by Jacek Korneluk

I always wanted to have a personal assistant “living & residing” on my computer and friendly interacting with me. I remember seeing a movie in 90s that was showing a unique software talking to a computer owner and calling him My Master. It was amazing, and it made my imagination run and search for a similar working solution.

I wanted desperately to get the same digital mysterious friend, or even better. The companion which was totally personalised to my needs and liking.

Unfortunately, almost all early available version of chatbots were disappointing for me, having accent problems or very limited capabilities. Siri, Cortana, Google Now, Alexa, or whatever the name and version have never attracted my real attention, until now! And this is the time everything changed.

I discovered the existence of multiple LLMs and Intelligent AI Agents which can be adapted to my entire needs, then help me almost effortlessly navigate a set of tools to get things done. Well, that’s where Agents came in!

AI Agents are software programs powered by artificial intelligence algorithms 1(Sameia, 2023). They are designed to simulate human-like behaviour, understanding, and decision-making processes. These agents possess the ability to analyse vast amounts of data, learn from patterns, and adapt their responses accordingly.

You can say agents are almost like super-powered helpers that can use a variety of tools based on what you tell them. They’re smart enough to even use the output of one tool as the input for another, making them incredibly convenient, flexible, and efficient.

Despite there being several different structures and multiple types of agents 2(Terra, J. (2023), for this article I will focus on two types of agents: Action Agents and Plan-and-Execute Agents 3(Langchain, 2023).

They are working differently but both aim to solve your problems and transverse the questions, even very complex and sophisticated ones.

Action Agents make decisions step by step, using what they’ve learned from previous actions to guide their next move. They’re great for simpler tasks, like finding information or answering quick questions n(Langchain, 2023).

Plan-and-Execute Agents work a bit differently n(Langchain, 2023). When you enter input, they take a step back and plan out the entire sequence of actions needed to achieve our entire task. It’s like a job of a manager who carefully maps out every step. Once the plan is set, the agent starts executing each action, one after the other, using the output of each action as the input for the next. It’s a well-coordinated performance that ensures smooth progress toward our goals and objectives.

So, the Plan-and-Execute agents are the masters of complex tasks.   They’re perfect for tackling big projects that require careful planning and attention.

You can ask, how the hell do the Agents work? I will try to explain it in a simple language.

Firstly, you give them some input, like a question or a request.

Then, based on what you need, the agent decides which tool (other software) to use and what input to provide to that tool. It’s like picking the right tool for the right job. The agent then calls the tool and records what it learns from it, which is known as an “observation” n(Langchain, 2023).

Then, using the history of tools, inputs, and observations, the agent decides on the next step to take. This process repeats until the agent can give you a direct response. It’s like having a very smart assistant who knows exactly what to do for you!

To create an Agent, we need a few components.

We are starting with the prompt template (where you are typing your questions), which takes your input to create a clear request for the agent.

Then we need the language model (LLM), which acts as the brain of the operation. It takes our input, the prompt, and what the agent has done so far to figure out what to do next.

Lastly, we need the output parser (syntax and grammar analyser), which takes the language model’s response and makes sense of it, turning it into the next action or a final answer. It’s like a good translator that helps the agent communicate with you effectively in our human language.

Sometimes it’s smart to combine both, Action Agents with Plan-and-Execute Agents n(Langchain, 2023).   It works better cos delegates smaller tasks to specialised workers while overseeing same time the entire project.

Agents are like our personal digital assistants, helping us tackle tasks most efficiently and effectively way. Whether it’s a “simple” Action Agent guiding us through simple tasks or a Plan-and-Execute agent leading us through complex ventures, they have the power to make our lives easier and more productive.

These intelligent digital “beings” are reshaping the way we interact with technology, therefore bringing us closer to a future where human-machine collaboration is super quick and intuitive.

Obviously, there are multiple possible use cases for AI Agents 4(Rebelo, M. 2023). From Healthcare to Business and Finance 5(Bansall S. 2023).

In this short writing, I will provide one example only – The Personal Tutors.

AI Agents have the power to revolutionise the way we learn. Imagine personalised virtual tutors that adapt to individual learning styles and preferences. AI agents can provide tailored educational content, offer real-time feedback, and identify areas where students may need additional support. You can even ask for 20/80 style learning but about it, I will write in a separate blog post. One way or another, learning becomes more accessible, engaging, and effective, catering to the unique needs of a particular learner. That is a good approach and solution to the lack of knowledge problem and desire to learn IMO.   I am looking forward to it.

Also, it is important to highlight that AI Agents must be designed to respect our privacy and data security. Adequate regulations and ethical guidelines should govern the collection process, storage, and use of personal information to maintain trust and protect our rights.

To achieve desired outcomes, it is crucial to remain aware of the ethical considerations, ensuring that AI agents are developed and deployed responsibly to ensure the beneficial impact of AI agents on society – Us I mean.

AI agents are not meant to replace Us (hopefully).  Most likely, they are designed to augment our capabilities and empower us to achieve more. By automating routine tasks and providing data-driven insights, AI agents free up valuable time and mental capacity for humans to focus on creativity, critical thinking, and complex problem-solving.

At this stage of my life (and I am 60), I love AI and AI Agents. You may like it or leave it. It is your personal choice.  Greetings from the author.

References:

1. What is an AI agent? (2023) Sameai. Available at: https://www.sameai.io/docs/what-is-an-ai-agent/

2. Terra, J. (2023) Agents in AI: Exploring intelligent agents and its types, functions & composition, Simplilearn.com. Available at: https://www.simplilearn.com/what-is-intelligent-agent-in-ai-types-function-article

3. Agents (2023) Langchain. Available at: https://python.langchain.com/docs/modules/agents/

4. Rebelo, M. (2023) What is an AI agent? Zapier. Available at: https://zapier.com/blog/ai-agent/

5. Bansall, S. (2023) 

Originally published at https://spektrumlab.io on Jun 21, 2023.