After four overhauls and millions of real-world sessions, here are the lessons we learned about context engineering for AI agents: manus.im/blog/Context-Engine…

Honestly, I'm surprised too — that's a legacy Manus from a few versions back lol...

Happy to see Manus hitting SOTA in @scale_AI & @ai_risks' new benchmark, but that 2.5% on the Y-axis says we've still got a loooong way to go😅

A new eval, Remote Labor Index, measures AI's ability to automate real-world, economically valuable projects from remote work platforms. Currently entirely unsaturated (max score of 2.5%) A great collaboration between @scale_AI and @ai_risks!

The all new web dev experience are now fully rolled out and available to all users. Behind it is still a unified general agent architecture, where every capability works together seamlessly: researching deeply to create sites, analyzing collected data, turning those insights into slides, and even emailing those slides to potential investors. Beyond web dev, this update also brings significant improvements in speed, intelligence, and overall usability. More details in our latest announcement blog: manus.im/blog/manus-1.5-rele… And as always, we're keeping our approach transparent. We recently shared a technical walkthrough on context engineering — including how features like unlimited context are implemented — in our latest webinar: piped.video/6_BcCthVvb8 Couldn't be prouder of what the team pulled off. 🙌

Context Engineering in Manus i had a great conversation w/ @peakji abt the design of @ManusAI_HQ + how these use context engineering. wrote some notes here (video link below). rlancemartin.github.io/2025/…

Grateful to be named in this year's MIT TR35 at 33. Fourteen years ago, I was fortunate to be on Forbes 30 Under 30 at 19. The past 14 years have been about learning, failing, and striving. Innovation is a journey, not a moment.

What's going on here @GeminiApp, aren't we supposed to be best buddies? 😂

Today a16z released the latest AI Top 50 list. As a small startup, it's truly an honor that we made it with not just one, but two products. The company is charging toward the $100M milestone, and I'm about to enter my 15th year in entrepreneurship. In the beginning, I thought starting a company wasn't that complicated: just follow your curiosity and build products that delight people. After years of stumbling, grinding, and learning, I've come to see that entrepreneurship really is just that: follow your curiosity and build products that delight people.

🚨 Announcing the latest @a16z top 100 AI apps! This marks the fifth edition 🎉 of our global ranking of consumer AI websites and mobile apps by usage. And in my opinion, it was our most surprising list yet! Our top takeaways + who made it 👇

Manus hits $90m run rate! Remarkable execution and a fantastic product being built in Singapore. Building exclusively on models from Anthropic, Google, and OpenAI, Manus is defining the next generation of ai powered productivity software for consumers and teams.

Just had a great conversation with @paulharapin at Stripe Tour Singapore. It's amazing to see Manus grow from launch in March to a $90M revenue run rate, with $100M just around the corner. Grateful for Stripe's strong support along the way!

Trying Manus Wide Research and loving it. If I were a product guy at that company, I’d be thinking of ways to make Manus more proactive. I’d love to have a daily or weekly call with the product where I can brainstorm research projects it could do for me then go off and do them. Almost everything on my to do list could be researched. In addition, there are continuous research projects I’d love to run with results being shared with me everyday. I would also have some sort of prompt optimization / personalization that it would help me modify weekly.

With “Wide Research,” @ManusAI_HQ is bringing about a fundamental change in agentic work. I instructed the tool to scan 100 scientific articles – including titles, abstracts, authors, journals, year, number of citations, institution, and country. Everything was automatically written into a table, sorted by relevance and impact. The result: an immediately usable overview of the global research landscape. Ideal for science journalism, strategies in health tech – or simply to have more informed discussions. Wide Research is rethinking research: – Parallel scanning instead of individual queries – Data directly usable for Notion, Excel, or video scripts – Saves time, reduces costs – without compromising on depth Here is the prompt:

Wide Research is our latest exploration in agent-agent collaboration. Built on our large-scale virtualization infrastructure, Manus can now autonomously dispatch a team of homogeneous Manus agents to work in parallel and aggregate the results. While building AI agents, we've been consistently inspired by classic systems research. Wide Research was directly inspired by the MapReduce paradigm introduced over 20 years ago by @JeffDean and Sanjay Ghemawat. As pioneers in large-scale distributed systems, Google encountered challenges others hadn't yet faced, and generously shared their solutions with the world. Today, as Manus pushes the boundaries of AI agents, we're running into a new class of problems that only emerge at scale. We'll continue to share what we learn along the way.

Choose the system architecture that benefits the most from switching to a stronger model, instead of comparing design choices using the same model. I treat it as a regularization factor in context engineering. It helps you discover more robust, generalizable architectures. It isn't just about future-proofing.

Forgot to include a key lesson in the blog: Context engineering can also overfit - not just to specific model families, but also to today's model limitations via premature optimization. At Manus, we never commit to an architecture based on static benchmarks. Instead, we treat models (M) as external variables and evaluate agent architectures (F) using two deltas: 1. Maximize F(M_strong) - F(M_weak) within the same model family 2. Minimize |F(M_a) - F(M_b)| across different model families We prioritize (1) over (2), and only consider (2) when models are agent-aligned. It's not about chasing SOTA. It's about ensuring that your agents remain orthogonal to models and future-proof as raw model capabilities evolve week by week. In the rising tide of model power, build speedboats, not pillars. 🚤

Today, Manus Integrates Veo3! Manus isn't just where you generate videos — it's where you craft cinematic experiences. Characters speak naturally with synchronized audio, scenes unfold longer to tell complete narratives, and everything feels sharper, richer, more alive than ever before. Veo3 allows for that extra visual finesse in comparison to our default model. The stage is set. The story is yours to tell. Available to Basic, Plus, Pro Members.

Thinking of you guys, while the rest of us are squashing bugs back in Singapore. 🫠