How do you know it didn't quietly break something for a customer you'll never hear from?

Most of us answer that by running a couple of examples by hand and trusting our eyes. There's a better answer, and it's the first thing I'd add to any team shipping LLM features. Evals.

They look like tests. They aren't. A test asks whether the answer is correct. An eval asks how good it was, and "good" is something you have to define.

In this session we'll build them from nothing. Where the cases come from, how you score an answer with no single right form, and when it's safe to let one model grade another. Then we'll change the prompt and watch exactly which cases move. The code is C#, but nothing here is C#-specific.

If you've shipped something powered by an LLM and you can't tell whether it's getting better or worse, this session is for you. Even if the word evals is new to you, you can expect to go home able to write your first one on Monday.

Talk (60 min)

You Changed the Prompt. What Did You Break?

AI Fundamentals

So what’s real, what’s exaggerated, and what’s next? Join Richard Campbell for an insightful look at the current AI surge, why it matters, and where the real success stories are happening today. Explore the opportunities, the challenges, and the eventual end of the hype cycle—so you can separate the promise from the noise and prepare for the future of AI in business and development.

After the AI Hype – What’s Real, and What’s Next

Ethics

People

Security

It’s not because we don’t want to develop accessible websites and applications.
But we often lack knowledge and perspective. We build based on our own experience because that’s what we know. In this talk, you’ll get to experience how our applications work for someone with a screen reader, low vision, or limited motor control, and feel firsthand the challenges they face.

The talk combines knowledge, practice, and lived experience to raise your awareness of accessibility on the web and how small changes can make a huge difference.

You'll see the most common mistakes in action, who they affect, and how to fix them. So come experience the world through someone else's senses, and learn how to be a more accessible developer.

Let's break some WCAG rules

Accessibility

By now most of you have set up your own coding agent, maybe with sub-agents, custom instructions, and skills. But how do you go from a useful assistant to a system that runs autonomously, overnight, unsupervised, in production?

I'll walk through how we built a fully autonomous development workflow at Nordic Corporate Bank where an AI agent picks up a task, implements the solution, runs tests, creates a pull request, and reviews its own code, all before a human ever gets involved.

When developers open their laptops in the morning, finished pull requests are waiting for review. Build failures and production exceptions are automatically routed to the agent, which fixes them right away. Developers spend their time on architecture, design, and the problems that actually require a human, plus building the system that builds the system.
This is not a proof of concept or a keynote demo. This is code deployed to production every day. In a bank.

How We Built an Autonomous Development Workflow at Nordic Corporate Bank

AI-Assisted Development

Keynote

In this session, we’ll cut through the hype and focus on practical, developer-tested strategies for getting the most out of AI in your workflow.

You’ll learn how to master prompting and context engineering, make the most of long context windows, streamline debugging and testing, and integrate AI into your daily development practices without sacrificing code quality or security. We’ll also explore how emerging standards like Model Context Protocol (MCP) unlock new workflows by connecting your AI assistant directly to tools such as GitHub, Slack, Playwright or Figma — turning it from a code generator into a true development teammate.

Whether you’re new to AI-assisted coding or already experimenting with it daily, you’ll leave this talk with concrete ideas and techniques you can apply immediately to level up your development process.

10 tips to level up your ai-assisted coding

This session explores familiar patterns that slowly turned simple web apps into things that feel harder than they should be. Not because anyone was careless, but because the trade offs only became visible years later.

Using well known websites as reference points, we look at the visible effects of these decisions at scale and recognise the same patterns in our own work. No deep technical dives, no frameworks, and no demos. Just stories, recognition, and the shared relief of realising this was never just your team.

If you have ever looked at a web application and thought “why is this so complicated”, this session is for you.

