I quickly tested Manus AI: I guess it has potential

When I mentioned Manus AI almost four weeks ago (at number 8), I wasn’t thrilled by the business model because it lacked a free tier. Now it sort of almost have one, if you’re lucky enough to be able to issue tasks that take 300 credits or fewer—because this is the free daily allowance they have just created.

What changed

From the notifications received from Manus AI:

May 11, 2025 — Introducing additional free access to Manus

Starting today, we’re launching additional access to Manus!

• Available to all with no waitlist
• One free daily task for all users (300 credits)
• A one-time bonus of 1,000 credits for all users

More value, more flexibility. Enjoy your free credits and invite your friends to try it out!

Then.:

May 16, 2025 — Manus image generation

Manus doesn’t just generate images. It understands your intent, plans a solution, and knows how to effectively use image generation along with other tools to accomplish your task.

The economics of Manus AI

The paid plans have expanded with a cheaper option, but let’s talk of the freebies:

• New users receive a one-off 1,000 credits
• Early adopters received an extra 1,000 credits
• “Refer a friend, get 500 credits each”: I noticed I only have 2 referral codes I could use
• Everyone, free or paid, receive 300 credits per day that get lost if not used
• Everyone who gives a rating to a performed task receives 100 credits
• The cost of a task is not fixed, but most tasks require between 450 and 900 credits (they claim simple tasks can cost as low as 200 credits, but good look with that)

I believe that the problem is obvious. Once the welcome credits are consumed, the free tier only offers 300 credits per day. Supposing you can imagine useful tasks that require less than 300 credits, you can get 100 back when you rate the answer, so the next day you can count on 400 credits (of which 300 would expire if not used). Only through repeatedly managing to run “cheap” tasks and accumulating the 100 credits for rating (as long as this mechanism still works) would one be able to run more “expensive” tasks once in a while. Not great.

On the paying side:

• Basic: $19/month ($16 if paid annually) get you 1,900 credits/month
• Plus: $39/month ($33 if paid annually) get you 3,900 credits/month
• Pro: $199/month ($166 if paid annually) get you 19,900 credits/month

Still anything but great!

Use cases: not your usual chatbot!

If it wasn’t already obvious to you, let me state it outright: Manus AI is not yet another chatbot that you ask this and that in a browser or in an app, or that you ask for occasional help with coding. As you should have gathered from Introducing Manus: The General AI Agent, Manus AI is an AI agent meant for complex tasks. And complex tasks are expensive. Think of Codex, only universal—for all kinds of tasks!

I nonetheless wanted to give it a try, now that they lure people with those 300 credits a day. Maybe I’ll find it useful in the end. In the free tier, mind you.

Caveat emptor: Manus AI accepts tasks, which are not conversations. You cannot come up with a teeny tiny follow-up question, “OK, but how about this?” or “You forgot about that.” Nope.

This shit is less and less attractive to me. For my occasional recourse to intoxicated AI robots, I use:

• For complex questions: Grok and ChatGPT.
• For simple, short answers: Claude and Copilot.
• On occasions: Mistral, DeepSeek.
• When I find it appropriate: Grok DeepSearch and DeeperSearch, Grok Think, Copilot Think Deeper, DeepSeek DeepThink, ChatGPT Think.
• For testing purposes: Qwen3.
• Rarely: Gemini (Flash and Pro);
• Almost never: Gemini Deep Research, ChatGPT Research.
• Never again: Perplexity, including Perplexity Research.

Where could I fit Manus AI into this?

Task #1: Understanding the clawback tax on drugs

There is a thing called “clawback tax” that is not the generic clawback agreement (Investopedia, Wikipedia English). It’s the stupid mechanism that applies to drug manufacturers in some European countries, and Wikipedia Romanian gives an extremely succinct and clumsy definition:

The clawback tax requires drug manufacturers – not distributors or pharmacies – to return to the Romanian state a portion of the profit made on sales of compensated medicines that exceed the amount allocated to them from the National Health Insurance Fund.

Clawback has a complicated definition: it is a 70 percent tax on any amount exceeding the volume of sales of medicines for the year 2009, plus a tax of between 5% and 12% on the total volume of sales.

The drug price cut applied since 2015, after an order that was supposed to take effect in 2009, was the move that drugmakers pay up to 25% of turnover for clawback.

It doesn’t make sense, right? I mean, are we in Communism or something? But don’t worry: it’s even worse in the United Kingdom! At least, this is what Manus AI told me.

