The Rise of Autonomous AI Agents
Nope, this isn’t about automating your Tesla, this is about a gazillion little specialized artificial intelligence agents running your life.
![Vector cartoon stick figure drawing conceptual illustration of group of government secret agents walking or marching in sunglasses. Vector cartoon stick figure drawing conceptual illustration of group of government secret agents walking or marching in sunglasses.](https://eu-images.contentstack.com/v3/assets/blt69509c9116440be8/blt02b5a0ad3dcbe796/65527f200c08f0040ad9fc0c/00AI-Zdenek_Sasek-alamy.jpg?width=700&auto=webp&quality=80&disable=upscale)
Zdenek Sasek via Alamy Stock
For the most part, AI agents are not spies -- though it would be foolish to expect them to act without gathering and using data. Let’s take a look at what these agents actually do.
“An autonomous agent is basically anything that can complete a task in response to things from its environment. And that means you don't necessarily have to tell it what to do. It just kind of knows what it needs to do,” says Justin E. Lane, co-founder and CEO of CulturePulse, who “has pioneered weird ways to study culture through AI,” lately via his company ’s multi-agent AI-created digital twinning methods.
Lane says a motion-activated porch light is an example of a very rudimentary autonomous AI agent while Rover the Mars-mobile is a good example of more complex AI agents. Complex use cases like Rover or a Roomba vacuum usually need a system or a team of many AI agents.
Most autonomous agents work in the real world, but some work in simulated environments, usually to perform various types of experiments or to make predictions on real-world events and outcomes. This type of AI agent is the most mysterious to the uninitiated.
Now on to the AI agents causing a stir today.
This class of AI agents can take a project and run with it all by itself, even to writing its own prompts, selecting the data it needs, and mapping a strategy.
One example is Godmode, which I personally used to find the best smart TV at the best price. I simply told Godmode to go find it for me and it went and searched all the websites for a TV to my specs at the best price. I also used it to find a cocktail dress that fit my size and unique style at the best price that could also be delivered the same day. Voila! Easy-peasey shopping!
But what happens to the SEO industry, online advertising, and retail websites when shoppers no longer shop or search online, and only AI agents do? Undoubtedly there are lots of changes ahead -- but maybe not all of them will happen immediately.
“Current popular autonomous agents like AgentGPT, ReAct and BabyAGI are fairly immature and buggy. These agents carry out steps based on the objective of the prompt, then the LLM takes action on behalf of the user,” says Muddu Sudhakar, CEO and co-founder of Aisera.
“Since they are based on general-purpose, foundational LLMs, this makes them more vulnerable to hallucinations and bias. These systems are imperfect as they can go into infinite loops and have a lack of enterprise-grade security,” Sudhakar adds.
But give them a hot AI minute and these agents will be much better at doing the tasks you give them. Meanwhile, you have to authorize each step one of these agents comes up with as a control measure. So, no worries, they won’t go too crazy. Maybe.
Companies can use autonomous AI agents for a lot of practical business purposes. The most obvious use case is to supercharge customer service.
“Enterprise-grade action bots are the next big thing in AI. These types of autonomous agents go beyond conversational chatbot capabilities to actively executing tasks and streamlining workflows. Rather than just answering questions, these sophisticated systems proactively engage with users and produce tangible actions to enhance customer experiences,” explains Aisera’s Sudhakar.
Action bots are based on micro LLMs like OpenAI’s recently released micro GPTs. Action bots are customized to the enterprise and its customers or users. These can be customized on a deep level to be industry-specific, domain-specific, and customer-specific.
“As a result, these fine-tuned agents have significantly improved accuracy and decreased hallucinations and bias,” Sudhakar says. “Action bots also have much tighter integrations with enterprise platforms like CRMs, ERPs, and RPAs, which ensures that data flows smoothly and tasks are executed efficiently. Finally, these agents have stronger security with SOC 2 compliance.”
AI agents working in simulated environments operate on their own beliefs. Wild, yes?!
“In more advanced logic systems, you can have certain probabilities and those probabilities are based on the beliefs of the agent. The beliefs of the agent can change in relation to the other agents in its environment and the other experiences that one agent has had that another agent hasn’t,” says CulturePulse’s Lane.
“In this way, what you’re looking at is an autonomous agent in a simulated society,” he adds.
The most common use cases for AI agents in simulated environments are to safely conduct many types of experiments, test the responses of AI agents, or to predict real-world outcomes in various scenarios. But the scene is almost always a little surreal.
