At the start of this year, we at AAI Labs began using AI agents in our daily work, and every team tried them, not just the technical one. Agents sent emails, looked up contacts and answered questions in Slack channels. Most of them were small assistants for a single task, and the mistakes we made back then came from not yet knowing how much context and how many limits they needed.
Since then, we have run internal AI training for our employees every week. It has been going for more than half a year now, and over that time we have noticed a clear trend: our teams keep getting smarter and more curious.
From personal assistants to business tools
The biggest change was not in the tools but in how people understand them. At first, an agent was a helper you could hand an email or a search to. Now teams see it as a tool for building apps, workflows and entire systems.
This shows up in two ways: teams build more complex systems together with engineers, and more and more often they solve their own problems.
Systems we build together
The best example is Sally, an agent that counts the week's sales from Pipedrive data. Its first version was barely usable. We only managed to fix it by working with the sales team: they knew best what to count and how, and the engineers turned that knowledge into a working system. Today Sally works well and is part of the sales team's routine.
Other internal systems came about the same way: for collecting invoices, for weekly reports, for overseeing sprints. Each one started as a team's problem and became a working solution once the people who had the problem and the engineers solving it started working together.
Their own solutions to their own problems
The other change is smaller, but maybe even more important. People in non-technical teams increasingly solve repetitive problems themselves. With tools like Lovable, they build small apps to make their own work easier, and often help colleagues along the way. One of these tools, for example, lets everyone check how many vacation days they have left.
These solutions do not appear on their own. To notice a problem and realise you can solve it yourself, you need to know what AI tools can and cannot do.
Why it works
The same tools were available from day one. The people changed, and a few things made that happen:
- Regular training. The sessions run every week, so people come back with new questions from their work and stories about solutions they have discovered.
- Expert support. Anyone who gets stuck can ask an engineer who builds AI systems every day.
- Mistakes are discussed openly. Our failures with agents were only useful because we talked about them and learned from them.
- Working together. In our mini-projects, mixed teams solve real problems, so engineering knowledge reaches the other teams too.
What this means for other companies
Buying an AI tool is easy. Getting a team to use it thoughtfully is harder: knowing what AI is good for, where it gets things wrong and what you can build with it. Without training, employees often stay at the first stage, where AI is just a chatbot for writing emails.
Good training has two traits: it is practical, teaching people to solve their real work problems, and it is led by people who build AI solutions themselves. That is how we run our AI workshops for companies: our engineers share what they have learned building systems for our own company and for our clients.