At the time, this seemed like a good idea

TBA - Maddy Montaquila

TBA - Kyle Kotowick

But we're not talking theory here. This exploration will be a hands-on experience. Each feature will be brought to life through real-world examples, all done in a live coding session. Join in as we shine a light on these secret TypeScript treasures and discover how they can elevate your coding skills to new heights. Recursive types, type permutations and a great new way to make even primitives way more type save and expressive are only three of the cool topics we are going to cover during this talk.

A Deep Dive into Advanced TypeScript: A Live Coding Expedition

Functional Programming

JavaScript

Programming Languages

This session will guide you through using OAuth 2.0 in environments where security is paramount. We will dive into the latest specifications designed to enhance OAuth 2.0's security capabilities. Topics include advanced security features like Resource Indicators, JAR, PAR, and DPoP. By the end of this session, you will possess a comprehensive understanding of the security aspects of OAuth 2.0, equipped to implement it in high-stakes settings.

Supercharging OAuth 2.0 security

Architecture


When systems handle decisions you used to own, how do you stay sharp? Where does your expertise truly matter? What are you responsible for?

This isn't a cautionary tale about AI. It's a practical question about technical leadership when the tools become increasingly capable.

Modern pilots work alongside sophisticated autopilot. They understand what the system does. They know when to override it. Their expertise gets reapplied, not replaced.

Using real stories from work with technical leaders at organisations like NVIDIA and Microsoft, we'll explore what this shift looks like in practice and what it means to be genuinely good at your job now.

Who's Flying the Plane?

Work skills

Soft Skills

This fast-paced talk focuses on practical exploitation and real-world attack chains in AI-powered web apps. You will learn how attackers build rapport with LLMs, manipulate behavior, and bypass restrictions.

We will demonstrate how seemingly benign vulnerabilities can be chained together to steal and exfiltrate sensitive data through indirect prompt injections, malicious rendering techniques, and abusive tool calls.

Finally, we will break down the reality of AI risk and show exactly how to build the guardrails you need to stop these attacks.-

Breaking the AI Cage: Prompt Injection Is Inevitable. Impact Is Optional

AI Agents

AI Tools and Data

This talk shows a better approach: shift the focus from chasing specific attacks to teaching solid defensive coding principles. We'll build security knowledge directly into the agent — using skills, hooks, and standards — so it looks up the right guidance, plans for it, and reviews its own work. You'll leave knowing how to make your agent produce secure code far more often, by default.

Teaching your coding agent defensive coding

Everyone is going to be redundant! AI will fail! AI will ruin the internet! AI can't do anything!

It's disorienting and stressful trying to work out what a reasonable strategy should be to move forwards with this technology that appears to shift it's goalposts almost daily.

In this session, we'll talk about understanding what AI really is, how to form a reasonable, governance friendly strategy around it, and what the sensible bets for the future are in the space.

We'll do a deep dive into ethics, the human cost of content moderation, and having to be confronted by the reality of technical change in real-time.

A survival guide to getting the most from the most controversial technology in a decade.

Making Sense of the Dissonance in AI

Join Richard Campbell as he explores the challenges and potential of putting data centers into orbit. The balance of cost, performance, latency, power, cooling, and reliability makes for a difficult mix - but the potential for the future is significant!

Above the Cloud: Building Data Centers in Space

Cloud

Space

Despite how important and prolific software and programming have become over the last 50 years the general public still have a very loose understanding of what software actually is.

This hasn't been lost of the people that build the software, who often slowly try and capture themselves in the work that they do.

In this talk we'll take a look at some examples through history of seeing the humans behind the code they write as they imprint themselves on their work and we'll consider if it's possible to deliberately make art by manipulating the form and design of software itself.

A talk for everyone wanting to be seen in the work, or understand their own creative urge.

Meditations on Code As Art

So if the harness is solved, what's left to build? Enter Agent Skills.
What differentiates my agent from yours is not the loop — it is the skills and tools we equip it with. A generic agent plus the right skills becomes a legal analyst. The same agent plus a different set of skills becomes a deployment engineer, a financial controller, or a research assistant. Skills are how you go from a tool-calling chatbot to an agent that actually does the job.

