Track 1

11:00 

Incremental modernization through CodeConcise

Alessio Ferri and Jonathan Passé

In today’s competitive market, legacy systems often hinder the rapid delivery of products and services that align with evolving customer expectations. As a result, legacy modernization is not only more critical–but also more urgent–than ever. Join us to explore how we are using a combination of deterministic techniques and AI to identify the core business capabilities buried within legacy systems. By uncovering “seams” in the code, we decompose modernization into manageable “thin slices” that can be delivered incrementally. This approach democratises modernization by providing continuous value and feedback throughout the journey–rather than deferring ROI until the end of a multi-year transformation programme.

11:55 

AI in action: Boost productivity and deliver business value with practical tools

Sara Michelazzo  

Thoughtworks is undergoing a bold AI transformation, integrating tools like Gemini, NotebookLM, and Agentic AI to boost both individual and enterprise productivity. In this session, we’ll share hands-on experiences - what worked, what didn’t, and how we’re navigating change. Expect live demos of AI-augmented workflows in Sales and Knowledge Management, including intelligent research for account planning, strategic meeting prep, and rapid proposal creation. We’ll also unpack our adoption strategies. Designed for practitioners and leaders alike, this talk offers practical lessons, adoption tactics, and clear impact metrics - ideal for anyone looking to integrate AI into daily workflows and drive real business value.
 

12:50 

Streaming: Real-time architecture for connected products in automotive

Baran Demir

Data-intensive systems demand architectures that can seamlessly handle real-time access and scalability. This talk dives deep into the technical execution of building such a system, focusing on a practical example from the automotive domain. We’ll present a scalable streaming solution designed for real-time data access, powered by event-driven microservices and MQTT for efficient, lightweight communication. The architecture integrates secure APIs and robust access control mechanisms to balance performance, security, and usability. Throughout the session, we’ll share the technical challenges we faced, the lessons learned along the way, and the tools and strategies that helped us deliver a reliable, future-ready solution.

Track 2

11:00 

The paradox of TDD and pairing: Why it's so difficult but so worthwhile

Paola Reyes Montalvo

This talk takes a candid look at the real-world experience of adopting Test-Driven Development (TDD) and pair programming. Drawing on a six-year personal journey, we'll discuss the initial frustrations of reshaping coding habits, the significant mental shifts required and the eventual rewards of these practices. You'll hear practical strategies to overcome common challenges, along with real-world examples of how TDD and pairing lead to more robust, maintainable code and stronger team collaboration. By the end, you'll leave with actionable steps to transform coding from an individual challenge into a shared, empowering success.

11:55 

Is clean code dead?

Sarah Schmid

In today’s fast-paced development world, conversations about clean code have become rare, but is it really less important? This talk is a reminder of why "good" code quality still matters, simple ideas to write better code and how to encourage others. Let’s bring clean code back onto the table.

12:50 

Enabling enterprises for generative AI agents

Fabian Nonnenmacher

Many organizations are integrating generative AI tools in an ad-hoc manner, missing the opportunity to build a strategic, unified capability. This talk offers a blueprint for enterprises ready to unlock the full potential of generative AI agents. We'll delve into the essential platform capabilities needed to support the entire agent lifecycle, from service creation and deployment to robust evaluations and fine-tuning. This platform empowers product teams to organically create and share AI tools and agents, enriching existing offerings and making them discoverable company-wide. We'll also spotlight the Model Context Protocol (MCP) and show how it allows various AI agents to work together seamlessly throughout your organization. To round things out, we'll tackle the enterprise challenges you can't ignore – security, auditing and practical governance that enables rather than blocks innovation.

14:30

Context-aware policy authoring for Terraform: Generating OPA policies with an MCP server

Óscar Cobles Montes

As developers use AI agents to generate Terraform configurations, ensuring compliance with internal policies becomes a critical challenge. This talk presents a new approach: Integrating an MCP (Model Context Protocol) server that connects AI agents with existing OPA (Open Policy Agent) policies. By querying these policies, the server extracts real-time constraints, such as allowed instance types, required tags or open ports, and feeds them back into the generation prompt. This allows developers to request infrastructure in natural language (“I need a server on AWS”) and receive compliant Terraform code by default. The session includes a prototype demo and explores how this pattern bridges policy, AI and Infrastructure as Code inside modern developer workflows.

15:25

Fighting the big ball of mud with business capability mapping

Jen Wallace

Are your systems drifting into a "big ball of mud"? In this talk, I’ll show how business capability mapping can uncover and resolve unclear domain boundaries in large, multi-team projects, without disrupting progress. I’ll share how we applied this approach across 10+ teams to gain alignment, reduce complexity and keep our architecture clean and scalable. Walk away with practical techniques to break free from architectural chaos, keep your teams aligned and your projects on track.

16:20

“Port & Mend" over rewrite - Reducing the cost and risk of dealing with legacy code

Matteo Vaccari  

Many legacy rewrite projects fall short of expectations, while the old systems still keep the business afloat. A practical alternative is to "port & mend" the legacy code. This involves minimal changes to adapt it to new platforms and technology ("porting") and giving it care through fixes, refactoring and testing ("mending"). The result? Maintainable, deployable and updatable code. While rewrites have their place, porting and mending are often less risky, more affordable and now even more accessible with advances in AI. This talk draws on real-world experience to explore the value and practicality of this approach.

14:30

Pairing with the machine: How AI agents are rewriting the rules of software development

Satish Kumar Nadarajan

What if your coding partner could understand your codebase, remember architectural decisions, and adapt to your workflow? This talk explores how autonomous AI agents are transforming software development by acting as intelligent pair programmers. Beyond tools like GitHub Copilot, these agents can analyze project context, suggest design improvements, recall team decisions, and participate in architectural debates. We'll cover key technologies like LLMs, memory systems, retrieval-augmented generation (RAG), and tool orchestration, with real-world VSCode prototypes. We'll also tackle challenges like hallucinations, dependency drift, and trust. Whether you're an engineer, tech lead, or platform builder, this talk reveals how AI-native development is becoming your newest teammate.

15:25

Training without sharing: Is federated learning the answer to data privacy in AI?

Emma Manna Loyarte

In sectors like healthcare, data is abundant but often unusable due to strict privacy regulations. Federated learning offers a promising solution: training models locally on sensitive data and sharing only the learned parameters, not the data itself. This talk introduces the core concepts behind federated learning and explores its practical potential and limitations. Within the scope of natural language processing, detecting domain-specific medical terms in patient records, we will examine how federated approaches can benefit from sensitive datasets without compromising privacy. Through a series of performed experiments, we’ll highlight the strengths and limitations of federated learning, as well as learnings that can help assessing its suitability for other use cases.

16:20

Application-driven infrastructure

Kief Morris

Even with automated delivery pipelines, cloud and infrastructure as code, managing environments and infrastructure is often a major drag for software delivery effectiveness. This talk describes an approach to building infrastructure as composable components that application teams can provision and configure themselves, changing the dynamic away from ticket-driven infrastructure silos.