Who am I?

Hello and welcome to my next article in the AI Knowledge Hub! 🙋‍♂️

My name is Jacob Kowalski - a corporate tax consultant from Poland who is really into AI stuff.

I create many really cool things with AI: chatbots, AI assistants, automations, agents, etc. 🤖 

I think we are living through the biggest revolution of our generation and want to share my thoughts about it.

Maybe some of you will find it interesting, or maybe you want to share your take on this too.

If you want to know more about me and what I am doing, check out my page with the button below. 👇

ChatGPT Agent is Coming!

Hi everyone! 👋

It is really funny - whenever I am writing an introduction for every article, one common thing comes to my mind first: AI development is so fast. 💨

After a couple of times, this is starting to sound cliché, but it's true - every week we get something new that could change the whole perspective of what AI can do for us. 🤖

The main inspiration for creating this article was the announcement of ChatGPT Agent - the new functionality of normal ChatGPT, which connects many of its functionalities into one to execute tasks more precisely. 🎯

Once I dived deep, I discovered that there are many of these kinds of "Web Agents" already, so I decided to test them out and… WOW. 🤯

These kinds of Agents, once developed, will be the new way we interact with the whole internet. 🔎

Okay, enough hype, let's get into it! 👇

The New Era of Internet: How AI Agents Will Handle 80% of Your Web Tasks

Your current interaction with ChatGPT is likely this: typing a question or prompt, waiting for response, some refinements and you are good to go. This tools are not doing anything besides that, they are only answering questions (more or less accurate).

But image having in your ChatGPT a digital assistant that books flights, fills out information forms when buying something, manages your emails, schedules meetings, and completes surveys while you focus on more important things.

This is AI web agents – and they're revolutionizing internet interaction.

Unlike chatbots that just answer questions, AI web agents actually take action. They can:

• Connect to your email inbox and draft, send, or organize messages

• Access your calendar to schedule meetings and check availability

• Fill out forms and surveys automatically

• Complete online purchases by navigating sites and executing transactions

• Research information from multiple sources into organized reports

This is only a fraction of what AI Web Agents will be capable of.

ChatGPT Agent Launch: Prepare for the New Way of Interacting with ChatGPT

OpenAI has just launched ChatGPT Agent, a feature that allows ChatGPT to perform tasks directly on websites instead of just providing information.

According to the OpenAI announcement ChatGPT Agent can complete complex online tasks on your behalf. It conducts research across websites, uploaded files, and connected sources like email, while performing actions such as filling out forms and editing spreadsheets - all while keeping you in control.

Key capabilities:

• Browse websites using a built-in virtual browser

• Fill out forms automatically on any website

• Access email and calendar to schedule meetings

• Create presentations and spreadsheets with real data

• Run code and analyze files for reports

• Log into websites securely with user permission

ChatGPT will intelligently navigate websites, filter results, prompt you to log in securely when needed, run code, conduct analysis, and even deliver editable slideshows and spreadsheets that summarize its findings.

ChatGPT Agent represents the biggest leap from "AI that talks" to "AI that acts." It combines several existing ChatGPT features into one powerful system that can handle the tedious tasks eating up your workday.

Important note: This function is available for ChatGPT Pro, Plus, and Team plan users.

According to OpenAI, everything sounds very optimistic, but let’s take a look how this Agent works in practice. I tested the Agent function based on 5 different tasks:

• Deep research analysis

• Creating a webpage

• Preparation of a report regarding a given industry

• Creating a spreadsheet with numerical data

• PowerPoint presentation.

Let’s see how ChatGPT Agent handled these. 👇

Deep Research Analysis

The first task is conducting a deep research analysis.

I prompted the AI to prepare a deep analysis of Bitcoin's current price, analyze the changes, explain the current price, and tell me if it's a good moment to buy.

Results below. 👇

First, let's break down the whole report into each section.

The Agent did well on analysis of current and past data, meaning that its capabilities for researching information are quite good."

Next, the Agent explained what the factors were behind the BTC price fluctuations in recent weeks. The explanations were clear, well-formatted, and rather understandable even for a non-crypto person like me.

The next part is about scenario painting. The Agent gave me three possible scenarios for BTC price changes. In this case, the formatting was not that good (the numbers could be better as 1.1, 1.2), but overall it painted a big picture of the possible future alongside with arguments for these scenarios.

