BONUS When the Team Becomes the Operating System for AI With Marko Taipale

BONUS: When the Team Becomes the Operating System for AI — Marko Taipale on the Twin Project at Solita Most teams trying to "use AI" end up with fast individuals and a slower system. Marko Taipale ran a two-year experiment at Solita that suggests the real bottleneck isn't the tool — it's the team's operating model. In this conversation, Vasco and Marko walk through the Twin Project with ISS Finland — two teams, same ERP pricing tool, one classical agile, one with generative AI in the room — and the lessons that became the CollabAI framework. The Twin Project — Two Teams, Same Product, One With AI in the Room "We had a luxury of: do whatever you want with AI, please get at least the same results, towards the same goal." In 2024, Marko's team at Solita was set up as the counterpart to an existing Scrum team building an ERP pricing tool for ISS Finland. Same product mission, two different operating models. The first days were chaotic and exploratory — the team tried over 120 AI tools, built a custom GPT to act as a stand-in product owner, and even sent a virtual assistant to sit silently in the other team's meetings so nobody from Marko's team had to attend. The framework grew out of what kept working, not from a plan written upfront. "Fast Individuals, Slow System" — Why Buying Licenses Doesn't Fix the Bottleneck "If you don't change your structures, AI won't do anything faster. The only thing that gets faster is the queues between your decision-making gates." The line from Marko's book lands hard once you have seen it inside a team. A Copilot license speeds up the individual — and then the individual sits and waits for the rest of the system: reviews, handoffs, refinement, stakeholder meetings. Those queues are exactly what AI accelerates, and the team feels even more frustrated than before. The real intervention is upstream, in how the team shares context and makes decisions together. Without that, AI just makes the existing inefficiency more obvious. Drifting in the Solution Space — Why Sense-Making Has to Happen Together "None of the real problems are so simple that a single person can solve them. If it's that simple, you should automate it." The early Twin Project team kept seeing what Marko calls drifting — small interpretive mistakes at the start of a task that twisted the solution into something unrecognisable later. Each person was reading the same docs and the same proxy-PO conversations, and each was leaving with a slightly different picture. Individual interpretation was not enough. They moved from individuals to pairs, then to whole-team sense-making sessions. The shared context only became useful when the team processed it together — and that processing turned out to be where the learning compounded. Never Leave the Daily — When Mob Programming Becomes the Operating System "This is happening so fast, we shouldn't actually leave the daily." The team started with vanilla Scrum, extended dailies, then ran multiple per day, then realised that the meeting was the work. They drifted into mob programming without naming it — a shared virtual machine where one person controlled the screen at a time, switching every few minutes. The agile labels came later, when someone read Mob Programming by Woody Zuill (see his earlier episodes on the Scrum Master Toolbox Podcast) and saw the team's own behaviour reflected back. The takeaway: when the pace of decisions exceeds the cadence of meetings, the team has to live inside the conversation, not visit it once a day. The 40-Prototypes Moment — When the Customer Joined the Mob "You waited 37 hours to get to this point where we get feedback." This was the turning point. Marko had built 40 different prototypes of the pricing tool in one hour, then walked into a weekly review where the customer pointed out the obvious: if the prototypes took one hour to make, the team had been waiting 37 hours to get the feedback that actually mattered. From that moment the client became part of the mob. New product directions started landing every five to seven minutes. The backlog quietly disappeared — issue management stayed, but for the AI's context, not for humans. When the product owner is in the room all day, the storage-and-handover layer stops earning its keep. The Regulation Layer — Why Sustainable Pace Gets Sharper, Not Softer, With AI "AI is a machine. It won't stop. That's why we need a regulation layer — and we have to regulate together, not individually." What broke first in the Twin Project was not the technology — it was the people. Cognitive load and the brain's hunger for clarity become the new constraint once decisions are flying every few minutes. The old Scrum idea of sustainable pace gets a second life here, but it has to be a shared pace, set by the team, not an individual one. Engagement is...