The AI Office That Began with One Brave Team
Thursday 8 October — Art of Leadership (I) · AI founders & systems
The table
The case
Subtitle: A fictional leadership case for discussion. Source: The owner's own case (Dražen Kapusta, 2026-09-13), verbatim — the eight questions are his
The customer request arrived at 8:42.
It required information from sales, operations and finance. Under the usual process, three people would search for documents, exchange messages and assemble a response.
At 9:18, a draft was ready for review.
It contained the relevant customer history, a proposed answer and two clearly marked questions requiring confirmation.
One of those questions mattered.
The requested delivery date conflicted with the current operating plan.
Marta, the team leader, called operations before approving the response.
By 10:05, the customer had an answer the company could support.
The team was pleased.
They were also clear about what had happened: AI had helped them prepare the work, and a person had resolved the commitment.
It was the first week of their new office experiment.
A Team That Volunteered
The company had discussed AI for months.
People had tried tools individually, and several had found useful applications. The founder now wanted to see how a whole team could work with AI across a shared process.
Marta’s five-person commercial support team volunteered.
They were experienced, curious and familiar with the frustrations of fragmented information. They handled enquiries that crossed departmental boundaries and knew where work slowed down.
They proposed starting with one category of existing-customer requests.
The founder asked what they needed.
Protected time to prepare. Access to the right internal information. Support from operations and finance. A clear agreement about what the system could prepare and what people would continue to authorise.
They also asked for permission to report mistakes openly during the trial.
The founder agreed and became the sponsor.
Giving the Experiment an Address
The team named the initiative the AI Office.
At first, it occupied one shared working area, a set of tools and a regular review on the calendar.
Their goal was concrete: prepare accurate customer responses with less searching, duplication and internal chasing.
They walked through recent requests and identified the work involved.
AI could help classify an enquiry, retrieve approved information, prepare a response and highlight missing details.
People would handle ambiguous requests, confirm commercial commitments and decide what to send.
The team assigned ownership of the information the process depended on. When a source was outdated, someone needed to correct it.
They also defined a practical fallback. If the tools were unavailable or the information could not be trusted, the team would continue through the existing process.
That allowed them to begin without making daily service depend on an experiment.
The First Useful Surprise
During preparation, they discovered that different colleagues used different versions of the same guidance.
The AI repeatedly surfaced the inconsistency.
The team could have treated it as a retrieval problem. Instead, they asked the relevant manager to settle which guidance applied.
Another recurring issue involved missing customer details. A short intake change prevented several follow-up messages.
Some early improvements came from making the work clearer before automation did much of it.
Marta saw this as progress.
The team was learning what a reliable AI-supported process required from the organisation.
The Mistake They Could Discuss
On the fourth day, a draft used a previous customer exception as if it were a standard condition.
The reviewer caught it.
The answer sounded plausible because the exception had been valid in its original context.
The team examined how it had entered the draft, adjusted the source handling and added a review point for similar cases.
Marta recorded the event alongside successful cases.
At the weekly review, she showed it to the founder.
He asked whether it meant the trial should pause.
“For this kind of commitment, the review stays mandatory,” she replied. “We have corrected the issue we found. We should now check whether the correction holds across the other relevant cases.”
They agreed on a focused check and continued within the defined scope.
The team learned that reporting a mistake would lead to useful examination. That mattered to how honestly they would work.
A Working Session with the Tribe
Marta and the founder brought the experiment to a COTRUGLI COlab session.
They used synthetic examples based on the kinds of requests the team handled.
Alumni took the roles of customer, operations manager and commercial reviewer. They introduced late changes, incomplete information and conflicting instructions.
The session showed that the process handled some cases well and became less clear when a request changed halfway through.
A participant who led a service business helped the team clarify where ownership should transfer.
Another suggested comparing the full time to a usable customer answer, including corrections and review.
At dinner, Marta spoke with leaders running similar experiments.
They discussed ordinary concerns: who maintained the sources, how colleagues reacted, whether the team was actually gaining time and what happened when the most enthusiastic employee was absent.
The shared work made the advice specific. The informal conversation made it easier to admit what still needed attention.
When Everyone Wanted an AI Office
By the end of the first month, the trial had attracted interest.
HR wanted help preparing internal responses. Procurement wanted to explore supplier enquiries. Another sales team asked whether it could use the same setup immediately.
The founder saw an opportunity to expand.
Marta saw the risk of turning a capable pilot team into the support desk for every new experiment.
Their process worked within a particular scope, with sources they understood and reviewers who knew the business.
Another team would bring different information, decisions and failure modes.
Marta proposed a repeatable starting package: choose one workflow, define the outcome, assign an owner, prepare the sources, set approval boundaries and run a measured trial.
Her team could share what it had learned and help another group begin.
It could not carry every group’s responsibility.
The Second Team
The founder now had to choose the next step.
He could expand the original workflow to more users, help a second team test a different process or establish a small central group to support adoption.
Each option needed people and ongoing maintenance.
He also needed to decide what the first team would do with any capacity it gained. Better customer follow-up was one possibility. More time to resolve recurring causes of enquiries was another.
The AI Office had begun as a local experiment.
Its next phase would test whether the company could spread what worked while keeping ownership close to the work.
Marta wrote a sentence at the top of the proposal:
“Every new team needs a business owner who is prepared to learn with it.”
The CDayZ Leadership Challenge
You are deciding how to develop the AI Office after its first month.
Choose the next workflow or team, define the support it will receive and establish how results will be assessed.
Your proposal must include ownership, information maintenance, human review, a fallback and the capacity needed to sustain the work.
Discussion Questions
The owner's "Eight Questions for the Room":
- Why was this team a useful place to begin?
- What should the first business outcome be?
- How should responsibilities be divided between AI and people?
- Who owns the information the process relies on?
- How should the company respond when the team finds an error?
- How could the COTRUGLI tribe improve the experiment?
- What should happen when other departments want to join?
- What would justify continued investment after ninety days?
Identify the characteristics of the people and workflow that made the experiment manageable and worth testing.
Define a measure that captures the full customer result, including review, correction and work passed to other departments.
Specify what the system may prepare and which decisions or commitments require human authority.
How would the team identify outdated guidance, resolve contradictions and maintain improvements?
What would determine whether to correct and continue, narrow the scope or pause part of the process?
Design a practical session that tests difficult cases and produces changes the team can use.
Choose an expansion approach that provides support while keeping responsibility clear in each team.
Include business value, service quality, maintenance effort, team capability and how any released capacity is being used.
A Lesson for the Next COTRUGLI Generation
A small team can give a company a practical way to discover how AI belongs in its daily work.
The team needs a meaningful outcome, reliable information, clear authority and support for learning from what happens.
The COTRUGLI tribe can strengthen that learning through realistic challenges and the experience of others doing comparable work.
The first AI Office began with five people willing to try.
Its future depended on the organisation’s willingness to learn alongside them.
Your comment on this chapter
A question for the table, a disagreement, what you would have done. The case lead reads every comment; the ones the table takes up enter the chapter as questions from the room, with your name.