Special issue: Agents For Your Autonomous Business (Manus, Operator, Proxy)

Good morning.

I wanted to send an issue specifically about a few recent Agents that are coming online.

Agents like these (Manus, Operator, Proxy) have been available and possible to build for a while now…

IF you know software development or have enough funds or experience to build them.

(I’m fortunate enough to both know myself how to do it, and have a team available).

Now they’re becoming widely available. For anyone.

Let’s dive in.

—Sam

IN TODAY’S ISSUE 🤖

• Prediction Fulfilled: General Agents in Q1, 2025

• Manus: The Autonomous Pioneer

• Operator: OpenAI's Web-Interacting Agent

• Proxy: Convergence AI Assistant

• Other Agents Incoming

Let’s dive in.

Prediction Fulfilled: General Agents in Q1, 2025

If you’ve been a reader of Bionic Business for a while, you know that I’ve repeatedly predicted these Agents becoming real in Q1 2025.

The code frameworks have been around for years now.

You can achieve similar, small-scale functionality if you can build them, or with agentic automation workflows, tool use, and so on.

Now?

If you don’t want to build them (or pay someone to do so), you can sign up for them.

Right now, as I’m publishing this, Manus is still on a waitlist for access.

I’ve had access since they launched and clients of mine have gotten theirs (took them a few days to get access). If I were you, I’d sign up for the list. No, I’m not getting paid a single penny to recommend this. 

Others, like Proxy (from Convergence) are available for $20/month (funny how OpenAI set this price and everyone has basically followed).

Operator is, of course, available if you’re paying them.

Anyway, let me give you an update on where we’re at with Agents like these.

The Big Picture

• Manus, Operator, and Proxy are the latest general AI agents, released in early 2025, with capabilities for autonomous task execution.

• Manus, launched March 6, 2025, is from China and claims to be the first fully autonomous AI, handling complex tasks like booking holidays.

• Operator, released January 23, 2025, by OpenAI, performs web tasks like booking tickets, available to US ChatGPT Pro users.

• Proxy, from Convergence AI, launched in January 2025, is customizable for various tasks, with basic access free and a Pro Tier for $20/month.

General AI agents, like Manus, are advanced AI systems that can perform a wide range of tasks autonomously, adapting to new situations without constant human input. 

Over the past 90 days or so, three notable agents stand out: Manus, Operator, and Proxy. 

These agents represent the cutting edge of AI right now, with capabilities that could change how you operate your business.

And yes, these are steps toward an autonomous business.

(See, I told you it was possible—and I’ve had my fair share of haters, doomers, and skeptics tell me I’m wrong. I appreciate their point of view, and I listen carefully. But I can’t deny what my applied AI studio has created and what a lot of other people are building. Because it’s, quite literally, happening).

Let’s get into them.

Manus: The Autonomous Pioneer

Manus, developed by China's Butterfly Effect and launched on March 6, 2025, is billed as the world's first fully autonomous AI agent. 

It can handle complex, real-world tasks such as booking holidays, buying property, and creating websites without human guidance. 

It uses multiple AI models, including Anthropic's Claude and Alibaba's Qwen, to execute these tasks, showcasing high autonomy and versatility.

Capabilities: Manus is claimed to be the world's first fully autonomous AI agent, capable of performing complex, real-world tasks without human guidance. 

Examples include booking holidays, buying property, creating websites, generating reports, and screening job candidates. 

It can autonomously browse websites, leverage various capabilities, and display its workflow in real-time, as demonstrated in launch videos.

So, basic functionalities:

• A sophisticated multi-agent architecture functioning like an executive managing specialized sub-agents.

• Cloud-based asynchronous operation that works in the background while notifying users only when outcomes are ready.

• Impressive demonstrations including creating biographies, building websites, and solving complex problems independently.

• A "Computer" window displaying real-time agent decision-making processes.

• Integration with various AI models including advanced language models.

Technical Foundation: Unlike single-model AI chatbots, Manus operates using multiple AI models, including Anthropic's Claude 3.5 Sonnet and fine-tuned versions of Alibaba's Qwen, enabling it to handle multi-step workflows through specialized sub-agents. 

Business Use Cases: So, what could you use Manus for?

A lot, as it turns out. A few examples below.

For SaaS Companies:

• Data Analysis & Reporting: Manus can autonomously analyze user behavior patterns, generate insights, and create actionable reports without human intervention.

• Financial Analysis: Capable of conducting thorough market analysis and creating financial forecasts for SaaS pricing strategies and expansion planning.

• Documentation Generation: Automatically creates technical documentation, help resources, and knowledge base content with minimal oversight.

For Agencies & Consultants:

• Market Research: Independently researches competitors, industry trends, and emerging technologies to inform client strategies.

• Content Creation: Generates high-quality marketing content, case studies, and industry analyses tailored to client needs.

• Client Reporting: Automates the creation of comprehensive client reports by aggregating data from multiple sources and extracting actionable insights.

For Media & Publishers:

• Content Development: Creates various forms of content including articles, newsletters, and technical manuals with minimal supervision.