“This is kind of like the 'Matrix,' in that you have people acting and interacting in a simulated world. Except of course, here, there are not actual humans powering these agents, but artificially intelligent agents, with different behaviors and psychologically realistic properties,” Lane says.
Essentially you can categorize autonomous AI agents into two types: basic and advanced. Some or many of either type, or a mix of both, can be used to comprise a system. But a system is not always needed. It depends on the capability of the agent and the complexity of the task or project you wish it to perform.
In any case, autonomous AI agents can exist in different environments including hybrid -- blended or mixed -- environments.
“The more advanced agents are almost like human assistants. They can function in effective environments, or simulated environments, or real-world environments and simulated environments,” says CulturePulse’s Lane.
“Agents in simulated environments help create artificial societies so that people can better understand conflict, war, and policies and so on. That creates a system that can be super, super useful,” he says.
One would think that autonomous AI agents are the best that AI agents can be -- or at least the final form of this particular branch of evolution. One would be wrong.
Consider the big picture here and the many different types of agents it will bear.
“What autonomous agents represent is the most ambitious attempt at combining generative AI models with engines that allow them to take action. Overall, this is an extremely powerful combination, and we can expect a myriad of new applications that embed both GenAI models and automated actions, just for more defined tasks, with clear steps and in ways that can be rigorously tested,” says Kjell Carlsson, Ph.D., head of data science strategy and evangelism at Domino Data Lab.
“In the very near future GenAI agents will abound, not just autonomous or semi-autonomous agents,” Carlsson says.
As AI agents take over more tasks and projects for their human overlords will that shift kill apps as we know them?
“Generative AI models can and should be used to support and amplify apps, not replace them,” says Sunil Senan, Infosys SVP and global head of data, analytics, and AI. “AI-powered apps can provide users with the ability to process information faster, connect unconnected data to build richer context and hence make quick, data-driven decisions, which improves their in-app experience.”
But that requires broad adoption of AI agents, not just a showy claim of “AI inside.”
“Simply adding chatbots to apps and services because competitors are doing so won’t cut it anymore, and organizations are realizing their AI-backed apps need to add more value to users,” Senan adds.
In other words, your app can’t just tell, it must show some real action. That means the faster a company switches to actionbots and other proactive AI agents, the better they will fare in this AI-fueled world.
Given the abundance of silly conspiracy theorists and the rise of the very real concerns of privacy and fairness advocates, people want to know exactly what autonomous AI agents are doing and why.
“The rise of autonomous, AI-powered agents has sparked fear -- stemming from a lack of understanding -- around misinformation, security, and sustainability,” says Infosys’ Senan. “AI agents need to have a clear purpose for what problems they are solving, coupled with a level of explainability and transparency, which is harvested with responsible AI -- the design and use of AI technologies that are aligned with human values, ethical principles, and legal requirements.”
Remember that the more things change, the more they stay the same. Take note of lessons learned in the past and see where they might apply to the work of AI agents, too.
“As we see more and more companies -- like a financial services company with its new GenAI assistant for financial advisors -- release AI-powered agents, it’s imperative that those entering this space apply learnings from the challenges of their predecessors,” Senan adds.
Given the abundance of silly conspiracy theorists and the rise of the very real concerns of privacy and fairness advocates, people want to know exactly what autonomous AI agents are doing and why.
“The rise of autonomous, AI-powered agents has sparked fear -- stemming from a lack of understanding -- around misinformation, security, and sustainability,” says Infosys’ Senan. “AI agents need to have a clear purpose for what problems they are solving, coupled with a level of explainability and transparency, which is harvested with responsible AI -- the design and use of AI technologies that are aligned with human values, ethical principles, and legal requirements.”
Remember that the more things change, the more they stay the same. Take note of lessons learned in the past and see where they might apply to the work of AI agents, too.
“As we see more and more companies -- like a financial services company with its new GenAI assistant for financial advisors -- release AI-powered agents, it’s imperative that those entering this space apply learnings from the challenges of their predecessors,” Senan adds.
During all the handwringing over the specter of a single, very big AI model becoming conscious and ruling over all of mankind, few noticed the multiple popups of tiny, specialized AI agents that can run an individuals’ life.
Here we’re talking about the often anonymous, autonomous AI agents that tend to work largely unobtrusively and unnoticed until OpenAI widened the spotlight with its recent announcement of a store full of micro GPTs.
Soon, an army of autonomous AI agents will be all up in your business, literally, and your life. That being the case, let’s take a look at what these truly are and whether they are friend, foe, or frenemy. At least now that we’re removing their invisibility shields, you’re more likely to see them coming.
About the Author(s)
You May Also Like