Gemini skills: write down once how your team works

What Google has announced

Google announced it on 30 September. A skill is a set of instructions you write once and Gemini applies whenever you call it up: how you want a newsletter drafted, say, or the tone you use with customers. To use one, you just type a slash and its name in the chat. You can combine several in a single request, such as a writing-style skill with a brand-guidelines one, and, according to Google's blog, they can include reference files as text, PDFs or images. Android Headlines adds that Gemini can spot prompts you keep repeating and offer to turn them into a skill.

The rollout is happening in stages: in the Gemini app for personal accounts it began on 30 September, and for business accounts Google expects to finish by mid-November. In Workspace it covers the Business and Enterprise editions, among others, for users aged 18 and over. Skills use SKILL.md, an open format based on Markdown (plain text with simple formatting marks), so a skill built on another platform can be copied across. For now they do not sync between the Gemini app and the Workspace apps: if you want one in both places, you have to create it in each. In business accounts, Gems will stop working no sooner than 1 March 2027, and any that remain will be turned into draft skills.

Lucía's notebook

Imagine you run a small travel agency with four people behind the counter. Lucía has been there longest and has her own way of putting a proposal together: first she asks about the real budget, then about the pace of the trip, and she never sends an itinerary without a plan B for a rainy day. All of it lives in her head and in a notebook with dog-eared corners.

For a few months now, the team has been using Gemini to draft proposals. Everyone prompts it their own way and everyone gets something different. Lucía's come out well because she knows what to ask for. One Tuesday in October, while Lucía is on holiday, a family walks in in a hurry and a colleague asks Gemini for a trip to Lisbon. The result is fine, but it looks nothing like what the agency usually sends.

Why does know-how stay in one person's head?

The scene is made up, but plenty of businesses will recognise it. In a small company, processes are not written down: people learn them by watching. That works as long as the person who knows is around. Nobody has time to write a manual, and when one finally gets written, it ends up in a folder nobody opens.

Something similar has happened with AI. The tools arrived without shared rules, and each person has built up their own prompts, which live in their chat history. The business's knowledge is still scattered, only now some of it sits in conversations nobody else sees.

The cost of everyone doing it their own way

That cost never shows up on an invoice, but you feel it. Proposals change depending on who signs them, and customers notice. New starters take weeks to pick up what everyone else takes for granted. And when the key person goes on holiday, or leaves for good, part of the way the business works walks out with them.

Skills will not fix this on their own, but they change the starting point. What used to be a manual forgotten in a folder can now be an instruction the tool applies every time someone calls it up.

How do you turn a recurring task into a Gemini skill?

There is no need to start with the whole business. One task that comes up every week and turns out differently depending on who does it is enough:

• Pick that task: a proposal, a reply to a complaint, a weekly summary for a client.
• Sit down with whoever does it best and note their steps and judgement calls in their own words: what they ask first, what is never missing, what tone they use.
• Add one or two good examples and the documents they rely on as reference, leaving out customers' personal data.
• Save it as a skill, try it on three real cases and fix whatever goes wrong.
• Decide who keeps it up to date and whether you need it in the Gemini app, in Workspace or in both, because for now it has to be created in each.

The most valuable part of the exercise is the text you write, and that text is yours: it is in an open format, so you can take it with you if you ever switch tools. One habit is worth keeping, too: whatever the AI prepares, a person checks before it reaches the customer.

How we help at digizone

At digizone we help companies get their processes in order and, when a general-purpose tool falls short, we build custom software that fits the way they work: from automating a recurring task to building an in-house assistant with the business's rules built in.

One example is Magnet-IA. The idea was to give boutique and independent hotels access to the judgement of a consultancy specialising in hotel revenue management (pricing and occupancy) and marketing. We built a copilot in which the AI answers using each hotel's real data, analyses its reports and suggests concrete actions in plain language. With every message, the agent stacks three layers: the global instructions, the hotel's profile and a set of adjustments only Magnet-IA can see, which it can fine-tune from a dashboard without touching any code. It is the same idea as skills, built into a product of its own.

Start with the task you repeat most

At Lucía's agency, the first version would be a single skill: the one for preparing trip proposals, written with her, tested on a week's worth of requests and corrected wherever it fell short. If it works, the next one could cover changes and cancellations. That is the idea behind an MVP, a minimum viable product: start with something small that already solves a real problem and let it grow with use.

If you would like to see how it would work for you, our free workshop, in Spanish, walks through it step by step.