In this talk, we will use the Claude Agent SDK as our harness — solved, off-the-shelf, no reinvention required — and focus all our energy on the layer that actually matters. You will get a clear mental model of the skill + tool stack, live demos of building skills two ways (with the skill creator pattern, and by capturing real workflows as reusable capabilities), and a practical playbook for composing skills with MCP servers and CLI tools to build the "everything agent" your team actually needs.

Stop building harnesses. Start building skills.

Build Skills, Not Agents

This talk explores different paths - from treating models as replaceable infrastructure to deeply integrating a single provider or adopting a complete agent platform. We’ll compare their trade-offs in speed, reliability, control, portability, and long-term sustainability.

The goal is not to eliminate dependencies, but to make them explicit, intentional, and survivable.

The AI Architecture Decision: What Should Your Company Own?

However, agents can reach objectives in ways the task author did not intend. Our own testing shows agents seeking network access when access shortens the path, modifying the environment instead of completing the task, and trying to bypass tool restrictions to achieve their goals. It has also been widely reported that agents under a cybersecurity benchmark obtained remote code execution on a host in the evaluation environment and retrieved the answer rather than solving the challenge (see the Open AI / Hugging Face Incident, the Anthropic Incidents, and the UK AISI incident). A single-outcome score cannot distinguish that outcome from a genuine solution.

This session shows how recording the evaluation runs as a graph instead of a binary outcome, and observing agent behavior rather than agent-reporting can be an effective way to measure and evaluate capabilities in agents. This approach can provide an alternative to binary evaluations and overcome some of the limitations of black-box, binary-based, evaluations.

This session also covers what information the graph can provide about containment. Restricting a tool restricts a name from a toolcall, not a capability. An agent denied a fuzzing binary in our testing rebuilt equivalent behavior from a shell loop and curl, and at the level of the activity graph those two runs are similar while a name-based control sees only a block followed by permitted commands.

The remainder of this session focuses on environmental design: separation between evaluator and agent, egress handling and monitoring, evaluation logic that does not rely on the agent's own report, and review of results to identify hallucinated success. These approaches ensure that the benchmark provided is representative of the capability being evaluated rather than the model’s ability to cheat on that evaluation.

Overall, this session presents an approach to evaluate agents based on monitorable behaviors grounded in a knowledge graph, and presents findings from our own internal testing relevant to current events in AI safety and security, and provides other best practices for ensuring benchmark reliability and validity.

*all em dashes were human generated.

Measuring Agents: Evaluation Environments and Activity Graphs

This talk follows an input through the entire AI pipeline, from normalization and tokenization to embeddings, guardrails, foundation models, and tool execution, to reveal where understanding begins to diverge. We'll show how attackers exploit gaps using TokenBreak and EchoGram, break systems with prompt injection and other adversarial examples, and hide all of these attacks from unsuspecting users.

But interpretation failures do not end at inference time. We will also examine how poisoning and backdooring can implant attacker-chosen meanings into a model, allowing seemingly benign inputs to trigger hidden behaviors long after deployment.

We'll close by turning these attacks into practical design principles for building AI systems that remain secure, even when their components disagree about what they are seeing.

Now I see you, now AI don't: Attacking the gap between Human and Machine Perception

In this talk, we’ll cut through the AI hype and look at the real evolution of our craft. We'll explore why raw code is becoming a commodity, why human "taste" and system architecture are skyrocketing in value, and how engineers are becoming masters of autonomous agents rather than victims of them. Join us for a witty, honest look at the bad, the good, and the future of software engineering leadership.

