The AI agent arena is buzzing, and the recent unveiling of OpenAI's ChatGPT Agent has only intensified the conversation. It's a pivotal moment, pushing the boundaries of what these intelligent systems can do. We're seeing a clear divergence in how companies are approaching this: on one hand, you have the 'foundation builders' like OpenAI and Claude, who are deeply integrating agent capabilities into their core models. Then there are the 'application specialists,' such as Manus, Genspark, Flowith, and Fellou, who are laser-focused on specific use cases and the flexible integration of external tools.
Manus, in particular, made waves earlier this year, sparking a significant interest in the general-purpose AI agent market. Following OpenAI's announcement, both Manus and Genspark were quick to weigh in, offering their perspectives. Genspark, on social media platform X, pointed out some areas where ChatGPT Agent's performance in the demo wasn't quite hitting the mark. Manus, meanwhile, went a step further, publishing a direct comparison showcasing the output differences between their systems and ChatGPT Agent on identical tasks.
To get a clearer picture of how these players stack up in real-world scenarios, a test was devised. Manus, Genspark, Flowith, and Fellou were tasked with independently reporting on the ChatGPT Agent launch, specifically in the format of a one-pager with design elements. The results were quite telling. Manus gave ChatGPT Agent a relatively high score, stepping away from their earlier critical stance. Genspark provided the most comprehensive information, including a structured comparison. Flowith, on the other hand, demonstrated the best understanding of the prompt, delivering a highly standardized and visually appealing one-pager.
Beyond this initial assessment, a more hands-on comparison was conducted, focusing on four typical tasks showcased by OpenAI: trip planning, graphic design, information analysis, and wedding planning. The 'wedding planning' and 'team sticker creation' tasks were particularly scrutinized.
For wedding planning, which involved multiple steps like recommending attire, booking hotels, and selecting gifts, Manus produced a lengthy guide but lacked direct purchase links or visual examples of clothing. Flowith encountered an issue with its dedicated fashion task line, resulting in missing information. Fellou took a considerable amount of time, over an hour, to gather information on itineraries, hotels, and dates through multi-threaded browser searches. While it presented a summarized report, it couldn't facilitate direct bookings. Notably, none of the tested products could achieve a closed-loop action, such as direct booking via Booking.com, parsing data from Zola wedding websites, or direct gift purchasing from Registry.
In the sticker creation task, all products could generate anime-style designs from uploaded mascot images. However, only Manus managed to add a sticker to a StickerMule shopping cart. The others provided ordering instructions but explicitly stated they couldn't handle payment submission or account verification.
This comparison highlights the nuanced capabilities of these AI agents. While ChatGPT Agent is making strides, dedicated platforms like Manus and Genspark are actively demonstrating their strengths and identifying areas for improvement, both in their own offerings and in the broader competitive landscape. The journey towards truly seamless and integrated AI agents is clearly ongoing, with each new development pushing the entire field forward.
