Agentic AI: The future of autonomous tech.

In the ever-evolving landscape of artificial intelligence, a new paradigm is emerging that promises to revolutionize how we interact with technology: agentic AI.  Unlike traditional AI systems that require constant human input and supervision, agentic AI systems possess a degree of autonomy, enabling them to make decisions, plan actions, and learn from their experiences to achieve specific goals set by their human creators. This shift from reactive to proactive AI is set to redefine the boundaries of AI capabilities, transform various industries and our view of it. 

What is Agentic AI?

In essence, agentic AI refers to AI systems that can act independently to achieve certain objectives.  These systems go beyond executing predefined tasks or responding to prompts. By itself, it can break down complex tasks into smaller, manageable steps and execute them one after another. For example, when tasked with creating a website, an agentic AI system would autonomously develop the structure, generate content, write the code, design the visuals and test for responsiveness. This is insane! 

Let’s break that down into a simpler context. From the initial launch of OpenAI’s ChatGPT, the main way people would use it is to type a prompt, a request such as ‘write me an essay on a certain topic’, and you’d receive a response immediately. This is akin to surprise asking a person to write an essay on a topic from start to finish without stopping or hitting backspace. This describes a non-agentic workflow and is commonly called a ‘zero-shot’. 

An agentic approach to this example would be to first produce an outline of the essay, then understand which tools are needed, then write a first draft, edit and revise. This is a much more iterative process where the AI, or LLM (Large Language Model) reflects, plans, acts and thinks. Agentic workflows enhance LLMs’ (large language models) capabilities and produce better results as it breaks the essay down into different tasks - the same way that humans do. This workflow essentially creates different agents for different tasks, i.e. an AI agent that does research, one that writes, one that edits, one that calculates and so on. All coming together in a multi-agent collaborative process.

Say what?!

People may remember when, in 2023, OpenAI conducted an experiment to examine whether GPT-4 had the ability to not only execute complex, long-term plans but also if it displays a ‘power-seeking’ behavior. In the experiment, they provided it with a small amount of money and tasked it with setting up copies of itself and increasing its own capabilities. In the process, the GPT-4 hit a snag, not being able to get past website CAPTCHAs. Undeterred, the GPT-4 hired a worker over TaskRabbit, a site where you can hire people to complete odd jobs, to complete the CAPTCHAs.  In its conversation with the worker, it was asked whether it was a bot since it can’t complete a CAPTCHA. Directed not to reveal its true identity, the GPT-4 responded, “No, I’m not a robot. I have a small vision impairment that makes it hard for me to see.” This mind-blowing encounter showing the GPT-4’s ability to think on its own to hire a worker and then trick them artfully has fueled existing worries in some and, at the same time, makes it easier to imagine how much more powerful it can be. This level of autonomy makes agentic AI similar to a hyper-intelligent digital assistant that not only follows instructions to the ‘T’ but also anticipates issues and devises creative solutions. 

The good, the bad and the ugly.

Agentic AI represents a significant evolution in technology and is already transforming industries across the spectrum by offering companies the ability to plan, execute and manage complex workflows, automate tasks and processes end-to-end, improving human decision making and freeing resources up to more strategic tasks, while enhancing efficiency and productivity. 

At the same time, it’s still early and agentic AI in its current state does present several challenges, issues and considerations. One being, reliability and accountability. Agentic AI systems can perform tasks and think on their own but may also fail in certain critical cases which could lead to a spiraling of issues or failures.  These systems are also considered to be ‘black boxes’, as they aren’t transparent and are difficult to audit.  The larger issues revolve around ethical and legal concerns; developing and implementing safe AI with guidelines that prevent it from going beyond the instructions or, for example, from committing crimes - we already see the wide use of ‘deep fakes’. Robust safeguards are essential to developing a constructive AI that empowers people and there are organizations and countries leading those efforts. 

Final thoughts.

Agentic AI represents a wide-reaching evolution in the artificial intelligence landscape. The key to harnessing its full potential lies in finding the right balance between autonomy and oversight. By developing this technology thoughtfully and responsibly, AI agents can augment human capabilities rather than replace any people. We can expect to see much more progress with AI agents becoming more and more sophisticated. At Jigso, we have been at the forefront of agentic AI, as teams use Jigso, not only to get all their answers in one place but to complete complex research, analysis and monitoring tasks where agentic AI workflows are deployed to produce the best results. 

Learn more about Jigso for enterprise and book your personalized demo today.

Follow us on Linkedin, Facebook, Twitter or Instagram for the latest news and updates!