Now, I’ve been trying for ages to understand the mechanism of this crap. Or, at least, the rationale. How would this benefit anyone? What’s the ethical part of this? It’s not a “sin tax,” right? One of the effects is that generic drugs or more generally cheaper drugs tend to disappear from the market, and it seems that the clawback tax is responsible for that. No wonder that communist measures destroy the market!

This clawback tax on drugs is not a Romanian invention. I seem to remember that Romania borrowed the idea from Hungary, and both countries later had issues with the justice because they somehow managed to apply this tax even to the VAT on drugs. Boehringer Ingelheim brought Hungary to the European Court of Justice (CURIA, EUR-Lex). I don’t remember at what level was settled the case for Romania.

Strange enough, as you’ll see, Manus AI doesn’t even mention Hungary in its report!

🤖 Here’s the task I gave Manus AI: “I don’t understand the clawback tax on drugs in Romania. What other countries have it? What is the rationale? Is it used for other products too elsewhere? How does it affect the generics?”

You can:

• Replay the task or read the result: Clawback Tax on Drugs in Romania and Global Impacts
• Read the online report: Understanding the Clawback Tax on Drugs

I was disappointed, and not only by the contents.

Reading the resulting Markdown documents from within the task’s view is pretty much OK. I’ll put here a screenshot as it would be seen by you, without the sidebar that I see as the owner of the task:

But the online report, for some reason, has all the bulleted lists centered, no matter what browser or OS I use, which is grotesque:

And I still couldn’t understand a bit of this shit. What are they trying to do? How is anyone still selling drugs in Europe?

Beyond Romania, the United Kingdom and Greece get special treatment, and Italy, France, Portugal, Belgium, Germany, and Croatia are briefly treated. But no Hungary. And, for fuck’s sake, are there any countries outside Europe that do that? In Asia, Africa, or Latin America? No fucking clue, and I cannot come with a follow-up question!

Quick takeaway (why “rebates” when it’s a fucking tax?!):

Cross-Country Comparison

• Across Europe, rebates are typically around 10%, with Germany being slightly higher at approximately 12%
• Romania, Greece, and the UK currently have some of the highest rates (23-27%)
• Belgium has one of the lowest capped rates at 4%
• Different terminology is used across countries, making direct comparisons challenging

It’s a shame to be European.

This task cost me 547 credits:

Being disappointed, I rated the result with 2 stars out of 5. Because it sucks big time.

Task #2: Use of “speak to,” “speak with,” “talk to,” and “talk with”

Later in the day, I asked Grok a question about the use of the verbs “to speak (to/with)” and “to talk (to)” in AmEn and BrEn, and the answers (because there were many follow-up questions and replies) didn’t satisfy me. The entire chat was in Romanian, but if you’re curious, it’s here. I’m not sure that an automated translation would make much sense.

And the idea came to me to ask Manus AI to dig up further.

🤖 To Manus AI:

Could you analyze the subtitles of 20-30 American movies of the ’60s to ’80s to determine the incidence and circumstances in which “speak to,” “speak with,” “talk to,” and maybe even “talk with” are used, in order to find out if my perception is correct that:
— “speak to” was preferred to describe 1-way speech, despite being perceived as more formal;
— “talk to” was preferred to describe 2-way speech (conversation, dialogue), even if it feels more casual;
— “speak with” became more common in the ’80s, also to suggest a 2-way speech;
— “talk with” wasn’t quite a thing yet.
Maybe some British films too should be analyzed to see how much the usage was different on this side of the pond.

You can:

• Replay the task or read the result: Analyzing Speech Phrases in American Movies 1960s-1980s
• Read the online report: Film Language Analysis: “Speak to,” “Speak with,” “Talk to,” and “Talk with” in American and British Films (1960s-1980s)

Funny thing, my assumptions seem to have been correct, and even the last one, which was declared as only “PARTIALLY VALIDATED,” was actually confirmed!

At least, if we trust what Manus AI did.

Executive Summary

This report presents a comprehensive analysis of how the phrases “speak to,” “speak with,” “talk to,” and “talk with” were used in American and British films from the 1960s through the 1980s. Our findings largely validate the hypothesized patterns:

1. “Speak to” was indeed preferred for one-way communication despite its formal connotation
2. “Talk to” was the dominant choice for two-way communication despite its casual nature
3. “Speak with” showed a significant increase in usage during the 1980s, particularly for formal two-way communication
4. “Talk with” was not yet fully established during this period, though it was beginning to emerge

These patterns reflect broader sociolinguistic trends during this period, including shifts toward more collaborative communication styles and the gradual informalization of professional discourse.

