Agile Heart · Andrija Arapovic · Agile Coach in Vienna
The AI transformation is just another transformation. I guide companies through it with what made agility strong: short cycles, clear accountability and teams that learn together. Only now people and agents work to the same beat.
Breather
A few years ago I was sitting in a meeting and someone said: “Scrum doesn’t work here.”
Today, in other rooms, I hear: “We need AI, fast!”
The sentences are new. The rhythm behind them is the same.
Which sentence did you hear last?
Pull it forward. I’ll show you how it sounded when the topic was Scrum.
Recognise
When companies introduced Scrum, people said the same things they say about AI today. The tools are new. I know the patterns, and the levers too.
“Scrum doesn’t work here.”
“AI does nothing for us here.”
Start small and after two weeks show something that works.
“We need Scrum.”
“We need AI, fast!”
Clarify the problem first, then choose the tool.
“The Daily is a waste of time.”
“The agents will handle it.”
Transparency: everyone sees what is being worked on, agents included.
“The Product Owner is overwhelmed.”
“Nobody is responsible for the agents.”
A role with a name, authority and coaching, so that it holds.
“Who decides when something is done?”
“Who signs off the agent’s output?”
A Definition of Done that applies to agents too.
“We’re growing faster than our structures.”
“Every department has its own tool.”
Orientation and shared rules first, then speed.
What changes
What stays
Understand
AI projects rarely fail because of the technology. They fail when nobody holds strategy, implementation and ways of working together. That’s why I work on all three, on one shared foundation.
Strategy
Roles, accountability and one goal. Who decides, who checks and what deliberately stays human.
From practiceXentral: hypergrowth from 30 to 280 employees, executive coaching of the leadership.
Implementation
Agents take over recurring work. With checks and sign-offs before anything goes live.
From practiceCyberSolutions: release cycle from 4 to 2.5 weeks, hotfixes per release from 5 to 0.
Ways of working
What carries everything: short cycles, honest retros, clear goals. Now with agents on the team.
From practiceScrum introduced or improved in more than ten organisations, from 6 to 150 people.
Since 2023, working with AI has changed fundamentally three times. Each phase has its own rhythm. Most companies are still in phase I or II.
Fascinating at first. Then come hallucinations, inaccuracies and flattery. The excitement fades.
Summarise our quarter.
Gladly! Your revenue is up 340%. A truly impressive achievement!
Sounds good, isn’t true. Nobody checked.
Suddenly everyone is building software. The results are mixed: what gets built without checks rarely holds up in production.
3 of 6 hold up in production.
Agents work on their own for hours and reach into your systems via MCP. What counts now is who steers them and who signs off.
The agent works. A human decides.
People and agents in the same sprint. Agile values matter even more here: short cycles, transparency, clear accountability.
Three people, three agents, one beat.
The strip shows the rhythm of each phase: isolated spikes, frantic building with breakdowns, agents running non-stop with one halt for sign-off, and finally a shared beat of people (red) and agents (dashed).
Breather
Before I recommend anything to a team, I listen. Who uses AI for what? Who decides what goes live? What does nobody say in the retro?
Eight questions are enough for a first impression. We’ll sort out the rest in conversation.
Measure
Eight questions across four dimensions: usage, sign-off, team and leadership. At the end you get named findings, a prescription in three steps and the format that fits. Nothing is stored or sent, unless you send it to me yourselves.
Question 1 of 8
All eight questions are answered. Your findings are alongside.
Findings
Phase II · Vibe Coding
Sample based on made-up answers. Answer the eight questions and your findings will appear here.
Breather
In February 2001, seventeen people met at a ski lodge in Utah. They wrote down four sentences that still hold today.
Every sentence has the same form: we value one thing more than the other. The other stays important. It just weighs less.
What would you put on the left side today?
Choose a word. It will be waiting for you at the end of the manifesto.
Next: StanceStance
We are learning to work with AI agents by doing it and helping others do it. Through this work we value:
In practiceEvery agent answer that touches a decision has a human reviewer. The agent delivers, the human judges and carries the responsibility.
In practiceThe Definition of Done applies to agents too: tests, checks, traceability. What can’t be verified doesn’t go live.
In practiceReview and retro include the agents. The team learns in every sprint, not once per licence.
In practiceEvery agent has a responsible person, named and visible on the board. If you don’t know who is responsible, switch it off.
The things on the right have their value. The things on the left matter more to us.
Help write the manifestoThe manifesto is written together
Which two things do you weigh against each other in your work with AI? Change the two terms and send me your sentence. I’m collecting them for the next version.
We value over .
AI Manifesto 2027Draft v0.1
Curiosity over control at any cost
How your sentence would read in the manifesto
Breather
Numbers come at the end of a story. At the beginning there is almost always the same moment: a team no longer knows which way is forward.
Three such moments from my engagements, and what happened next.
Trust
From my engagements, with the numbers from my CV. Each story says what follows from it for teams with agents.
In less than nine months the company grows from about 30 to almost 280 employees. Structures and processes are missing. Scrum is used in places, but poorly: knowledge and experience are lacking.
I support the teams as Scrum Master, coach the junior Product Owners closely and do executive coaching with middle management. In acute situations I help hands-on, for example with release planning.
