with plans to get its own agentic AI models into millions of businesses. Even Amazon, which has been late to the AI party, has finally announced the release of its updated Alexa, which will be able to book reservations at restaurants on your behalf without you needing to ask it to.
The Potential Benefits of Agentic AI
There are a lot of obvious upsides to agentic AI. For the end user, the biggest benefit is convenience. Why waste hours of your time planning a vacation, looking for flights and hotels, and then booking them all, when you can just ask AI to do it all for you? If agentic AI models can work successfully, then it’s clear that many people could use them to save themselves significant time.
Another major benefit is that agentic AI can learn your habits, something that you’ve probably already experienced to some extent. For example, if you use Spotlight on your iPhone to open apps, you’ll see a list of suggested apps based on those that you most often open at that specific time of day. Agentic AI will be able to take this to the next level, learning what you like to do when, and suggesting actions or even taking them on your behalf.
Some agentic AI models can even access your emails and calendar and use the information to take action on your behalf. For example, it can see the email reminder from your dentist, check your calendar, and book you a dental appointment, all before you’ve even had a chance to look through your emails.
Many people use Alexa to control their smart home devices (although there are many more powerful options). However, agentic AI would mean that you don’t need to use voice commands or apps to control your smart home; everything would happen automatically. Walk into your living room in the evening and the lights could turn on, the blind close, and smart TV power up and go straight to the next episode of the show you’re in the middle of watching.
Related
Perplexity’s AI Shopping Agent Proves Unreliable in Early Tests
Perplexity’s new AI shopping agent takes its time when making purchases, if it can at all.
The Significant Risks Posed by Agentic AI
There’s a saying in tech that if you’re not paying for the product, then you are the product. For example, Facebook is free to use, but costs a lot of money to run. The company makes that money by targeting you with ads based on the information that you freely give to it, such as your age, gender, location, interests, relationship status, shared posts, comments, and more.
For agentic AI to work properly, it needs access to almost everything. If you want it to run your schedule, then it needs to be able to read every single one of your emails and have access to your calendar. If you’re using agentic AI to run your smart home, it may also be able to hear and see everything you’re doing through smart speakers and security cameras that you grant access to.
This is a gold mine for companies that want to offer you agentic AI services. You’re effectively giving them access to almost every aspect of your life, and they can use that information however they wish.
This isn’t the only issue, either. If your agentic AI is cloud-based, then your entire home is potentially at risk of being taken over and controlled by someone other than your AI. Your devices could be hijacked, or the feed from your home cameras intercepted.
Ultimately, the agentic AI may simply get things wrong, with potentially serious consequences. AI doesn’t even have to go rogue, like HAL in 2001: A Space Odyssey; all it would need to do would be to turn out the lights when you’re chopping onions, and you could end up getting hurt.
I Want Agentic AI For Some Tasks but Not for Everything
It’s possible that many of these issues could be mitigated. You can already run your own large language models (LLMs) at home, and so it should be possible to run open-source agentic AI locally, meaning you don’t have to share your information with anyone else or put your home at risk of cyberattacks.
However, even if agentic AI was perfectly safe, there are plenty of things that I just wouldn’t want it to do. For example, companies are touting the idea that agentic AI could book your vacation for you, finding hotels and flights and booking them on your behalf. I don’t want AI to choose where I stay on my vacation; I want to be able to choose for myself.
I don’t want AI to choose my groceries for me or book my next haircut or reserve a table at a restaurant of its choosing for Valentine’s Day. These are all things that I still want to be able to do myself. Giving AI the ability to make payments on my behalf is about as far from what I want as it can get.
That’s not to say there aren’t some benefits I would like. Being able to run agentic AI locally to help to automate my smart home sounds like the perfect fit. AI that can learn my habits and the way that I use my smart home, and make things happen by magic or even suggest automations that I hadn’t considered is exactly what I want.
Just Because You Can Doesn’t Mean You Should
The pace of AI development is incredibly fast, with new breakthroughs happening all the time. The trouble is that with so little time to stop and think, companies are developing features because they can rather than because they necessarily should.
An AI that can choose your meals for the week, order your groceries, and tell you when you need to start cooking sounds great in theory. But what happens if the AI wasn’t aware that you hate carrots, or that you really just want to have takeout that weekend? Handing over complete control to AI isn’t always going to be as good as it sounds.
That’s not to say there aren’t benefits. Having AI prepare your grocery list for you that you can tweak before ordering would definitely be a big time-saver, and that’s the way I hope agentic AI will go; suggesting what to do rather than just doing it all itself.
Agentic AI has the potential to make our lives easier, but at what cost? Ultimately, we’ll all need to consider whether we’re willing to give up our privacy and personal choice for the sake of convenience.