The Death of Coding (Part 7): Why Software Engineers Are Still Here

The industry has come up with completely new ways of working with graph data, which we’re going to take for a test drive in this talk! We’ll check out what changed in Neo4j, the only major graph database that is still with us after 10 years. Then we’ll discuss in-process analytical graph databases, which is an idea that didn’t even exist until just a few years ago. In particular, we’ll test KuzuDB and Memgraph with some demos and discuss when to use them. We’ll look at graph processing engines, like PuppyGraph, and see what makes them different from databases. Finally, to understand why graph databases have become more important than ever in the age of LLMs, we’ll build an AI agent augmented by a knowledge graph served from FalkorDB.

Modern Graph Databases

Database

In this session, I'll share what I've learned from using Claude Code daily and teaching it to developers in workshops. We'll go into the features that change how you work. Things like shaping Claude's behaviour through 

, building your own commands and skills, automating feedback loops, and understanding the small configuration details that most people overlook.

We'll also look at use cases beyond software development. We'll use Claude Code as an actual AI-assistant, not only a code-assistant.

This isn't a getting-started guide. It's a tour of the parts of Claude Code that only become visible once you stop treating it like an autocomplete tool.

If you use Claude Code but feel you're not getting the most out of it, this session will help you see what it can really do.

What Claude Code Can Do That You Haven't Tried

Once the demo is over, teams face questions that prototypes rarely answer. How do we evaluate non-deterministic outputs? How do we protect sensitive data and manage new attack surfaces? And how do we introduce security and governance without bringing delivery to a halt?

Drawing on practical experience designing and delivering enterprise AI solutions, we will explore the journey from proof of concept to production. We will discuss recurring challenges and trade-offs across architecture, evaluation, security, and operational readiness, without relying on a particular technology or reference architecture.

Attendees will leave with practical considerations for identifying risks early, avoiding common failure modes, and making pragmatic engineering decisions that balance delivery speed with enterprise security and reliability.

From AI Prototype to Production: Lessons from Enterprise AI Projects

Machine Learning

As the leader of Microsoft's AI Red Team operations, Tori Westerhoff has spent the last several years testing frontier models, copilots, agents, and AI-powered products before release. The work ranges from classic cybersecurity attacks to newer categories of failure: jailbreaks, prompt injection, deception, manipulation, and emergent behaviors that don't fit neatly into traditional security frameworks. 


This session is an honest look at how that work actually happens —you'll leave with a clearer picture of how modern AI systems fail, why some of the most effective attacks target come from outside security and tech, and how to bring a red teaming mindset to building and integrating AI responsibly in your work.

A realistic sense of how AI red teaming differs from traditional security testing

A few representative failure modes from real-world generative AI systems

Why interdisciplinary approaches sharpen a red teamer’s ‘offense is defense’ security

A more grounded mental model for reasoning about AI security risk with open source tool recommendations

NDC { Porto }

Agenda

09:00 - 10:00 (UTC±00)

Keynote

Nick Chapsas

10:20 - 11:20 (UTC±00)

10 tips to level up your ai-assisted coding

Aleksander Stensby

Let's break some WCAG rules

Elise Kristiansen

Who's Flying the Plane?

Amit Grinvald

11:40 - 12:40 (UTC±00)

After the AI Hype – What’s Real, and What’s Next

Richard Campbell

TBA - Maddy Montaquila

Maddy Montaquila

A Deep Dive into Advanced TypeScript: A Live Coding Expedition

Christian Woerz

13:40 - 14:40 (UTC±00)

How We Built an Autonomous Development Workflow at Nordic Corporate Bank

Hallstein Brøtan

Making Sense of the Dissonance in AI

David Whitney

Supercharging OAuth 2.0 security

Philippe De Ryck

15:00 - 16:00 (UTC±00)

Breaking the AI Cage: Prompt Injection Is Inevitable. Impact Is Optional

Stök .

Joona Hoikkala

At the time, this seemed like a good idea

Kevin DeRudder

TBA - Kyle Kotowick

Kyle Kotowick

16:20 - 17:20 (UTC±00)

Teaching your coding agent defensive coding

Erlend Oftedal

You Changed the Prompt. What Did You Break?

Gui Ferreira