Within four weeks, according to the Senior Product Manager, the situation is “turned around”: from a disoriented team to a stable one. Two new squads form following the LeSS approach, with one backlog and two Product Owners. Product and engineering tribes grow together through the coaching of their leads.
When tools grow faster than structures, you get the same chaos as in hypergrowth. Orientation first, then speed.
Employees in nine months
One dot stands for ten people: about 30 before, almost 280 after.
Sprint-Burndown
Volkswagen wants to develop more of its software itself. At the new competence centre Cariad, a tool is being built for internal specialists such as safety engineers. In one year the Release Train grows from one to 17 teams, spread across Germany and the Czech Republic.
I introduce Scrum in two inexperienced teams: a UI/UX team and a system team. Plus hands-on coaching, support with PI Planning, objectives, roadmaps, epics and backlogs, and close exchange with the Release Train Engineer. My approach: make the purpose and value of Scrum understandable instead of prescribing rules.
After four weeks, according to the Product Owner, the UI/UX team is “turned around”: from an overwhelmed team to a stable one that increasingly plans sprints with the PO on its own. The system team, an external supplier with hurdles of its own, uses the transparency of Scrum to assess and adapt its situation.
Agents don’t relieve an overwhelmed team, they overwhelm it faster. Stabilise first, then accelerate.
Teams in the Release Train, after one year
the first teamteammy two teams
UI/UX team after four weeks
overwhelmedstable, increasingly self-organised
Development is spun off from Cyberport and offers its own e-commerce platform on the market. Cyberport becomes one of several customers. One team is already running, a new data science team is starting.
I look after the development team as Scrum Master, introduce the new team to Scrum and coach business stakeholders, Product Owners and test and QA managers. I synchronise several teams following the LeSS approach and use Kanban for support alongside Scrum.
The release cycle drops from four to 2.5 weeks, hotfixes per release from five to zero. Agile and non-agile teams, such as DevOps and SAP, work in one shared release plan with one shared workflow.
Shorter cycles only work with quality assurance that grows along with them. It’s no different with agents.
Release cycle in weeks
Hotfixes per release
Open a row for the situation, role, result and what follows from it for AI teams.
Every engagement has left traces: methods, questions that are coming back with AI, and the formats that grew out of them. Choose a point and follow the path from experience to offer.
Engagement
The connections are my own reading, not a measurement.
I don’t come from computer science but from art, film and online business. That’s why I see teams as people making something together.
Business administration at the Vienna University of Economics and Business, then Digital Art at the University of Applied Arts Vienna, graduating with a Magister degree in 2005.
Camera, editing and post-production for industrial and commercial film, plus web design. Freelance and at production companies in Vienna.
Co-founder and later sole operator of an online start-up. Plus product management for a gaming platform and project management in online marketing.
As Product Owner at a weather service I introduce Scrum to the company and take responsibility for two of Austria’s most visited portals. I have worked in an agile way ever since, out of conviction.
As Agile Coach I help build an e-commerce competence centre: 30 developers at six locations, a relaunch, later the spin-off as a service provider in its own right.
Certified Scrum Professional, Product Owner with Roman Pichler, Scrum@Scale with Jeff Sutherland in New York, Kanban KMP I and II. Alongside that, engagements at ERGO, a financial software company and a data security provider.
Training in systemic coaching and ICF coaching skills. Plus coaching within my network, including the managing director of an e-commerce company with annual revenue of over one billion euros.
Xentral grows from 30 to 280 employees. At Volkswagen (Cariad) a Release Train grows to 17 teams. At UNIQA I support a Release Train of 150 people.
An art start-up for AR exhibitions, a museum in Vienna, then building a data platform at enercity. Alongside that, I work on what agents mean for teams.
Agile transformation meets AI transformation. I guide teams on the way to the agentic enterprise, with the values that have always carried the work.
How I work
Training
Breather
Nobody has to buy a transformation in one piece. Most collaborations begin with twenty minutes and one honest question: is this a fit?
After that we go step by step. After each step, you decide whether to continue.
How much time do you have to begin with?
Choose a time frame and I’ll mark the matching format.
Next: BeginBegin
Every format ends with something that belongs to you: a decision, a running workflow or a team that carries on by itself. The formats can be booked individually or combined.
How a collaboration can run · choose a row
Schematic. Every format starts with a free conversation. If none fits, I’ll say so openly. Stopping is a good result too.
I also give talks and workshops on the four phases and the AI Manifesto 2027.
I see the opposite, if anything. Facilitation, resolving conflict and accountability stay human. The role is changing: anyone guiding teams of people and agents needs more clarity about roles and sign-offs, not less.
Yes. AI is often the reason to finally talk about accountability, short cycles and shared goals. I bring both: introducing Scrum and working with agents.
It helps. I work with developers on equal terms, with hands-on experience in Docker, Kubernetes and Git and architecture knowledge from object-oriented development. The technology itself is built and maintained by your people or your partners.
Both. I live in Vienna and have worked on site with clients in Vienna, Munich and Hanover, and just as long fully remotely. My working languages are German and English.
Prices will follow. The intro call is free in any case. After that you’ll get a clear offer for each format.
Then I’ll say so. Stopping is a good result too, if it saves you time and money.
A few sentences are enough: what it’s about, who’s involved, by when. I reply personally. Or take the quick check first and send me the findings.