Imagine you’re deep into a market research project that requires accuracy, depth, and a lot of data. You might think, “I’ll just ask ChatGPT or Claude to create a report,” and while they can give you a good outline, they often miss the little details that make a report truly valuable.
It’s not that these AI tools can’t do the job—they’re incredibly powerful. The key is in how you use them. To really get the best out of AI for research and writing, you need to go beyond a basic approach.
Today, AI agents are changing the way researchers and writers work. The long hours spent sorting through data, organizing thoughts, and polishing every detail are becoming a thing of the past. AI can now handle much of that work, letting you focus on what’s most important—coming up with insightful conclusions and crafting a clear final product.
But let’s think bigger. Imagine not just one AI agent, but a whole team, each focused on a different part of your project. They work together smoothly, helping you through every step of research and writing, from gathering information to fine-tuning your analysis. Now that’s a real game-changer.
Enter Collaborative Multi Agent design
Collaborative Multi-Agent Architecture is like having a team of AI specialists, each with its own area of expertise or task. These AI agents don’t work alone; they collaborate, sharing information, ideas, and progress in real-time to produce a unified, well organized result. It’s like having a group of experts working together on a project, each contributing their unique skills, but with the added speed and precision that AI brings to the table.
This is by no means a novel idea, the system i developed is based on GPTResearcher an open source project that employs agents for research and writing. With our system, we have used the researcher and added extra agents to take this capability a step further to create really powerful documents that hopefully can be used in the real world!
How does this compare with vanilla ChatGPT or Claude?
Now, I have been fiddling with AI Agents for over a year now, and this design philosophy intrigued me. Having a team of robots working for you as a team, together without me having to monitor them sounded exciting! I manage a team of engineers every day (and they are great!) but having agents do the work for me with NO supervision, doing exactly what i want them to do, sounded like a great idea!
To put this idea to the test, I decided to create a team of AI agents specifically for research and writing. I won’t get into the technical stuff here—that could be a whole blog series on its own. Instead, let’s focus on the main question: how did the results turn out? Is this team of agents really better than using just ChatGPT or Claude?
To find out, I gave the same prompt to ChatGPT, Claude, and my newly assembled team of agents. Then, I compared their outputs side by side. I put each of their responses into a PDF for a clear comparison. Here’s the exact prompt I used with all three systems.
Prompt
Create a research report on how climate change affects purchasing patterns in the world.
Do latest research as of today
Result of ChatGPT using GPT-4o
I tested ChatGPT by asking it to create a research document for me. Since ChatGPT usually aims for concise responses, I prompted it to “Generate the longest document possible” for this test. The result? A 5-page document produced over two chat sessions. Impressive, but it had a few quirks.
The content was generally clear, well-organized, and up-to-date with current research. However, there were some noticeable gaps. Even though it found relevant information, ChatGPT didn’t add sources and citations correctly—an unexpected miss for an AI meant to help with research. Also, the document lacked a table of contents and detailed formatting, both of which are key for a polished, professional look.
Result of using Claude
I tested Claude with the same prompt I used for ChatGPT, and the results were surprisingly similar. Claude produced a well-structured 9-page document, but it reached a limit and cut off, noting that it had exceeded its content capacity.
Claude did impress me by generating a table of contents and mentioning that it had access to research data up until 2024. However, it didn’t perform any real-time research online. Like ChatGPT, Claude also missed including citations altogether.
In the end, while Claude created more pages and delivered a well-organized document, it shared the same limitations—no citations and difficulty handling very large tasks smoothly.
Result of Multi Agent Collaboration
This is where things got exciting. I put my multi-agent system to the test with the same prompt but limited it to creating up to 10 sections. The results were stunning.
The system quickly sprang into action, with agents researching, writing, reviewing, and publishing as if they’d been working together for years. The result? A 16-page document that was deeply researched, well-structured, formatted with tables, and complete with citations. The document wasn’t just thorough; it was also thoughtfully organized in terms of both structure and content.
The real advantage of this approach is the built-in feedback loop. After each agent completes a task, a supervising agent reviews the work, much like in real life. This process leads to a more polished, deeply researched, and well-reviewed report, outperforming both Claude and ChatGPT in terms of quality and thoroughness
Conclusion
After trying out ChatGPT, Claude, and the multi-agent system, it’s clear that AI can really help with creating content. While ChatGPT and Claude did a decent job, the multi-agent system truly impressed me. It produced a 16-page document that was not only detailed and well-organized but also included proper formatting and citations. Theoretically, this system can create a document thats 100’s of pages long, and can be customised to write not just research reports, but stories, plot lines, market research documents, the possibilities are endless.
That said, even though the multi-agent system did a fantastic job, it’s not perfect. It still needs a human touch to add those final finishing touches that make a report truly polished and professional. But as a starting point for research and writing, it’s hard to beat. It gives you a strong foundation that you can easily build on with just a bit of extra effort.
In the end, the multi-agent system is a powerful tool that, when combined with human insight, can lead to some pretty amazing results.
Check out the documents yourself and tell me what you think on our socials!
