AI Agents Compared: Manus, Lovable, and Gemini for Real Business Tasks

Manus: The Only Agent That Could Do the “Dumb” Work

The task was simple: visit each website, read through it, compile a table, send it over. Routine stuff. But it turned out to be a real filter for AI agents.

Here’s what was tested in practice:

• Manually transferring product cards from competitor websites into a Replit admin panel. Pure copy-paste — but only Manus could pull it off.
• Collecting partner contacts for a rental service. Again, a trivially simple task — and again, only Manus managed it.
• Sending invitation emails to partners.

Manus is expensive, credit-hungry, but when something needs to be done “by hand” — it delivers.

Why Manus Still Requires Constant Supervision

The problem is that Manus loses context easily and can start working in the wrong environment entirely. You can’t leave it unsupervised.

That’s where the biggest disappointment hit. The expectation was that while skiing, it would quietly grind through the boring work. Instead, it constantly demanded attention — asking for confirmation, checking whether it was doing things right, even arguing about the best way to complete a task. Every chair lift ride turned into a back-and-forth with Manus.

Lovable: The Fastest Way to Launch a Hypothesis

Lovable was used for instant starts.

Specific use cases:

• Landing page for a rental service
• Packaging an idea to attract partners
• MVP for initial hypothesis validation

In a single day: landing page, domain, email — and the product is already ready to show. One day, done.

Lovable impressed immediately with its simplicity. But the more time spent working with agents and AI tools in general, the more limiting it started to feel — hard to explain exactly, more of an intuition. There’s also a practical issue: no mobile app, and loading from a browser is painfully slow.

Gemini: The Unexpected Leader in AI Agents for Finance

The most surprising discovery was Gemini from Google.

The specific case: building a financial model for a decorative materials business.

Neither ChatGPT, nor Genspark, nor Manus could handle the financial model. Gemini could:

• Calculate
• Build tables
• Understand the economics
• Design and format the output

At some point, seriously considered switching to it full-time. The only thing that held back was the inconvenience of working from mobile devices. But it’s genuinely top-tier — and definitely something to use more extensively going forward.

The Real Conclusion About AI Agents Comparison and Business Automation

Right now, it’s simply not possible to send an agent off to work and go relax.

Agents require:

• Constant attention
• Task decomposition — and this is a major insight on its own. Break tasks into tiny pieces. Agents choke on large ones.
• Verification at every step
• Understanding of the underlying architecture — you still have to know the fundamentals

They accelerate the start, but they don’t remove responsibility.

For now, the best setup is human + AI, not “AI instead of human.”

In that sense, the experiments are only just beginning. The hope is that next year the ski slopes stay distraction-free, with agents only checked in the evenings — and ideally at a slightly lower cost.

To be continued.