Finally, it provided me with a recommendation on whether it's a good time to invest in Bitcoin right now. Of course, ChatGPT cannot give financial advice or be one-sided. However, it gave me a rather satisfactory response with arguments for investing today and brought my attention to potential risks.

If you are interested in reading entire document - full report here.

Creating a webpage

The second one aimed to turn the Agent into a professional graphic designer.

The goal was to create a landing page for an Italian Pizza Restaurant, which should include a menu, prices, opening hours, and should encourage clients to book a table.

Here are the results. 👇

When it comes to webpage creation, the Agent is not fully ready to do it, but it helped a little and gave me some inspiration.

First, it created a test webpage which looks like this:

Nevertheless, it also provided me with some images and visualizations of how the webpage could look:

As of now, the Agent is not quite ready for this kind of task, but I bet in the future it will be, because we have so many tools that could help us with this.

Industry report

The next task revolved around an industry report with graphs and tables included.

This idea came to my mind as in my job I prepare sections with industry analysis, so the Agent feature could be very helpful here.

My goal was to prompt the Agent to create a deep report for the EV (electric vehicles) industry with graphs, infographics, tables, etc. included, in order to copy-paste it into my file.

Below you can find how the Agent handled this task. 👇

As in the Bitcoin case, it performed really well with gathering and formatting information.

It provided me with so much more than I asked for, and I think this will be the most common option for me to use.

Please find some sections from the report below:

Unfortunately, it is not capable of creating or copying graphics, making its own charts, etc., but besides that, it performed very well.

Full report here.

Spreadsheets with numerical data

Here's the corrected version:

The fourth task tested how well the Agent could gather and format numerical data.

In this case, I prompted it to gather the inflation rates (from Eurostat) of 5 European countries: Germany, Spain, France, Poland, and Italy, from 2020 to 2024.

Here you can see how it went. 👇

After prompting and 9 minutes of waiting, the Agent came back to me with this:

Also it created a spreadsheet with the data.

For reference I also checked the inflation rates on Eurostat page:

As you can see it worked perfectly. This is also the function that I will be using most of the time.

Powerpoint presentation

The last task was about creating a presentation in order to check if it can combine image and text generation in slides. The presentation was about Large Language Models (LLMs) and should provide basic information for non-technical students.

Here are the results. 👇

Generally it looks good, but very basic to me. The information included is too short and sometimes the graphics are out of place as you can see below.

It worked for around 60 minutes on this and in my opinion, the results were not worth it, I recommend using some other tools (more info in the next section).

Full presentation here.

ChatGPT Agent Best Alternatives Manus.ai & Genspark

As you can see, ChatGPT Agent is not an ideal web agent for every use case. It has some strengths but also cannot perform some tasks or performs them in poor quality.

This is why some other tools have emerged to enable tasks that OpenAI cannot do yet.

The first is manus.ai - a platform which is able to do similar tasks but can also create a full website from scratch. I highly recommend testing it; as a free user, you can use some of its capabilities.

Below you can find the same website that I wanted ChatGPT Agent to create, developed by manus:

The next alternative is genspark.ai - which is quite similar to the previous tool but can also download certain files, fact-check, make calls, or even create podcasts about given topics.

I tested the podcast function and it is actually insane! The quality of this is even better than some "human" podcasts! It also has the capability of connecting to your Gmail, Notion, or calendar to automate certain tasks. This also works incredibly well.

I highly encourage you to give it a chance and try to create something powerful.

Summary

Congrats, you have made it to the end. 🔚 

To recap everything covered above, ChatGPT Agent is a quite nice and helpful feature; most use cases (at least for me) will be gathering and formatting data, both numerical and text. It is not something revolutionary but nice to have when you are working with ChatGPT anyway.

I also tried it with recommendations and buying something through the chat window, but for now it needs more improvement in the whole process. At the moment I prefer typing and checking products "manually," but in some time, as this function gets some improvements, we could experience a change from Google to ChatGPT shopping.

Also, the alternatives are in quite a good place as well; manus and genspark are very convenient tools for their use cases and could be more interesting as the development of Web Agents increases.

Thanks for reading.

Jacob