There comes a point, when using artificial intelligence every day, when the problem stops being getting an answer.
The problem is having to always correct it in the same way.
“We don’t use this term.”
“This is not our client.”
“This promise is too strong.”
“The product works like this, not like this.”
The draft arrives in a few seconds. Then the real work begins.
This is also where we should read the passage that Google is preparing from Gems to Gemini Skills. The idea is to make reusable instructions that today we risk rewriting, copying, or reconstructing with each new conversation.
For those who use Gemini sporadically, relatively little changes. For a company that has incorporated it into recurring activities, however, the issue becomes more interesting.
From Gems to Skills: The Difference Is in How You Work
THE Gems allow you to create customized versions of Gemini, configured with instructions and reference materials.
Skills shift the concept onto the task.
They are sets of reusable instructions that Gemini can call up when needed and, most importantly, combine within the same job.
The difference seems subtle until we bring it into a real process.
If we regularly prepare content for a company, we can get guidance regarding tone of voice and how to construct a specific type of document. We don't necessarily need to compress everything into one huge instruction that we have to lug around every time.
It is a potentially more modular management.
But there is one point that risks being overlooked: Making a bad statement reusable means making the error reusable as well.
Before the Skill comes the method.
The corrections you make every week are already useful material
To understand whether this innovation could make sense for the company, I wouldn't start with Gemini.
I would start with the corrections.
Let's take ten pieces of content approved in the last few months and compare them with their respective first drafts. What did we change each time?
We might discover that titles are systematically becoming too promotional. Or that product pages are always missing technical information. That terms we don't typically use commercially are being used. That every newsletter needs to be shortened because the first version tells a lot and says little.
If a correction keeps coming up, there's probably a rule behind it that everyone in the company already knows, but that no one has ever written down.
That rule can become an instruction.
And this is where Skills begin to have a concrete interest.
Time saved is measured by the approved version, not by the speed of the first draft.
A Skill Doesn't Automatically Know Your Company
Google also provides examples of how to transform a blog post into content for different platforms, applying guidelines on tone, format, and working methods.
It's an easy to understand usage.
But if we have a generic article upstream, we will very efficiently obtain five generic contents.
The problem remains.
Instructions can tell Gemini how much to write, what words to avoid, what structure to follow. They can't invent the expertise that should be behind it.
If we're talking about an industrial product, someone needs to provide real-world features, conditions of use, compatibility, and problems encountered by buyers. If we're talking about a service, we need to know where that service ends, what it includes, and what requires a different assessment.
This is information that is difficult to find in a well-written prompt.
We find them by talking to those who sell, produce, install, and assist the customer.
Then AI can help us organize them.
Multiple Skills Together? Useful, Until They Start to Contradict Each Other
The ability to combine different instructions is probably one of the most interesting aspects of the new approach.
And it's also one of those that needs to be tested most carefully.
One skill might require very concise text. Another might require a detailed explanation of every technical feature. Both make sense individually. When used together, you have to decide who's in charge.
It's a much less technological problem than it seems.
It happens already in briefs: marketing wants a short text, sales wants to include all the information, the SEO person asks to develop a theme and in the end someone has to establish the priority.
With Skills, that decision doesn't disappear. It just needs to be made clearer.
And when products, business conditions, or internal procedures change, the instructions must also be revised. Otherwise, the AI will continue to perfectly apply a rule that the company stopped following six months earlier.
Where we would try the Skills first
I wouldn't start with "let's automate marketing.".
It's too broad to tell if anything is working.
It's better to take an activity we know well and that is repeated: a product sheet, the transformation of an article into a social post, a specific sales response, a document always built with the same structure.
Then we need two versions: how we do it today and how we do it with Skills.
Then we can look at what's really happening.
How many corrections are left? Which ones? How long will it take before the material is usable? Are the errors that continue to appear procedurally or do they involve information and decisions that the AI cannot know?
If after three tries we are still rewriting the same things, adding more instructions may not be the answer.
Perhaps the upstream process is not clear enough.
When will the transition from Gems to Skills arrive?
Here it's best to distinguish between what Google has announced and what each company can actually use on their account.
The documentation relating to the transition provides for an automatic migration of Gems to Skills, with different timeframes: the indicated calendar starts from November 2026 for personal accounts and continues in 2027 for Workspace accounts.
However, the information available on access to the new features is not completely uniform, and some indications still refer to limitations relating to Gemini Spark and geographical availability.
Translation: Before we build a business process on top of it, let's check what's actually available in the account that will use it.
In the meantime, there is work that should not be lost.
Write down the rules that currently exist only in people's heads. Organize reference materials. Understand what information needs to be verified before something is published or sent to a client.
It's useful with Gemini. And it's also useful without Gemini.
The interesting question is not how many things we can entrust to AI
In DigiFe We use and observe these tools mainly from one point of view: what happens to the final result.
Producing an article in five minutes instead of fifty seems like a huge improvement. It's much less so if we then have to check every statement, change half the tone, and add all the information that makes that content truly our own.
With Skills, Google is trying to reduce some of this friction.
But technology only takes you so far. Someone still has to decide what's worth saying, what the company can promise, and why a customer should choose that particular offering.
If you're already using AI and find that you keep correcting the same things, I'd start there.
Not from a longer prompt.
From the corrections.
