Lynn Raebsamen, CFA

Welcome to 2025, the year of the AI agent—or so the tech hype machine would have you believe. Autonomous agents are being hailed as the next big thing, promising to revolutionize work, streamline daily tasks, and maybe even make your coffee while writing your emails. Every major tech company and startup is racing to build these digital assistants, with claims that they’ll handle everything from booking trips to crafting business strategies. But behind the glossy headlines lies a messy reality: not all agents are created equal. Enter Manus AI, a system that’s making waves for all the wrong reasons.

Manus AI: The Hype vs. Reality

Unlike DeepSeek’s LLMs, which excel at language processing and generation, Manus operates as an autonomous agent, performing tasks like information retrieval, data processing, code execution, and web automation. It’s designed to tackle everything from resume screening to website creation, using a multi-model approach that adapts its reasoning to each task.

Sounds impressive, right? Well, only if you enjoy watching your digital assistant spiral into loops and glitches like a robot trying to solve a Rubik’s Cube blindfolded. Despite the hype, early users have reported confusing answers, frustrating delays, and never-ending loops that make the experience more comedy than convenience.

The Shortcomings of Manus AI

Glitches Galore:

Manus AI’s ability to execute tasks sounds great in theory—until it starts looping endlessly or gets stuck in repetitive cycles. It’s like watching Bill Murray relive Groundhog Day, but instead of waking up to “I Got You Babe,” Manus wakes up to error messages and browser tabs that just won’t close. Testers have reported frustrating delays and failures when asking Manus to handle complex tasks. For example, one user tasked Manus with booking conference tickets, only to watch it scroll through browser tabs like an indecisive shopper at a Black Friday sale. Each loop is a new chance for Manus to get it right, but so far, it seems stuck in its own AI version of Punxsutawney.

Borrowed Brilliance:

Despite its ambitious claims, Manus relies heavily on existing models rather than introducing proprietary innovations. This creation is less a revolutionary breakthrough and more a clever patchwork of existing tech, stitched together with the finesse of a mad scientist on a caffeine binge. Let’s peek under the hood of this “groundbreaking” system:

1. Claude Sonnet: Manus’s poetic soul, borrowed from Anthropic. It’s like giving a Shakespeare-quoting parrot a job as a creative director.
2. Qwen finetunes: Alibaba’s contribution to this AI potluck. It’s as if Manus raided Alibaba’s digital pantry and came out with a gourmet AI casserole.
3. Linux sandbox: The playpen where Manus builds its digital sandcastles. It’s like giving a toddler a beach bucket, but instead of sand, it’s filled with ones and zeros.
4. Web browser control: Manus’s way of surfing the internet like an octopus trying to navigate eight smartphones simultaneously.
5. File system management: The digital equivalent of giving Manus a filing cabinet and watching it play office.

This AI Frankenstein might impress at first glance, but it’s essentially wearing last season’s AI fashion with this year’s accessories. It’s like showing up to a tech conference in your grandfather’s suit, but with a smartwatch and VR goggles.

In the end, Manus is less of a “DeepSeek moment” and more of a “deep sigh” moment. It’s the AI equivalent of reheating leftovers and calling it gourmet cuisine. But hey, in the world of AI, sometimes yesterday’s models with today’s marketing can create tomorrow’s hype.

Security Concerns:

Manus’s ability to interact with external systems and retrieve files has sparked fears about privacy and security risks. Without proper safeguards, an autonomous agent with access to sensitive data could inadvertently create vulnerabilities—a scenario that’s less “helpful assistant” and more “digital liability”.

The Bigger Picture

The hype around AI agents isn’t limited to Manus; it’s part of a larger narrative dominating tech media in 2025. From Salesforce’s Agentforce platform to Nvidia’s predictions about agentic AI reshaping industries, the promises of automation and efficiency are everywhere. Yet experts caution against getting swept up in the excitement. While simple use cases show promise, sophisticated applications still require significant technological maturity. Manus exemplifies this disconnect between ambition and execution. It represents a bold step toward autonomy but struggles with reliability—a powerful concept that remains frustratingly out of reach for most users.

Manus AI may be riding high on the wave of agentic AI hype, but its shortcomings highlight the challenges of building truly autonomous systems. Glitches, over-reliance on existing models, and security concerns suggest that Manus is more “work-in-progress” than “game-changer.” Manus stands as both a cautionary tale and a glimpse into what might be possible—if developers can iron out the kinks. So while 2025 may be the year of the AI agent, it’s clear that some agents still need training wheels—and maybe a good dose of humility before they claim to revolutionize our lives.

The Fun, the Fails, and the Future of AI