A lot has been written about artificial intelligence at work, and much of it is noise. But a handful of pieces of advice keep coming up in the most serious research and among the most listened-to experts. We have gathered and summarised them with a small business, an organisation or a freelancer in mind. The underlying idea is the one we argue for at Arkadium, a project of Opengea, the cooperative behind Intergrid: an AI that thinks for us makes us poorer; one that helps us think better renews us (from the essay “From intelligence to wisdom”, in Catalan).
1. Try it on your real work
This is the first of Ethan Mollick's four rules. He is a professor at the Wharton School and one of the most cited authors on the subject (Co-Intelligence, 2024): “always invite AI to the table”. Nobody can tell you where it will be useful for you; you have to find out by trying it on real tasks: a difficult email, a summary of a long document, a description of a service for your website.
2. You are the one who reviews and decides
This is Mollick's second rule and the most repeated piece of advice. A Harvard study with 758 BCG consultants measured it: on tasks AI is good at, those using it completed 12% more work, 25% faster and with 40% higher quality. But on a task where AI failed, those who relied on it got it right less often than those who didn't use it. The researchers call it the “jagged frontier”: you can't tell in advance where it will fail, and when it fails it doesn't look like it. That is why a person always reviews AI output before it is sent or published.
3. Give it context, as you would a new colleague
The third rule: treat it like a person, but tell it what kind of person to be. Answers improve a lot if you explain who you are, who the text is for, what tone you want and what it must not change. For example: “You are the customer service person at a small accounting firm. Rewrite this reply so it is kind but firm. Do not change any dates or amounts.”
4. Start with one specific process and measure it
A 2025 MIT report found that 95% of the generative AI pilots at the companies studied delivered no measurable return, and that the problem was not the technology but how it was built into the work. The lesson for a small business: pick a repetitive task that costs you hours, try AI on it for a few weeks and compare the time before and after. If it works, make it a habit; if not, move on to another.
5. Give your team a tool and some rules
According to the Microsoft and LinkedIn Work Trend Index, 78% of people who use AI at work bring their own tools, often with personal accounts and company data inside. Better to decide:
- Choose one tool and use it with work accounts.
- Keep conversations out of training: in ChatGPT, Gemini, Copilot or Claude, under privacy or data settings, turn off the option that uses them to improve the model. The business versions already commit to this by contract.
- Write half a page of rules: what you use it for, what must never go into it (passwords, bank, health or confidential customer data) and who reviews the output.
6. Train your people
The OECD surveyed more than 5,000 SMEs in seven countries: 31% were already using generative AI, 65% of those said it had improved staff performance and 83% said it had not changed how many staff they needed. The main obstacle: half said their staff lack the skills to use it well. The EU AI Act also asks organisations that use AI to take measures to train the people who use it. It doesn't take a long course: one session to share what works and what doesn't already helps.
7. Watch out: scammers use it too
Phishing emails now arrive without mistakes and in flawless language, and Spain's cybersecurity institute INCIBE has already documented cases of a manager impersonated with an AI-cloned voice. The advice from fraud experts is simple and it works: always confirm an urgent payment or a change of bank account through another channel, by calling the usual number, not the one in the message.
8. Assume tomorrow will be better
Mollick's last rule: “assume this is the worst AI you will ever use”. What doesn't work for you today may work in a few months. Every now and then, try again the tasks you had ruled out.
And one piece of advice from us: AI talks about your website too
More and more people, instead of searching on Google, ask an AI assistant “where can I find a carpenter in Vic?” or “which accountants work with cooperatives in Girona?”. The assistant answers with names and links to the websites it understands well. It helps if your website clearly says what you do, where and for whom, with opening hours, contact details, some frequently asked questions and your customers' languages. And there is a decision to make: whether to let AI companies' crawlers read it, because if you block them all you may also stop appearing in their answers. If you don't know what yours does now, write to us and we'll check.
It all comes down to one idea: neither fear nor haste, but good judgement. We will soon show you how we are building AI into the tools you already use with us.