Manus - Hottest Thing This Week

This edition, unlike previous ones of The Agentic Era - will be solely focused on Manus AI - we couldn’t miss this ☺️ 

A Video that went viral

Probably everyone saw the video of Manus AI performing 50 different tasks, setting somewhat of a dystopian scene. But is that video what said general agent can really do or what we, AI enthusiasts, want to see?

Not another Deepseek moment for China?

If you remember, admiration from the Deepseek’s release lasted a couple of days - maybe a week. After the admiration, came criticism of US AI scene - questions such as why China went open source? Why so many US companies, who had a head start couldn’t achieve what Deepseek did. Tone changed. Completely.

All of the US tech and AI influencers (and I’m not suggesting this was in any way briefed) as one, defended mainly OpenAI and reassured the world of the United States’s capabilities in Artificial Intelligence.

Nobody could say Deepseek wasn’t impressive - it bloody was and still is, however, everyone, defending US AI scene said as one - this is a small setback - wait, our guys have stuff up their sleeves.

Well, a couple of days ago OpenAI announced 20,000 USD per month AI Agents. I’m not sure how much Manus AI will cost - but I’ll bet three months for OpenAI’s Agents - it won’t be a portion of what Altman’s offering.

With the announcement of Manus (Probably should have worked more on the naming part) some of the biggest US media outlets - acknowledged that there is a new player and that it’s impressive what we see but - weren’t too keen on focusing on positives - maybe - rightfully so!

There were lots of reports of Manus not handling tasks as well as users expected or maybe, that’s the main reason why they’re still in closed beta.

New Manus Moment

Use cases include gathering lists, from different databases, gathering public opinions, analysing different stocks’ performance and public sentiment on topics of your choice, as well as editing your podcast or getting a set of icons with your brand identity.

Use cases vary - but let’s get into tiny details - how Manus works (or how it should work)

Manus leverages Antropic’s Claude 3.5 Sonnet and Alibaba’s open-source Qwen, together with independently operating agents that act autonomously on a wide range of tasks - that’s why Manus claims to be the first general agent on the market.

How Manus Really Works

As only 1% of the wait list got access to Manus, Here’s TL;DR of all the reviews on the internet:

• Manus experienced crashes on different tasks, on some, it became “lazy” - completing only a portion of the task.

• Manus has a higher failure rate (they are working on it) than ChatGPT DeepResearch and Manus’s per-task cost is about 2$, a tenth of DeepResearch's cost.

• Manus is an intuitive, easy-to-use tool, primarily for analytical tasks that require research but has a limited scope because it failed to go through captcha and paywalls

• On some tasks, Manus takes a long time to complete (there were examples shown of tasks taking as long as a couple of hours

• Everyone says the same - when Manus will get more stable - it’ll be great to use

Whenever I’ll try it first hand - Manus Review Edition will be on its way from The Agentic Era.

Y Combinator’s roundtable on Vibe Coding

AI Agents Infrastructure Stack

You might have seen this stack, gathered by Madrona on your LinkedIn threads. Not many might know that this is only part of what John Turow has covered in newsletter, where this image has been taken from, so I’d like to urge everyone to check out the full newsletter - here