Like most of my fellow ‘technology journalists’, I was intrigued by the idea of having an ‘AI Agent’ catering to my work needs. In all my years of serving my profession as a humble contributor, I have always yearned to offload some of my basic, repetitive tasks to someone, while I can focus on the greater things, i.e., brainstorm on new concepts and write. Sometimes, I have also felt the need to have a brainstorming partner, which is a tough find in the industry. Believe me, the urge to find a like-minded brainstorming partner is real.

Hence, when ‘Agentic AI’ started dominating the headlines, I always wondered whether I could have a smart assistant with whom I could discuss and offload some of my repetitive tasks – a helper that could speed up my research and fact-checking process. 

I figured that my personal assistant could be an AI agent, i.e., one that researches, organises information, performs tasks, and even makes decisions within the boundaries I set. The internet is full of examples where smart coders and engineers have created truly intelligent assistants capable of handling the boring stuff in the background while the humans concentrate their efforts on creativity and productivity.

I, however, have no idea of coding, and hence, figured out that I need to rely on a frontier-level AI chatbot to create my first AI agent.

And what better medium to do it than ChatGPT, that too in the free tier!

But where do I start?

The three levels of AI agents

Level 1: Personal AI assistant (No coding required)

This is the easiest and fastest way to create an AI agent. Using ChatGPT’s custom GPT capabilities, one can create an assistant with specific instructions, knowledge, and responsibilities. Think of it as hiring a specialised virtual employee and giving them a detailed job description.

GPT-based agents can:

– Conduct research

– Summarise articles

– Generate content ideas

– Analyse documents

– Answer domain-specific questions

– Follow personalised workflows

Most people can create one in less than 30 minutes. Seems easy for someone new to agentic AI.

Level 2: An AI Agent that takes actions

The second level goes beyond conversations. These agents can interact with tools and applications to perform tasks such as:

– Reading emails

– Managing calendars

– Creating spreadsheets

– Gathering information from websites

– Preparing reports

– Organising research

Instead of merely suggesting what needs to be done, these agents can actually do much of the work themselves.

Level 3: Fully autonomous AI Agents

This is where things become truly powerful. And serious. Advanced agents can:

– Operate continuously in the background

– Access APIs

– Connect with databases

– Execute workflows automatically

– Monitor events in real time

– Trigger actions based on specific conditions

These systems typically require knowledge of coding, cloud infrastructure, and more sophisticated development tools. They are often used by startups and enterprises to automate complex operations.

I chose ‘Level 1’ assistant for my journey: Here’s why

As someone who writes about AI and technology every day, I realised that much of my workflow involves repetitive tasks, i.e.,

– Tracking announcements from AI and tech companies

– Researching technical concepts

– Finding story angles

– Helping with article structuring

– Generating headline ideas

– Drafting image-generation prompts

– Summarising lengthy reports.

These are exactly the kinds of activities an AI agent can help with. Moreover, I prefer taking action on any activity personally, and unless I have confidence with the AI managing my basic work needs easily, I chose to stay away from automation for the time being. 

Also, instead of building a complex autonomous system, I decided to create a specialised AI assistant inside ChatGPT Projects. 

The reason? It is easy.

Step 1: Define my assistant’s role

The first thing I did was write a clear mission statement to ChatGPT. Rather than creating a generic assistant, I gave my ChatGPT a specific purpose:

“You are my AI and technology journalism assistant. Your role is to help me research emerging technologies, identify newsworthy developments, generate story ideas, suggest headlines, explain technical concepts, and assist with article production.”

Remember that the more your specify your needs, the more useful the agent becomes.

Step 2: Telling my ‘AI Agent’ what to do

Next, I outlined the tasks I wanted the agent to perform. My list included:

Research Support

– Monitor major AI companies like OpenAI, Anthropic, Google, Perplexity, etc.

– Explain new models and technologies

– Summarise research papers

– Compare competing products

Editorial assistance

– Help with headline options

– Suggest article structures

– Suggest unique angles

Make tech jargons easy

– Convert technical jargon into plain English. The goal is to help with understanding.

– Explain tricky AI concepts for mainstream audiences.

– Use as many examples and analogies as possible.

Creative assistance

– Create image prompts that suit my articles the best, complying with the tonality of my publisher.

This essentially became the agent’s job description.

Step 3: I defined my writing style

Many YouTube guides suggested that an AI assistant becomes significantly more useful when it understands how you write. Hence, I uploaded a few examples of my published articles and instructed the assistant to study my writing pattern, looking for my preferences:

– Toneality

– Sentence structure

– Vocabulary

– Headline style

– Storytelling approach

I also specified that the research drafts should:

– Be accessible to general readers.

– Avoid excessive technical jargon.

– Prioritise clarity.

– Maintain a conversational tone

– Focus on the business impact of AI

Step 4: Creating a ‘Research Workflow’

One of the most valuable features of an AI agent is consistency. Hence, I asked my agent to consider the following while doing research on any topic:

– Summarise the topic.

– Explain why it matters.

– Identify key stakeholders.

– Highlight opportunities and risks.

– Suggest article angles.

– Generate headline options.

– Recommend visual concepts.

Did my personal ‘AI Agent’ work? 

It has been a few days since I started involving my assistant in my workflow, and safe to say, the output is far from my expectations. It left me puzzled – it wasn’t working as expected. But the agent isn’t at fault for this.

As we discussed above, the agent does exactly what it’s told, i.e., it follows my lengthy prompt. With the first couple of days of testing, my AI agent has gotten a basic understanding of what I do and how I work. It comes up with ideas following thorough research, suggests headlines that go with the tonality of my publication, and even comes up with useful suggestions for tweaking my story drafts. 

For example, when news broke about Anthropic releasing the Fable 5 model, my AI agent helped with understanding the tiered offering and how it differed from Mythos 5. It also listed all the advantages, disadvantages, and cited research works on the topic. However, it struggled with coming up with story ideas and angles that could be of interest to my readers.

There’s always a 50 per cent chance that my personal AI agent has useful suggestions at my disposal. For all the other times, it needs help with more prompts to get to the desired results.

Hence, just like raising a child, I need to train my AI assistant like a new employee. I need to fine-tune the prompts and make it understand my work needs better. Hopefully, this should start paying dividends in the weeks to come.

However, where my ChatGPT-based AI agent shines is in these departments:

– Headline suggestions.

– Creating cover image

– Helping with SEO keywords

– Shaping my rough ideas.

Most importantly, it allows me to spend less time on repetitive research and more time on analysis, and composing my copies.