• Image Analysis: Analyzes visual content for marketing applications and audience engagement strategies.

• Editorial Assistance: Provides fact-checking, research support, and content recommendations based on audience data.

For Freelancers:

• Workflow Automation: Streamlines administrative tasks to allow freelancers to focus on billable work.

• Client Research: Conducts in-depth research on potential clients and their industries to inform pitches and proposals.

• Content Creation Support: Assists with drafting, research, and editing across multiple content formats.

Operator: OpenAI's Web-Interacting Agent

Operator, released by OpenAI on January 23, 2025, is designed to perform tasks on the web, such as booking concert tickets or filling out forms. 

Operator can interact with web pages like a human, typing, clicking, and scrolling. It's currently available to ChatGPT Pro subscribers in the US, with plans for broader rollout.

Capabilities: Operator is designed to perform tasks on the web, such as booking concert tickets, filling out forms, and shopping online. 

It uses its own browser to interact with web pages, typing, clicking, and scrolling like a human. 

Powered by the Computer-Using Agent (CUA) model, built on top of GPT-4o, it combines vision capabilities with advanced reasoning to execute multi-step tasks autonomously.

A handful of use cases:

• Fill out forms, and make reservations independently.

• Open a browser and interact with websites just as a human would.

• Process natural language requests like "Find tickets for the next concert at the Sphere".

• Include safeguards such as "takeover mode" for sensitive information and confirmation requests before finalizing transactions.

Use Cases: It supports categories like shopping, delivery, dining, and travel, automating tasks such as planning vacations or making restaurant reservations. 

It sometimes fails at this, but overall, it works pretty well.

Users can monitor its actions through a dedicated window, ensuring transparency, and it can self-correct if it encounters issues.

Even though it’s made for general-purpose use, you can still use it for a ton of business tasks.

Whatever you’re doing for your business online, you can have Operator do some of it.

Business Use Cases: A few ideas for you below on what you can do with Operator.

For SaaS Companies & Agencies:

• Customer Support Automation: Handles refund requests, reservations, and order modifications without human intervention.

• Competitive Analysis: Monitors competitor pricing, availability, and trends in real-time across multiple platforms.

• Multi-platform Research: Gathers information across websites using visual recognition rather than APIs, enabling broader data collection.

For Freelancers & Consultants:

• Automated Job Applications: Scans job boards and submits applications based on freelancer qualifications and preferences.

• Client Research: Gathers background information on potential clients across multiple platforms to prepare for pitches and meetings.

• Administrative Tasks: Manages bookings, scheduling, and communications across platforms that lack formal integration options.

For Media Organizations & Publishers:

• Content Research: Gathers information from multiple sources simultaneously, creating comprehensive research briefs.

• Distribution Management: Handles content submission across platforms with varying interfaces and requirements.

• Audience Monitoring: Tracks engagement metrics across platforms without needing separate API access for each.

Proxy: Convergence AI Assistant

Proxy, from Convergence AI, was launched in January 2025 and serves as an AI assistant capable of performing tasks autonomously, such as navigating the web and filling out forms. 

It is customizable and has automation built-in, allowing users to set and forget key tasks. 

Basic access is free, with a Pro Tier available for $20 a month, offering unlimited sessions and enhanced features.

It’s good but not as robust as Manus, right now anyway.

Capabilities: Proxy is an AI assistant designed to perform tasks autonomously, navigating the web, filling forms, and handling various tasks. 

It is customizable and has automation built-in, allowing users to set and forget key tasks, such as scheduling or email management. 

It can handle multiple tasks simultaneously and learn from user instructions, positioning itself as a generalist extension of the user.

You’re looking at things like:

• Takes control of your mouse and keyboard to complete web-based tasks on your behalf.

• Handles everything from shopping research to travel booking while requiring only final approval from users.

• Allows users to submit multiple prompts concurrently without waiting for previous tasks to complete.

• Provides a streamlined interface for monitoring agent progress.

Access and Pricing: Basic access is free, offering 5 sessions daily with 2 parallel sessions and one automation. 

The Pro Tier, at $20 a month, provides unlimited sessions, 5 in parallel, and 20 automations, enhancing its utility for both personal and professional use.

Market Fit: Proxy is noted for beating competitors like OpenAI’s Operator in benchmarks, suggesting strong performance in consumer-friendly, out-of-the-box readiness.

Even though it’s billed as general use, for the general public, you can still use it for business tasks.

Business Use Cases: Similar to Manus and Operator, plenty of things you can have it do.

(All you do is chat with it and make your requests—no automation needed to set things up).

For SaaS Companies:

• Cross-platform Workflow Management: Navigates and operates across multiple systems including Salesforce, G Suite, Slack, and specialized platforms.

• Customer Feedback Aggregation: Collects, analyzes, and summarizes customer feedback from multiple channels.

• Code Review Automation: Streamlines development workflows by handling routine code reviews.

For Agencies & Consultants:

• Multi-client Management: Maintains separate workflows and processes for different clients while learning client-specific preferences.

• Marketing Campaign Monitoring: Tracks performance metrics across multiple platforms and generates comprehensive reports.

