Don’t Sleep On Genspark’s Super Agent

Genspark is better than the much-hyped Manus, yet it has sailed under the radar.

Yet another Thursday post in the “Hot Takes” format.

TL;DR

Genspark’s Super Agent is a genuinely competent general agent that reasons and performs complex tasks, but it’s been overlooked in the avalanche of agent hype.

What is it?

Last week, Genspark announced its “fast and reliable” Super Agent:

That video currently sits at a criminally low 32K views compared to Manus AI’s 700K+, certainly not helped by the monotonous presentation accompanied by a generic 1990s TV infomercial soundtrack.

But under the hood, Super Agent is surprisingly effective at orchestrating dozens of moving parts to complete complex tasks.

Super Agent appears to be powered by a reasoning model—perhaps o1 or o3-mini—which processes your request, then calls on Genspark’s specialized agents, LLMs, image models, video models, and other tools based on the task.

Source: MainFunc

Among other things, Super Agent can use:

• Image models to make visuals

• Video models to create short clips

• Text-to-speech models to generate voices

• Coding skills to write functioning code

• Deep research agents for, uh, deep research

• …and lots of other stuff.

It also has a “Call for me” option that uses a voice agent to call up and talk to real people and businesses on your behalf, but I haven’t tested this one myself.¹

Super Agent beats both Manus AI and OpenAI Deep Research on the GAIA Benchmark for general assistants:

Source: MainFunc

Let’s be honest, though: We’ve been inundated with such benchmarks lately.

These days, my eyes usually glaze over when I see another comparison chart, which helps explain why the announcement got only a bland mention in my last roundup.

But having now used Super Agent for several real-world tasks, I can say that this is the first time I’ve been impressed by a general AI agent.

In the past week, Genspark Super Agent:

• Helped me research, consolidate, and code the GPT-4o swipe file for my paid subscribers.

• Found a fast charger for my phone in a Danish online store. It nailed this task, arriving at the same recommendation I did after longer manual research. (Read the chat.)

• One-shot coded a landing page² based on a simple request, done as a top-level test of its capabilities. (Read the chat.)

• …and did equally well in several other minor test tasks.

In short, it just works!

Genspark lists 11 real-world examples, and I’d say they’re representative of Super Agent’s actual capabilities.

How do you use it?

It’s stupid simple:

1. Go to genspark.ai (sign up for a free account if needed).

2. Type your task or request into the big box on the front page:

   

3. Watch Super Agent break down its thinking as it works through your task.

   

At each step, you can click the “View” button to expand the details and see the agent’s thinking.

Once the initial task is finished, you can request changes via chat.

Here’s Super Agent’s one-take output for the above request:

Genspark gives you 200 daily credits for free, which is decent for testing text-based and coding tasks but probably won’t be enough for stuff that requires more advanced third-party models.

Why should you care?

AI agents have been hyped relentlessly since 2023 when BabyAGI and AutoGPT were making a splash.

But the gap between flashy hypothetical demos and agents that actually work has been hard to close.

In October 2024, we started getting first glimpses of functioning general agents³ when Anthropic launched “Computer Use.”

Soon, we had OpenAI’s “Operator” in January 2025, Convergence AI’s “Proxy” in February, and finally “Manus” in March.

But they all came with caveats like waitlists (Manus), invite-only or expensive research previews (OpenAI), the need to install and run clunky local scripts (Anthropic), and so on. Proxy is excellent, but it’s primarily confined to web browsing tasks.

To me, Genspark’s Super Agent is the first general agent that runs entirely inside a virtual environment, doesn’t require access to your computer, can reason and take many diverse actions, and is readily accessible by everyone.

It’s one of those rare moments in AI where reality exceeds the marketing hype.

To be sure, Super Agent is not immune to the usual AI blind posts like hallucinations, imperfect web browsing, and so on. At the same time, I find that it can usually fix things based on simple feedback without messing up something else in the process.

The best part? You don’t have to take my word for any of this.

Just sign up for a free account and take Genspark’s Super Agent for a quick spin.

You might be as positively surprised as I was.

🫵 Over to you…

Have you tried Genspark Super Agent? What were your impressions? Have you had the chance to compare it to Manus or other AI agents?

I’d love to hear your thoughts about it.

Leave a comment or drop me a line at whytryai@substack.com.

---

Thanks for reading!

If you enjoy my writing, here’s how you can help:

• ❤️Like this post if it resonates with you.

• 🔄Share it to help others discover this newsletter.

• 🗣️Comment below—I love hearing your opinions.

Why Try AI is a passion project, and I’m grateful to those who help keep it going. If you’d like to support my work and unlock cool perks, consider a paid subscription:

1. For now, “Call for me” is only available in the US and Japan.

2. You can nitpick over the crappy brand image made by Flux or minor link issues, but all of this is easily fixable with a single round of feedback.

3. We also had more specialized “Deep Research” agents from several competitors.