Agile Meets Architecture Conference, Day 2 Recap
Nick Tune and Eduardo Da Silva on the Kesselhaus Stage
The second day of the Agile Meets Architecture Conference 2025 in Berlin has drawn to a close, and I'm leaving with a profound sense of satisfaction and inspiration. It was a privilege not only to attend but also to serve as a Program Chair and, this afternoon, as the track host for the Maschinenraum stage. The energy throughout the day was noticeable – a blend of focused learning and vibrant community interaction.
One of the standout aspects of this conference is its intimate scale. It fosters an environment where connecting with speakers and fellow attendees feels natural and easy. Engaging in conversations with numerous friends and colleagues from the Domain-Driven Design, Software Architecture, and Team Topologies communities was a highlight. There's a special kind of positive energy when passionate professionals gather, and this event had plenty of it. Even keynote speakers like Jurgen Appelo were readily available for chats, adding immense value to the experience.
The sessions I attended today offered a wealth of insights. Here are my takeaways:
Human Robot Agent: Scouting AI Beyond Hype and Doom - Jurgen Appelo
Jurgen Appelo, known for his contributions to modern management practices like Management 3.0, delivered a thought-provoking keynote on the human angle of Artificial Intelligence. He moved beyond the usual hype and doom narratives, focusing instead on how AI interacts with human capabilities and limitations. Jurgen pointed out that our brains aren't optimized for tasks involving extensive memory, repetition, or calculation – areas where AI excels. He even shared an anecdote about replacing human beta readers with AI for his books, noting AI's superior consistency. However, he cautioned against outsourcing the discovery of insights solely to AI.
Jurgen Appelo with an interesting quote
The talk explored the concept of "vibe coding" as a way to address bottlenecks, suggesting AI helps manage volatility. Appelo argued for the increasing importance of generalists, as AI takes over specialized tasks. This shift might transform "T-shaped" professionals into "M-shaped" ones, capable across multiple domains. He posited that AI necessitates agility more than ever, potentially leading to more dynamic team structures, akin to Heidi Helfand's concept of dynamic reteaming, rather than traditional static teams. For me, the connection Jurgen drew between the decline of static, long-lived teams in favor of dynamic reteaming, fueled by AI's ability to reduce cognitive load – illustrated by the firefighter analogy where teams form rapidly without traditional storming/norming phases – was a key insight. This dynamic interplay definitely warrants further evaluation in the sociotechnical space.
Appelo also touched upon the disintegration of linear value streams into more dynamic "value networks", driven by human-machine collaboration. He emphasized shifting focus from mere products to the overall user experience, where AI can be a significant enabler. Finally, he introduced the MARVIS framework (Modularity, Ambiguity, Reflexivity, Volatility, Intricacy, Scalability) to highlight complex areas where human involvement remains crucial, especially concerning reflexivity – the need for human oversight and the potential skill deterioration if we rely too heavily on AI without practice.
How Flow Works and other curiosities - James Lewis
James Lewis with a quote from Accelerate
James Lewis, a Software Architect and Director at ThoughtWorks known for his work on microservices, delved into the dynamics of workflow in software development. He framed software development as knowledge work, where value increases as information moves through the value stream. Unlike physical inventory, delayed digital work (like old Jira tickets) represents stalled knowledge flow. James utilized Value Stream Maps to highlight wait times versus value-creating cycle times, urging a systems-thinking approach.
He broke down common bottlenecks: coordination dependencies, scheduled batch processes, and queues. Making these visible inherently reveals the underlying economics. A significant portion of the talk focused on queuing theory (M/M/1 vs. M/M/c models), illustrating how service capacity impacts throughput. He contrasted traditional Ops teams (limited capacity 'c') with the near-infinite capacity of IaaS, explaining the success of cloud platforms like AWS in improving flow.
Drawing on Reinertsen's principles, James discussed congestion collapse (catastrophic drops in output when load is too high) and the importance of controlling queue size rather than just cycle time to maintain optimal throughput. The core message, backed by complexity science and information theory, emphasized that understanding and managing flow often leads to reducing batch sizes – a key factor in efficient software delivery.
Domain-Driven Transformation: How to improve the structure of legacy systems - Carola Lilienthal
Dr. Carola Lilienthal, an expert in software architecture and author of "Domain-Driven Transformation", addressed the challenge of revitalizing legacy systems, often referred to as "Big Balls of Mud". These systems, burdened by technical debt and complexity, consume development capacity with bug fixing rather than feature creation. Carola proposed Domain-Driven Transformation as a path to break down these monoliths into manageable, less coupled components, thereby reducing cognitive load.
The transformation process begins with collaborative domain rediscovery using techniques like EventStorming or Domain Storytelling to understand the business domain independent of the existing software. This understanding forms the basis for defining a target architecture centered around subdomains and Bounded Contexts, visualized using a Context Map. The crucial third step involves comparing this target architecture with the actual system structure to gauge the distance, potentially using metrics like the Modularity Maturity Index (MMI) which Carola and her company WPS developed.
Carola outlined strategies based on the system's current state. For tangled systems, stabilizing the technical foundation (build automation, framework updates, error resilience) while strategically strengthening the domain model (using value objects, reducing inheritance) is key. She differentiated between domain types like Pipeline, Blackboard, and Dialog domains, noting that "Pipeline" domains are often showcased because they are easier to decompose, while others present unique challenges. The final step, "do the move," involves executing the refactoring and restructuring based on the established plan.
