- The Inktelligence
- Posts
- The Inktelligence - July 21, 2025
The Inktelligence - July 21, 2025
The Buzz Around AI Agents

Image created and animated with Midjourney
This past week has been a significant one in terms of the new developments around AI agents. First we have Perplexity releasing a preview of Comet, their AI assistant-driven browser, followed by Open AI’s release of Agent Mode.
My impressions of Comet
First let’s take a look at Comet. It’s based on Chromium, which is the open-sourced version of Chrome so you’ll get the look and feel of Chrome. This also means that all your plugins will easily transfer over to Comet.
Here’s what you’ll see on the home screen. You can add other widgets besides the ones that come as default. There is a sidebar where you can chat with the “Assistant”.

Basically, it’s like having a web browser that works in concert with the AI. You can do things like ask the AI to open tabs, close tabs, fill out forms, etc. The benefit of having this work in the browser is that it can also interact with apps that are running locally. You can point to “localhost:XXXX” and have the AI interact with the app.
An example of this is running Ollama locally on your machine. Now I can have an AI interact with another AI. :)
Here’s a video of me asking Comet to gather some information and then creating a new Google Doc and compile all the information that it found. I think performing research is a good use case for it.
Another cool feature of Comet is that you can turn on “Voice Mode” and just start talking to the AI Assistant.
After I posted this on X, Aravind liked my post - nice!

At the moment Comet is only available to Perplexity Max ($200/month) subscribers. Others can get on the waitlist. But you can also skip the waitlist if you have an invite. I was able to secure my invite from someone who works at Perplexity.
BTW, I have an extra invite to Comet. The first person to send me an email with the subject line “Comet, Please!” will get it. Please email me at [email protected]
ChatGPT Agent
Next comes ChatGPT with “Agent Mode”. If you’d like, you can read OpenAI’s official release. This is ChatGPT’s foray into AI agents. You might be wondering how this is different from Operator which was released earlier in the year.
While Operator mainly focused on the ability to control a web browser and take actions on it, Agent does more by combining 3 different capabilities. According to OpenAI:
“At the core of this new capability is a unified agentic system. It brings together three strengths of earlier breakthroughs: Operator’s ability to interact with websites, deep research’s skill in synthesizing information, and ChatGPT’s intelligence and conversational fluency.”
In order to test Agent’s capability, here's the prompt that I used:
"I want to create a LinkedIn carousel about the features of the newly released 'Agent' from ChatGPT. I want 6 slides in the carousel. Make it look good using mid century modern colors and serif font."
Will it be able to do it? Yes and no.
It was able to follow most of my instructions except it did not create a LinkedIn carousel. Instead it created a Powerpoint slide deck. It was able go out to gather information from the web, and then created the slides using mid-century modern colors and a serif font.
I didn't specify what the slides should cover, just that I wanted 6 slides. It decided on its own what slides to create.
The entire session took about 15 minutes. I shortened the video so that everything including the results are captured in about a minute and a half.
Agent is now rolling out in stages. Pro subscribers should have access to it now. Over time, it’s becoming available to Plus, Team and Enterprise subscribers as well. I do find that it runs a bit slow thus detracting from the user experience. But hopefully that will improve as OpenAI adds more GPUs to their datacenters.
There are some other agentic AI platforms like Manus and Genspark that can run code and have access to other tools as well. They are quite promising based on my experiments with them. Both these platforms run on credits (which I’m not really a fan of) vs. an all-you-can-eat plan. You can get some free credits to try them out. The problem with a credit system is that you use up the credits even if the task does not complete properly.
Find out why 1M+ professionals read Superhuman AI daily.
In 2 years you will be working for AI
Or an AI will be working for you
Here's how you can future-proof yourself:
Join the Superhuman AI newsletter – read by 1M+ people at top companies
Master AI tools, tutorials, and news in just 3 minutes a day
Become 10X more productive using AI
Join 1,000,000+ pros at companies like Google, Meta, and Amazon that are using AI to get ahead.
Well, that’s it for now.
Till next time,
Hock