Can an SME actually use Manus or Genspark
Yes, and they're better than you think. But the job they do and the job your business needs are usually not the same one. Three questions tell them apart.
The most common question we’ve had this half-year: “can I just use Manus, or Genspark?”
Short answer: yes, and they’re good. But the problem they solve and the problem your company has are usually not the same problem.
What they genuinely do well
Don’t underrate these tools. Ask Manus for “the top twenty frozen food importers in Hong Kong, as a comparison table” and it opens a browser, works through them one by one, and hands you a spreadsheet. Genspark’s AI Slides builds a ten-page deck in about a minute, laid out better than most people manage by hand.
A few years ago that was two days of an intern’s time. Now it’s the length of a coffee.
So the first piece of advice is simple: if you haven’t tried them, try them. The free tiers are enough to tell whether they’re useful to you, and the answer is usually yes.
But producing a thing and running a process are different jobs
For one-off deliverables, a general agent is strong. What actually eats time in a company is rarely one-off — it’s the thing somebody does again every week.
Three questions separate the two.
One: do you have to feed it your own information every time?
A general agent starts every task from a blank page. Your pricing logic, your supplier list, the three contract clauses that killed a deal last year — it knows none of it. You can paste them in each time, but that’s manual work. You haven’t saved the time, you’ve moved it.
A process needs the opposite: the information goes in once and is there every time after.
Two: does the fifth colleague doing this get the same result?
Prompts drift. The same task written by you and by a colleague gives different output; written by the same person today and next month, also different.
For a one-off that’s fine. If the task runs weekly and the output goes to a client, “different every time” is the problem itself. It’s also why we suggest starting from one process rather than from a tool.
Three: after a month, can you say what it cost?
Both are credit-based. A single deep-research run can consume several hundred to a thousand credits, and you can’t see the number before you press go. Fine for personal use. Harder to put in a budget.
How to split it
This is how we split it ourselves, and what we suggest to clients:
- Exploration, one-off research, a deck you need this afternoon — general agent. You’ll enjoy it, and it’s about twenty to twenty-five US dollars a month.
- Weekly, repeatable, goes to a customer, expensive when wrong — make it your own process: your data, fixed steps, a colleague presses one button.
It isn’t either/or. Several of our clients run both, and should.
One aside: notice where your data is sitting
Manus changed hands twice in a year. Meta announced a roughly US$2bn acquisition in December 2025, Chinese regulators ordered it unwound in April 2026, and by June Meta had completed the operational separation, with Manus returning to independent operation.
That isn’t a reason not to use it. It’s a reason to ask, before your customer list and pricing logic go in for the long term: will this be the same company next year?
Your own workflow doesn’t have that question. It’s one of the reasons client data sits on the client’s own server rather than on ours.