Carola Lilienthal showcasing the MMI Metric
Now we’ve given them every freedom that they want and they still don’t do what we want - Michael Mahlberg
Michael Mahlberg on stage in the Maschinenraum
Michael Mahlberg tackled the nuances of autonomy and motivation in teams. He started with a simulation demonstrating increased happiness and productivity with shifted decision-making, seemingly advocating for self-managing teams. However, he quickly pivoted, arguing that complete autonomy without guiding goals or constraints isn't fulfilling. He used the analogy of traffic lights or sports rules – constraints that enable, rather than restrict, effective action.
Michael emphasized that true autonomy requires supporting infrastructure and mutually agreed-upon "road signs" or guiding principles, which teams often lack the opportunity to negotiate. This lack impacts motivation, which he discussed through the lens of intrinsic versus extrinsic factors. He cautioned against the simplistic "intrinsic good, extrinsic bad" dichotomy, referencing Self-Determination Theory's concept of autonomous motivation tied to values and suggesting the distinction is sometimes used as an excuse for poor leadership.
Effective leadership, according to Michael, involves a mix of styles: signalling desired behaviours, directing when necessary, leading by example, and inspiring. He connected this to delegation, referencing the Tannenbaum and Schmidt continuum and Management 3.0 delegation types. Key principles included not delegating what hasn't been delegated to you and only delegating if you're prepared to accept the outcome. The talk provided practical examples of how to frame delegation effectively.
Architecture: The absence of self organization - Gerrit Beine
Gerrit Beine introducing his ideas
Gerrit Beine offered a philosophical perspective on architecture and self-organization. He posited that self-organization is a fundamental process in the universe, leading to structure formation in physical and social systems alike. However, he argued that human endeavors aimed at achieving specific goals inherently require reducing self-organization by imposing constraints and direction.
Gerrit challenged the common interpretation of the Agile Manifesto's statement about the best architectures emerging from self-organizing teams. He contended that architecture, by defining fundamental structures and principles (as per ISO 42010), actively limits self-organization to ensure consistency and guide development within certain boundaries. Therefore, architecture represents an intentional absence of complete self-organization. His revised take: the best architectures emerge from teams capable of defining rules to create appropriate structures from disorder.
He proposed three principles for architecture: Responsibility (awareness of decision consequences), Empowerment (enabling non-tech people to control their use of technology), and Agency (building based on local interactions for sustainability, maintainability, and security). Ultimately, architecture acts as the boundary and pathway that directs the ever-present force of self-organization towards desired outcomes.
Riding The Rollercoaster of Change: Turning uncertainty into growth - Sofia Katsaouni
Sofia Katsaouni delivered a deeply personal and insightful talk on navigating change. Sharing her own experiences, including moving to Finland in 2014, she framed change not just as a challenge but as an opportunity for growth, emphasizing self-awareness, curiosity, experimentation, and stability.
A key point was adding an "acknowledge" step to the agile inspect-and-adapt cycle. By consciously acknowledging our feelings and situations, we empower ourselves to make deliberate choices. Sofia explored the complexity of emotions during change, noting that opposing feelings like excitement and anxiety can coexist, sometimes accompanied by guilt. She advocated for curiosity – towards ourselves (What fears or opportunities does this change bring? What do I risk losing?) and towards others, fostering connection.
Instead of large leaps, Sofia recommended small, manageable experiments to navigate change, allowing for course correction. Crucially, she stressed the importance of finding stability amidst flux – identifying constants that provide grounding. For her, these included self-awareness, friends, workouts, nature, and her partner. The talk encouraged identifying our own constants and using acknowledgment, curiosity, small experiments, and grounding constants as tools to turn the uncertainty of change into a path for growth.
Sofia Katsaouni sharing personal stories in her talk
Shift-Left for Learning: With the Help of Enabling Teams - Tsvetelina Plummer & Berrin Akvardar
Berrin Akvardar and Tsvetelina Plummer explaining the shift left on learning with many practical examples
Tsvetelina Plummer and Berrin Akvardar presented a compelling case for evolving organizational learning, stating that great work stems from a great learning culture. They critiqued the traditional top-down learning flow (leadership defines strategy -> training sessions -> employees consume) as slow and positioning employees as passive recipients. This model often fails because leadership might not accurately grasp the actual learning needs across the organization.
Their solution involves "shifting learning left," integrating it more closely with development teams, inspired by the Enabling Team concept from Team Topologies. This moves away from scheduled training towards continuous enablement. They questioned who comprises these enabling teams, playfully suggesting roles like "LearnOps Engineers". The core idea is to make learning opportunities more accessible, timely, and relevant to the teams' immediate contexts.
They shared a practical example from GfK/NIQ's "we.innovate" initiative, a community of practice designed to break down knowledge silos. Key principles included community-driven topic selection (with voting), empowering internal "enablers" (trainers), and collaborative event organization. Incentives like awards and badges were used, but crucially, learning wasn't tied to performance targets. The initiative was highly successful, boosting visibility, encouraging participation, and receiving excellent feedback, demonstrating the power of shifting learning left.
Waving goodbye!
It was an enriching day filled with valuable discussions and perspectives. The Agile Meets Architecture conference schedule delivered a strong program, and the vibrant community made it a truly rewarding experience.
Now, as the conference wraps up, I feel a mix of sadness to leave this inspiring environment and excitement from all the ideas buzzing around in my head. My brain feels heavily inspired, and I'm eagerly anticipating a large book order I placed during the event – hopefully, it arrives tomorrow! I truly can't wait for the next edition of Agile Meets Architecture. I wish everyone a safe trip home and look forward to seeing many familiar faces again soon, perhaps at ComoCamp, DDD Europe, or Craft Conf!
Martina Freers, Michael Mahlberg, Jörg Müller and Dajana Günther during the closing session
