You can roll out a new tool with a training session and a manual. Or you can let people work out for themselves what it is good for. In early October we took the second route: two weeks, one safe environment and a rule that everyone takes part.
The IZ26 challenge is over, and it went better than we hoped. Our Mitigate AI platform now has 18 new skills that colleagues built themselves, and they are used by developers, project managers, team leads and our accounting team. One idea went beyond the platform altogether - more on that below.
Start with your own problem
People learn a new tool fastest when they use it to solve their own problem. So the challenge had a single task: find something in your work that keeps coming back and annoys you, and hand it to AI. Nobody had to build anything big. Everyone had their own workspace where they could try, get it wrong and try again. The deadline was short, and taking part was mandatory.
The quiet middle is normal
After eight days, fewer than half of the team had started. That is normal: most people get going when the deadline is close, and in the middle it matters not to panic.
The reasons were human. Some were waiting for a free moment, some did not know where to begin, and some did not want to show unfinished work. What helped was a personal message to everyone who had not started yet. In the second week almost everyone joined in, and in the end only two colleagues did not take part.
Two weeks did what months of manuals and training sessions had not. People started using our platform on their own because they were working on their own problem, not on someone else's example.
Ideas no management plan would have listed
Nobody built a grand "AI strategy", and that was a good thing. Developers wanted their timesheet to fill itself in from their calendar and the work they had done. Project managers wanted to see each morning which tasks had not moved for a week. Accounting wanted the fuel report to put itself together. These are small things, but each one costs somebody an hour every week, and none of them are visible from a manager's desk. You cannot write a list like this into a rollout plan. Only the people who do the work can.
When four people build the same thing
While judging, we noticed that ideas kept repeating. Four colleagues built a timesheet draft independently of each other. The project managers' morning briefing came in three versions.
Our first thought was to pick the best one. But then three people would learn that their work had been for nothing. So we did something else: we merged all versions into one skill, took the strongest part of each, and asked every author for permission and for a test on their own data. The skill got better, and every author's work is at its core. And repetition is the clearest signal of where the team hurts most.
The idea that went further
The challenge was won by our developer Ivars Belovs with Desk agent, a small Mac app that puts the platform's assistant in the menu bar. You can call it up from any app, and it sees what you are working on: the selected text, the open email or web page, a file you drop on it. Selected text can be translated, shortened or turned into a reply with one click. The app sends and deletes nothing without your approval.
From a change management point of view, this idea is worth more than it first seems. The biggest obstacle to using a new tool is that you have to open a separate window and remember that it exists at all. Desk agent works the other way round: the assistant comes to people where they already work. It can help in the accounting system, an old internal app or a client portal that has no AI of its own. Because the app works with a platform workspace, it has the same skills and access rights as that workspace. Desk agent is already on our product roadmap, and we are now working out which direction to take it.
What happens after the challenge
After a challenge, what matters most is that the good ideas do not stay in their authors' workspaces. So we spent the week after the deadline moving skills to where they help others.
| Workspace | For | New skills |
|---|---|---|
| Mitigate General | Everyone | Timesheet draft, error overview and release risk score, next task, dependency monitoring, code change review, technical text retold in plain language for clients, Mitigate documents and presentations |
| PM Helper | Project managers | Daily briefing, sprint health, task research, Slack message to Jira task |
| Team Leads | Team leads | Code security audit |
| Hestija | Accounting | Fuel reports, software invoice reconciliation and transfer, holiday report |
| Mobile | Mobile team | Android sources and library check |
Some skills also got ready-made schedules. A project manager can switch them on and get the morning briefing at 7:30 on working days, and the dependency report at the start of every month. That way a tool stops being something you have to remember to open.
When colleagues tested the merged skills on their own projects, they found bugs our tests had missed. The whole team found more in two weeks than any test plan. And the AI costs of the entire challenge were surprisingly small.
What we would suggest to others
If you have a new tool to roll out, ask every employee before the training: what in your work keeps coming back and annoys you? Give people a safe place to make mistakes and a short deadline. Do not worry about the quiet middle, a personal message helps then. And plan a week of work after the deadline so that the best ideas reach everyone.
In two weeks we got 18 tools that nobody would have put in a plan, and a team that now looks for what else it can hand to AI on its own. The skills keep growing because their authors improve them based on colleagues' feedback. And every new idea that helps one person can reach everyone.
Facts checked on 22 October 2026.