The “Key Findings” section includes a table that is broken in the final web report. From the Markdown:

Key Findings

1. Phrase Usage by Communication Direction

Phrase	Primary Direction	Notable Characteristics
“Speak to”	One-way	Formal, often hierarchical, used for addressing or announcing
“Talk to”	Two-way	Casual, versatile, used for dialogue and conversation
“Speak with”	Two-way	Moderately formal, collaborative, professional
“Talk with”	Two-way	Casual to moderate, emphasizing collaboration and equality

2. Regional Differences

American Films:

• Showed more dramatic evolution across decades
• Led the shift toward increased usage of “speak with” in the 1980s
• Demonstrated greater flexibility in phrase selection by the 1980s
• Earlier adoption of “talk with” as a legitimate alternative

British Films:

• Maintained more consistent usage patterns across all three decades
• Preserved stronger formality distinctions between phrases
• Showed slower adoption of “speak with” as an alternative to “speak to”
• “Talk with” remained relatively uncommon even in the 1980s

3. Decade-Based Trends

1960s:

• Clear distinction between formal and informal communication
• “Speak to” dominant for formal, one-way communication
• “Talk to” common for casual interactions
• “Speak with” and “talk with” relatively uncommon

1970s:

• Beginning of more flexible usage patterns
• “Speak to” remained common in formal contexts
• “Talk to” continued as the preferred casual form
• “Speak with” began to increase in frequency
• “Talk with” showed slight increase

1980s:

• “Speak to” maintained its formal usage but became more restricted to truly one-way scenarios
• “Talk to” remained dominant for casual communication
• “Speak with” showed notable increase, especially in American films
• “Talk with” became more established, though still less common than other forms

4. Hypothesis Validation

All four original hypotheses were either validated or partially validated by our analysis:

• Hypothesis 1 (VALIDATED): “Speak to” was preferred for one-way speech despite its formality
• Hypothesis 2 (VALIDATED): “Talk to” was preferred for two-way speech despite its casual nature
• Hypothesis 3 (VALIDATED): “Speak with” became more common in the ’80s for two-way speech
• Hypothesis 4 (PARTIALLY VALIDATED): While “talk with” wasn’t fully established, it was beginning to emerge by the 1980s

The 25 American films and 8 British films used in the study:

American Films (1960s-1980s)

1960s

1. The Graduate (1967)
2. Psycho (1960)
3. Breakfast at Tiffany’s (1961)
4. 2001: A Space Odyssey (1968)
5. Bonnie and Clyde (1967)
6. The Good, the Bad and the Ugly (1966)
7. Easy Rider (1969)
8. Planet of the Apes (1968)

1970s

1. The Godfather (1972)
2. Taxi Driver (1976)
3. Star Wars (1977)
4. Jaws (1975)
5. The Exorcist (1973)
6. One Flew Over the Cuckoo’s Nest (1975)
7. Rocky (1976)
8. Annie Hall (1977)
9. Apocalypse Now (1979)

1980s

1. The Breakfast Club (1985)
2. E.T. the Extra-Terrestrial (1982)
3. The Terminator (1984)
4. Back to the Future (1985)
5. Blade Runner (1982)
6. The Shining (1980)
7. Raiders of the Lost Ark (1981)
8. Die Hard (1988)

British Films (1960s-1980s)

1960s

1. Lawrence of Arabia (1962)
2. A Hard Day’s Night (1964)
3. Alfie (1966)

1970s

1. A Clockwork Orange (1971)
2. Monty Python and the Holy Grail (1975)

1980s

1. The Elephant Man (1980)
2. Chariots of Fire (1981)
3. A Room with a View (1985)

A nice little chart:

4 CSV files are also provided.

This task, being more complex (finding SRT files, analyzing them, etc.), took 849 credits!

But I rated the result 5 out of 5. It’s not bad.

Conclusions

Is it good? Is it OK? Is it meh-ish? It’s difficult to assess the usefulness of Manus AI for the average Jane or Joe, and even less so after only two tasks. It’s just that it’s expensive!

In one task, it mostly failed. In another one, I’m quite satisfied with the results. UPDATE: Or maybe not so much.

I’m sure Manus AI will find its customers. But it needs to improve, if you ask me.

Maybe I’ll ask it to create a picture for me. I wonder how much it’s going to cost, but also how to define it as a task. There must be something to analyze, to assemble, to use in a more elaborate way. Some other time.