This Week in AI: Manus Decoded + Member Spotlight on Healthcare Innovation

Cutting through the noise in AI

Welcome to the Women Who AI Newsletter, your weekly update on what actually matters in AI when you’re focused on building and scaling startups.

Was this forwarded to you? Click here to subscribe and join our community of women founders shaping the future of AI.

This Week’s Deep Dive: How Manus Really Works

China-based Butterfly Effect recently launched Manus, an AI agent generating serious buzz. Manus is an AI agent system designed to automate complex tasks by acting like a team of digital teammates. It doesn’t just chat; it works in a virtual environment, collaborates with humans, and learns from interactions.

While impressive for research and analytical tasks, Manus struggles with system crashes, large text processing, and accessing paywalled content. Check out this MIT Technology Review article to learn more.

Let’s break down how it works.

What can Manus do?

Manus interacts with the world using tools you recognize:

• Browser: Visits websites, takes screenshots, and scrolls (like a human user).

• Terminal: Runs commands (e.g., running python files).

• Text Editor: Writes and edits files.

• Search: Pulls real-time data from the web.

All tools run in a safe, isolated sandbox.

The Sandbox: Virtual Machine (VM)

Manus creates and maintains a dedicated Linux environment for each user, allowing it to install tools, save files, and build on past work within this virtual computer. What makes this particularly powerful is your ability to take control at any moment using VNC technology (similar to Zoom screen-sharing), letting you seamlessly step in to edit code or guide the AI mid-task whenever needed.

Prompts

Detailed system prompts that serve as Manus’ blueprint. These instructions define how Manus interacts with its environment, manages tasks, and handles errors, ensuring it acts efficiently and safely.

You can browse some of these prompts here.

Roles

Manus uses specialized agents (think: a team of experts):

• Planner: Maps out step-by-step strategies (“Research → Analyze → Build”).

• Specialists: Browser wizards, coding pros, and data organizers handle specific tools.

• Knowledge Manager: Acts as a librarian, storing and retrieving key info.

• Orchestrator: Ensures everyone stays on track without losing context.

Memory System

Manus employs a memory system that continuously proposes fresh information through its Knowledge Base while maintaining long-term files that function as extended memory.

The system records every action in a comprehensive timeline via its Event History feature, allowing users to visualize the decision-making process.

Why This Matters for Founders

For founders navigating the complex demands of early-stage companies, Manus represents a potential force multiplier for small teams. Virtual machine architecture allows you to build custom workflows that grow with your startup (market research → report drafting → presentation design). The collaborative interface means you can delegate complex research, market analysis, or competitive intelligence gathering while maintaining quality control, stepping in when human judgment is needed. Its persistent memory system accumulates institutional knowledge about your business, leading to improvement over time.

Setting Up OpenManus Agents

Still waiting on your Manus invite code? Don't want to send all your startup secrets to a mystery server? You have options! OpenManus is an open-source project that replicates Manus' functionality, built on top of Browser Use (the open-source resource we shared last week).

Quick Setup Guide

Visit https://github.com/mannaandpoem/OpenManus and follow the Installation Instructions.

I followed Method 2: Using uv (Recommended). Here's what you need to know:

1. After initial installation instructions, open config/config.toml and add your OpenAI API key (get one at https://platform.openai.com/api-keys).

2. Note that any line with a # in front is commented out. Focus on filling out the two fields that aren't commented.

3. Make sure to run playwright install before starting.

4. Unlike the official Manus, OpenManus doesn't run a full virtual machine. Instead, it uses Playwright to navigate a web browser.

Performance

I tested OpenManus with Claude Sonnet 3.7 (using model = "claude-3-7-sonnet-20250219", base_url = "<https://api.anthropic.com/v1>" in the config file) and found it could handle all the same tasks that MIT Tech Review tested with the official Manus. For longer, more complex tasks, I needed to increase max_steps (default is 20) in app/agent/manus.py. I also tested a local configuration with Llama 3.2 via ollama, which worked out of the box with the sample config code. Overall very worth it, if only to get an understanding of how the system works on a relatively small codebase.

Member Spotlight: Ayushi Sinha, Turmerik AI

Meet Ayushi Sinha, founder of Turmerik AI, who's leveraging her unique background to build a company at the intersection of AI, healthcare, and equity. Coming from a family of doctors but choosing the "CS route" at Princeton, Sinha developed Turmerik AI's business model during her time at Harvard Business School—evolving from an expert data labeling company that gave her "free access to amazing CEOs and heads of hospital systems."

For founders navigating regulated industries, Sinha's approach offers valuable lessons: "You can't be a 'random AI girly' in healthcare, you need anchors of credibility." Her multi-pronged strategy includes partnering with respected MDs who co-design and vouch for their tools, building hybrid teams that balance technical expertise with domain knowledge, and taking a phased approach to compliance. As she puts it: "I'm not a doctor, but people on our team are doctors, and our customers are well-known doctors. I'm standing on the shoulders of giants."

Read our full interview with Ayushi to learn about her founder journey and how Turmerik AI is addressing critical diversity gaps in healthcare data →

Jobs We Love

AWS DevOps Engineer (Contract) at Nyxion – Cambridge-based Global Impact Assessment is looking for an experienced AWS DevOps Engineer to build and maintain cloud infrastructure for Nyxion. This contract role combines technical expertise with social impact, requiring strong proficiency in AWS services, Infrastructure-as-Code tools like Terraform, CI/CD pipelines, and containerization with Docker and Kubernetes. They're particularly interested in candidates with experience in machine learning applications and impact measurement.

If you're passionate about applying your cloud engineering skills to meaningful work, this position offers the perfect blend of technical challenge and purpose.

To apply, email your resume to [email protected].

Reply With Questions

We want this newsletter to address the real challenges you're facing. Is there a specific AI development you'd like explained? Jargon we included but didn't properly explain? A business problem you're trying to solve with AI? Reply directly to this email with your questions, and we'll tackle them in next week's edition.

Share the Knowledge

If you found value in today's newsletter, please consider forwarding it to other women in your network who are building, or thinking about building, in the AI space. The more we grow this community, the stronger our collective impact becomes.

Here's to building the future of AI, together.

Lea & Daniela