• Automated Research: Conducts targeted research to inform client strategies and recommendations.

For Media & Publishers:

• Content Aggregation: Gathers information from multiple sources for research and reporting.

• Document Summarization: Creates concise summaries of lengthy documents and research papers.

• Publishing Workflow Management: Coordinates content across multiple platforms and publication channels.

For Freelancers:

• Task Automation: Manages repetitive tasks like email management, scheduling, and basic client communications.

• Job Opportunity Research: Scans job sites for relevant opportunities (though with some limitations on volume).

• Administrative Support: Handles invoicing, contract management, and client communications across platforms.

Let’s Compare These Agents

While each platform has unique strengths, they share common benefits in automating repetitive tasks, productivity, and helping you focus on strategic initiatives rather than routine operations.

A few key differentiators include:

• Manus: Excels in autonomous execution of complex tasks with minimal guidance.

• Operator: Specializes in navigating web interfaces without requiring API access.

• Proxy: Focuses on learning and adapting to specific workflows with long-term memory.

Other Agents Incoming

There is lots more cooking, including specialized agents (for example, I know of at least 8 potential agents made for online marketing, or email marketing, social media, etc.)

Frankly, I think there’s more value in a specialized agent than general-purpose agents. Manus, Operator, and Proxy are very useful—but they’re made for two markets at the same time: consumer and business. 

Anyway, some other agents to look for below.

NVIDIA Agents at GTC 2025

NVIDIA continues to position itself as a leader in Agentic AI, focusing on infrastructure, models, and development tools to enable enterprises to build reliable, efficient, and scalable AI agents.

They want to be the agent operational foundation that you build Agents on top of and with. It’s a smart move—selling picks and shovels to the golddiggers. 

Llama Nemotron Family of Models:

• NVIDIA launched the Llama Nemotron family, a series of open reasoning-focused large language models (LLMs) built on Meta’s Llama 3, enhanced with NVIDIA’s post-training techniques for improved reasoning capabilities.

• These models provide a foundation for creating advanced AI agents that can operate independently or as connected teams to handle complex, multi-step tasks.

NVIDIA AI-Q Blueprint:

• A new open-source framework for connecting AI agents to large-scale enterprise data and tools, enabling agents to query multiple data types (text, images, video) and leverage external tools like web search and other agents.

• Provides observability and transparency into agent activity, reducing complex task completion times from hours to minutes.

• Described as supporting "digital employees" by enabling collaboration between agents and optimizing performance across different agentic frameworks (e.g., Salesforce’s Agentforce, Atlassian Rovo, ServiceNow AI platform).

Salesforce Agentforce 2dx

Released on March 18, 2025, Agentforce 2dx expands Salesforce's digital labor platform with new tools for deploying autonomous AI agents in enterprise businesses. 

Key features include:

• Tools for both low-code and pro-code developers to configure, test, and deploy agents.

• Advanced analytics for monitoring and improving agent performance using real-time data.

• Agentforce API for backend integration and Invocable Actions for embedding agents in Salesforce business logic.

• MuleSoft integration for creating agent topics and actions using natural language.

• Agentforce Steps in Slack Workflow Builder for no-code automation.

Anthropic's Claude with Computer Use

Anthropic has enhanced Claude with new "Computer use" capabilities, allowing the model to interact with desktop environments in a human-like manner. 

The latest version, optimized for Claude 3.7 Sonnet, includes:

• Tools that enable Claude to manipulate a computer by moving a cursor, clicking buttons, and typing text.

• A multi-agent loop architecture where Claude requests tool actions, applications execute them, and results return to Claude.

• Enhanced thinking capabilities in Claude 3.7 Sonnet to provide insight into the model's reasoning process.

OpenAI's Agent-Building Tools

On March 11, 2025, OpenAI launched a comprehensive set of tools designed specifically for building, deploying, and scaling AI agents. 

This release includes:

• A new Responses API to streamline agent building.

• Built-in tools for web search, file search, and computer use.

• A new Agents SDK for orchestrating both single-agent and multi-agent workflows.

• Integrated observability tools for tracing and inspecting agent development.

Lots more in the works.

But you should get started with the 3 above, as they can do quite a lot for your business.

What questions do you have about agents? Or agentic workflows?

What problems are you trying to solve? 

What kind of analysis would help you?

What are you trying to get done—this week—that perhaps AI could do for you?

Reply, let me know, so I can help you.

I can’t help you if I don’t know what you need.

Tell me a little bit about your business, what you sell and do, and then answer the questions above.

Oh, also, before I forget:

Cortex is open again for new subscribers.  

Go here now to sign-up if you’re interested.

As a reminder, with Cortex, every month you get:

• 2-3x full, unrestricted email issues on various AI news, tools, automations, use cases.

• 1x Signals issue, focused on strategic use of AI in your business.

• 1x Circuits issue, focused on tactical, how-to implementation in your business.

Together, they’ll help you make moves in your business to go autonomous—and scale with AI.

Cortex is open for a few days, get in now if you’re interested.

Talk soon,
Sam Woods
The Editor