Alma Media has been actively building its AI capabilities for the past few years. With a dedicated AI program guiding the work, AI is not a side initiative but a clear part of the company’s strategy. Across the organization, there are already advanced users, and in some teams, agent-based workflows are part of everyday work.
At the same time, the focus has been on bringing the whole organization along. Alma has developed multiple ways to support this, and the Learning Sprint is one concrete way to do it in a focused, hands-on format especially for those who are just getting started.
At Splended we often hear the same question from organizations: we get that AI agents are a thing now, but how do we build them?
Together with Alma Media we set out to answer that question in the most practical way we could. Instead of lectures or one-off workshops, we designed a Learning Sprint: eight two-hour sessions where cross-functional teams would build their own AI agents in Microsoft Copilot Studio. Not watch demos. Not follow along but to build.
Our coach Santeri Kallio met with participants twice a week over four weeks. Each session started with a 20–30-minute teaching block on a new concept, followed by hands-on work where participants built and tested things on their own. Teams from HR, finance, legal, and IT all worked side by side, which turned out to be one of the best parts. People saw how the same technology could solve very different problems depending on the context.
From ideas to working agents
We started by asking people what kind of agent they would most want for their own work. The answer was clear: something that finds and surfaces the right information from internal systems. That is where most of them were losing time every day simply locating the right document or answer buried somewhere in the organization.
Then we asked them to build one in the first session. As one participant put it:
“Surprise: the first experience is that it helped us with navigation and how we should build the agent.”
Within an hour, people who had never touched the platform were getting useful answers from their agents.
By the second sprint session, we had over 30 concrete agent ideas on the table. What struck us was how grounded they were. Nobody was pitching sci-fi scenarios. People wanted agents that could handle things they were dealing with every week: sorting through shared inboxes, answering recurring questions about company policies, preparing for meetings, and consolidating data from multiple systems. One team proposed an agent that helps managers navigate tough conversations around feedback and performance. The ideas came straight from daily frustrations.
Learning by building
Things clicked during the prompt engineering sessions. We gave participants two intentionally bad agent instructions and asked them to fix them. The feedback was unanimous: every single line needs to be more specific. People started asking questions they had never considered before. Who exactly should receive a meeting summary everyone in the company, or only the people who were invited? What does “professional tone” mean in this context? What should the agent do when it doesn’t know the answer?
One group came up with a prompt structure that others quickly adopted: Objective, Instructions, Context, and Stop conditions. That framework reframed the whole exercise. The question was no longer whether AI can do something, but whether you can describe what you want precisely enough.
Later, when participants connected their agents to real company data sources, they hit the kind of friction you only discover by doing. Some sources worked out of the box. Others required workarounds because of login walls or processing delays. One participant loaded a large internal document library and had to wait a while for it to process, but when they tested it, the agent gave solid answers. Another added a collective agreement document, and the agent could handle basic questions about it on the first try.
People were testing with their own documents and workflows, which is what made it stick.
What we learned
By the sixth sprint session, when we introduced more advanced concepts like structured conversation flows, the group had split into two camps and we were happy about that. Some participants were already thinking about linking multiple agents into a single workflow. As one person wrote:
“I will definitely use these, because the more I can automate what agents do, the more I see it truly benefiting my work.”
Others were more measured:
“I can’t say yet. I have to test and look into these more first.”
People who were ready to run had the room to push ahead. People who needed more time weren’t being rushed. A sprint-based structure can accommodate both.
We also picked up useful feedback about the format itself. Some participants, especially those starting from scratch, wanted a more guided walkthrough at the beginning before branching out on their own. That is something we are incorporating into future programs.
A strong starting point for AI adoption
One participant summed up the experience well:
“This is one of the best ways to learn AI: hands-on, practical, and directly tied to everyday work. Santeri is a true professional who managed to support participants at very different levels and keep everyone progressing. This kind of approach is taking us in the right direction in Alma’s AI strategy as we introduce agentic workflows across the organization.”
Another participant put it simply:
“This was a good activation to start working with my own agents. It pushed me to turn ideas into practice.”
What made this collaboration work was the rhythm. Two sessions per week kept the momentum going without overwhelming people. The teaching-then-building format meant nobody could passively sit through a presentation. And the eight-sprint arc gave enough time to move from “what is an agent?” all the way to multi-step workflows with real data sources and autonomous behaviors.
By the end, participants across four different business functions had working agents tied to their own processes. They also walked away with a clear sense of where AI agents help and where they don’t which is hard to get from a presentation or a demo.
In short, this was not about introducing AI. It was about building the capability to use it in everyday work.
