I got ChatGPT Agent to deliver a truly useful report - and then NotebookLM took it to the next level

What happens when you take a billion dollars of computer power and let it research a question for 12 whole minutes? A lot more than if you use that same power for just a few seconds, that's for sure.

Since ChatGPT Agent is so new, I've been exploring its capabilities and trying to understand what kinds of problems it's best suited for. This time, I asked it to compare cloud storage services. It took 12 minutes to come up with an answer.

Also: 9 programming tasks you shouldn't hand off to AI - and why

Because this is a learning exercise, I also asked ChatGPT 4o to do the same thing. Our old familiar chatbot responded in about two seconds. There was definitely a difference.

(Disclosure: Ziff Davis, 's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

Keep reading to the end. Once I add NotebookLM to the mix, the results go from interesting to astounding…

Research cloud storage services

The prompt I gave both ChatGPT Agent and ChatGPT 4o was this:

I need to store 10 terabytes in the cloud. The data will be synced from two users and two local servers. Find me prices, and compare and contrast the advantages and disadvantages of each option.

The first big difference is that Agent asked for more clarification.

This, in turn, resulted in my thinking about my needs more specifically, and then telling Agent that I wanted cloud storage services compatible with my Synology servers.

Look at all of them, but they must be able to accept file syncing. For the servers, they should be compatible with Synology servers.

ChatGPT 4o did not ask for any clarification, so some of the recommendations it made were not compatible with my servers.

You could argue that wasn't the fault of 4o, but really of the person writing the prompt. And you would be right, sort of. Sure, had I added that as a criterion, it's possible that 4o would have limited its response to Synology-compatible options. But that pushes the work back on me, not on the AI.

Our whole goal in learning what these systems can do is to find out how they can help us, not how they can put more demands on us. Agent helped because it asked for some clarification. It elicited a more complete instruction than what I originally gave it.

Also: The best AI for coding in 2025 (and what not to use)

That's the same kind of behavior you'd want from an assistant or a team member. As a manager, I've often encouraged my team to ask questions when they need clarification. The last thing I want is to have someone go off on a project for a day or so, only to come back with useless results because they couldn't be bothered or were too intimidated to ask questions when they needed more information.

As a leader, I've been fortunate that most of my team members have been highly interactive. But over the years, I have had one or two employees who've always stubbornly wanted to go it on their own, without external input, clarification, or guidance, despite my encouragement to be more engaged with the team. Those tenures never ended entirely positively.

Now, here, Agent is doing just the sort of clarification interactivity you'd want from an assistant. Right away, that puts a plus in its column vs. the ChatGPT 4o LLM.

Agent gave me a list of 17 services in a table that contained the plans it was describing, price for as close to 10TB per month as it could find, and some important notes. For some services, like AWS and Google, it described different variants of their offerings.

ChatGPT 4o also gave me back a table, but it only listed six options.

Both provided summaries of advantages and disadvantages for the services they spotlighted. Agent did so after the table, and after its analysis and recommendations section. ChatGPT 4o opened with the advantages and disadvantages section, which made it a little harder to ascertain the scope of the overall response right away.

Agent's response started off with a very helpful general considerations section that provided a solid and workable framework for the various options.

Agent then followed up with its version of the advantages and disadvantages discussion.

Agent then wrapped up with a specific set of recommendations by comparing the various services.

Finally, Agent provided a top-line summary that gave solid guidance for further exploration.

Overall impressions

As is often the case with basic ChatGPT 4o, I was left feeling like I had only been given a cursory, quick answer. I got some insights on what to explore, but felt like there was still a lot of work to be done.

Agent gave me back a fully workable briefing. It asked for clarification regarding how I wanted to use the services, and then tailored its response around those needs.

Also: Hacker slips malicious 'wiping' command into Amazon's Q AI coding assistant

I am familiar with most of the services Agent discussed, and while I didn't double-check all the current pricing, the service descriptions were accurate and so was the pricing, at least for the services I'm familiar with.

Google's storage pricing, for example, was sourced directly from a Google support page with pricing.

In my first article on Agent, I showed how Agent wasn't always accurate or complete in its responses, but it could sometimes shine. This is an example of a project where it does, indeed, provide considerable help.

Also: Is ChatGPT Plus really worth $20 when the free version offers so many premium features?

I've gone down the research rabbit hole on enterprise-level storage before (I have 60TB in the cloud), and it's always been a long haul. If Agent continues to grow in reliability, I can definitely see this becoming a very helpful adjunct tool for research and analysis.

Bringing NotebookLM into the mix

Despite knowing the risks of crossing the streams, I decided to give the Agent report to NotebookLM. I first used the share button at the end of the Agent report, but the download button didn't do anything.

So then I just selected the whole report and copied and pasted it as text into NotebookLM. That sacrificed the table formatting, but at least I was able to get the information into NotebookLM through its paste text option. Then I asked NotebookLM to generate one of its amazing conversations.

NotebookLM took about ten minutes to generate its audio overview, but the results were surreal, in a good way. NotebookLM took the ChatGPT Agent report and ran with it, broke it down into its own understandable version, and even added some value on top of what Agent provided.

This might be AI's first magical combo. Use Agent to do some virtual desktop computer usage and then provide a deep research report. Then augment that report with an audio overview generated by NotebookLM. The final overview ran 14 minutes.

Also: My 8 ChatGPT Agent tests produced only 1 near-perfect result - and a lot of alternative facts

So, in less than 45 minutes (12 for Agent, 10 for generating the podcast, and 14 to listen to the conversation), it's possible to get a very strong overview of whatever topic you set Agent loose upon.

I know I come across pretty suspicious of AI, the potential dangers to society, and its boneheaded moves. But combining Agent with NotebookLM's conversation mode is truly next level, at least in this test case.

• You can view the full ChatGPT Agent session here.
• You can view the ChatGPT 4o transcript here.
• You can listen to the NotebookLM conversation here.

Have you tried using ChatGPT Agent yet? What kinds of tasks do you think it's best suited for? Do you prefer the fast-but-shallow approach of 4o, or the more methodical deep dive from Agent? Have you ever paired tools like ChatGPT and NotebookLM for research? How do you see this evolving in your own workflows? Let us know in the comments below.

---

You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.

Get the morning's top stories in your inbox each day with our Tech Today newsletter.