- AI For Humans - The Newsletter
- Posts
- Is Manus AI's Next DeepSeek Moment?
Is Manus AI's Next DeepSeek Moment?
Don't call it Deeper Research: The powerful new Chinese AI agent that's turning heads
Today on AI For Humans The Newsletter!
Manus this an exciting week in AI
GPT-4.5 is now available to ChatGPT Plus users 🎉
And Hedra’s new v3 will animate anything with two eyeballs and an orifice
Plus, our can’t-miss AI feature of the week!
Welcome back to the AI For Humans Newsletter!
If you’ve been listening or watching us for a bit, you know that AI Agents are all the rage. Just last week, we discussed OpenAI’s potential $20,000/month coding agent that is looking to upend the entire software market.
But what if I told you there’s a new AI Agent on the market that can to do a lot of what OpenAI and Anthropic’s agents promise to do soon?

The Manus Homepage looks familiar…
Manus is an autonomous AI agent developed by Chinese startup Monica, designed to independently execute complex tasks without continuous human input. It operates asynchronously in the cloud, allowing it to continue working even when users are offline, much like OpenAI’s Operator. Unlike Operator, using Manus for now is free if you have access (it’s in a limited beta).
What makes it different? Well, outside of being fast and free, Manus can code in addition to performing research like OpenAI’s Deep Research product. We used it to give us the best possible strategy for late-game Balatro and it did a remarkably good job.
On the other hand, we attempted to get it to ‘vibe-code’ a game called ‘Endless Walking Simulator’ and at first it looked like magic (it created a fifteen step process and was churning through them one by one) but once it got to around step nine it hit a ‘fatal error’. In part, this was because Manus was trying to bite off too much, but you’d hope the system would better understand what was happening.

Manus tries to do too much sometimes…
It also turns out that it might not be magic, just clever engineering and prompting around Anthropic’s Computer Use + a number of additional tools. Is that as bad as a lot of people want you to think? We’re not so sure. A good wrapper of an LLM system is still incredibly useful.
One thing it won’t be for long (we assume) is free. This has to be costing a fair amount of money to run and we suspect before Manus becomes public we’ll be seeing a business model or, maybe like DeepSeek, they’ll just try to hammer away for market share. China wants as big an AI foothold as they possibly can get.
Overall, we’re pretty impressed with Manus as a whole and it’s important to remember this is the worst is will always be. AI Agents are set to transform the entire world and, if Anthropic’s Dario Amodei is to believed, might make 100% of code writing obsolete in just 12 months.
More on Thursday!
-Gavin (and Kevin)
3 Things To Know
Hedra Launches v3 Of Their Lip Sync Model
Early pioneers in the space, Hedra, are pushing well past basic lip sync to generate increasingly believable animated avatars. Unlike most tools, it even handles animated characters. Unlike Runway Act-One it doesn’t required a driving visual performance, just an image + some audio or a script.
GPT-4.5 Begins Trickling Down to The Everyman
Just last AI year (last week) in the newsletter we discussed OpenAI’s controversial new GPT-4.5 model, available only on the $200/mo Pro plan. Now Plus users are getting limited access, and as a happy Team plan user I can share I now have 4.5 in my model picker as well.
Sudowrite Launches Muse, an AI for Fiction
As AI leaders race to develop more powerful email drafters and flight sim generators the team behind Sudowrite claim they’ve shipped the first AI designed for authors. It’s been tough sledding for AI writing apps in the past, with many coming and going (pivoting to copywriting) in the past couple years, but we’re here for the specialist tools purpose-built around a specific craft.
We 💛 This - Runway’s Restyle Video
AI video champ Runway shipped a sneaky good feature this week to alter the look & feel of any footage.
Simply provide a restyled first frame, that is, a version of the first frame of your footage that’s been run through a tool like Magnific to be modified while maintaining the same structure.
Some of the examples we’re seeing in the wild are really cool. Interestingly, the input video quality kind of doesn’t matter. The framing & positioning do, but it means you can shoot just about anything, on just about any hardware, and restyle it to your heart’s content.
Are you a creative or brand looking to go deeper with AI?
Join our community of collaborative creators on the AI4H Discord
Get exclusive access to all things AI4H on our Patreon
If you’re an org, consider booking Kevin & Gavin for your next event!