[AI SPRINT] $20,000 AI Agents, and is Manus the Best AI Yet?

Last week, OpenAI’s Sam Altman told investors they were working on AI agents at price points up to $20,000 per month! While details were scarce, this announcement raises important questions about what these AI agents could mean for the future of business.

During a discussion with investors, where Altman also mentioned OpenAI is currently losing $5 billion per year, he outlined three tiers of AI agents under development:

• A “high-income knowledge worker” AI for $2,000 per month

• A software developer AI for $10,000 per month

• A PhD-level scientific researcher for $20,000 per month

When most businesses aren’t even giving their employees ChatGPT for $30 a month, $2,000, $10,000, or $20,000 a month seems absolutely crazy to many.

But it isn’t.

To grasp this shift, you need to understand that AI evangelists believe the future of AI is that companies will essentially outsource their knowledge work to AI agents. The promise is that AI agents are super-smart, don’t complain, work 24/7, don’t require desks, benefits, or team-building activities, and are infinitely scalable. If that comes to pass, AI agents could become a very compelling way to reduce costs and increase services in a company.

To understand where AI agents stand today, it’s helpful to compare their cost savings against traditional human roles. The numbers show a clear trend: for repetitive tasks, AI agents are dramatically cheaper—but for expert roles, the gap narrows.

The first three roles show roughly an 85% cost savings compared to human workers, but this likely reflects the current limitations of AI agents rather than their long-term value. Today, they primarily handle routine, automated tasks with less overall impact than human employees.

For example, AI sales agents handle basic outreach, similar to an entry-level rep reading from a script. But soon, they’ll go much further—using natural voice conversations to engage customers (have you tried Boardy?), handling objections, and even closing smaller deals without human input. They will know your product inside and out, trained on your sales and marketing literature and entire product catalog. They will be able to automatically create presentations and proposals for customers and will likely close deals with zero human involvement within the next 12–18 months. Think of it like having a highly trained virtual sales team working around the clock, and at that point, I expect they’ll cost a lot more than $500 per month.

At first glance, OpenAI’s higher-priced agents may seem expensive. But unlike sales and admin agents, which automate routine tasks, AI developers and researchers generate value through complex problem-solving and innovation. Their cost is closer to human wages because they’re designed to perform at an expert level—potentially replacing multiple employees in the same role.

In that scenario, I would expect their cost to move much closer to human levels, probably around that 20% discount to humans.

The other factor to consider for software developers and researchers is that their value is likely significantly higher than human employees because they generate value through innovation. Beyond closing deals, they will identify opportunities for innovation and autonomously deploy solutions.

Imagine you’re running a business, and your AI software developer spots a critical issue—a supplier’s order didn’t process. Instead of waiting for a human to notice, it instantly diagnoses the issue, fixes the error, and completes the order—all before you even know there was a problem.

And it goes further than that—AI agents are already creating massive value in pharmaceutical and scientific research, identifying new methods and ideas that may be worth millions or even billions of dollars. Just last week, an AI identified what may be a better and cheaper version of the popular Ozempic.

With these agents running 24/7 and scaling to handle many tasks at once, a single AI software developer or researcher will likely be equivalent to multiple human employees in the same role. That’s why I don’t think these costs are unreasonable. We just aren’t there yet, but I’m confident we will be in a surprisingly short time.

What Should You Be Doing About It?

You may not like this future, but it is the reality, and it pays to start preparing for it. If you are a worker in one of these roles and are worried about being replaced, your best move right now is to learn AI and position yourself as a leader in how to use it.

AI agents will still require management, and if you are the most knowledgeable about AI in your function, you might be the one helping to find, implement, and monitor your future AI coworkers. Your job will shift toward setting goals for AI workers, monitoring them, and improving them—an increasingly valuable skill.

If you are a business leader: the greatest risk in business today isn’t AI taking your job—it’s someone who knows how to use AI taking your business.

Becoming an AI-enabled organization is not just about purchasing AI agents. It requires reassessing your position in the market, understanding AI’s impact on your industry, upskilling your workforce, integrating AI tools and agents, and transforming your company to stay competitive as AI reshapes the landscape. It’s a big task, but the most important thing is to start and make progress every month.

That’s the point of the AI SPRINT newsletter—every month, you need to “sprint” to implement AI, learning more and adopting it further and further into your organization. And I’ve even given you a project plan for that: the AI SPRINT Framework. You can review the past newsletters for a review of each step, but just keep working to build more AI capabilities every month! Then you will be ahead of your competition:

Now on to Manus AI: the latest Chinese AI Disruption

Manus AI (https://manus.ai) burst onto the scene last week with its launch announcement on X. Its AI agent combines multiple discrete AI capabilities: a high-quality reasoning agent powered by Anthropic’s Claude, and a highly capable computer operational sandbox. This combination appears to significantly outperform current operator products from OpenAI and Anthropic, delivering faster performance, greater capabilities, a better user interface, and public file sharing.

What sets Manus AI apart is its seamless integration of multiple AI capabilities, enabling it to operate fully autonomously. Unlike typical AI assistants, which require users to prompt them step by step, Manus AI can take a goal—like conducting a financial analysis, creating a research report, or potentially automating a hiring process—and complete it end-to-end with minimal input. You monitor and instruct the AI as it completes the work.

This provides the clearest glimpse yet of what AI agents will soon be capable of—and why they will be so disruptive to knowledge work.

As another small Chinese startup, Manus AI has parallels to DeepSeek’s recent announcement, further fueling the debate over whether China has caught up in the global AI race. Manus AI’s performance in benchmarks like GAIA—which tests AI agents on real-world problem-solving—suggests that it is ahead of OpenAI’s Deep Research, at least in planning and task execution.

With limited access, demand has skyrocketed—some invitation codes are even being resold for as much as $1,000. That level of market enthusiasm is rare, even in the AI space, and indicates how much anticipation is surrounding this technology.

However, there are also concerns. As with other Chinese AI products, questions about data privacy and security have surfaced. While the hype around Manus AI is undeniable, access remains highly restricted, making it difficult to assess its real-world performance. And US-based AI companies will likely replicate its capabilities soon.

Unfortunately, I don’t yet have access to Manus AI to test it myself. But from what I’ve seen in public demos, it looks incredibly compelling. Here are a few examples you might want to check out—watch the recording first, then review the final results:

1. Tesla Stock Analysis Recording and Final Results

2. Creation of a Middle-School Physics Lesson Application and Final Results

3. B2B Prospect Research Report and Final Results

You can view plenty of other examples at manus.im/usecases.

Closing Thoughts

Manus AI is a significant step forward for AI agents, and I expect OpenAI, Anthropic, and others to replicate its approach quickly. If you want a glimpse into where AI agents will be in business over the next 12 months, look no further than what Manus is demonstrating today.

With AI agents now in action—and some priced at up to $20,000 a month—the real question is: how will you adapt your business to leverage them?

Did someone forward this newsletter to you? If you're not already signed up, you can subscribe to AI SPRINT™ for free here.