AWS Summit
Zurich 2026
Your next steps in AI and Agentics: from pilot to production, safely and at scale.
We’ve curated the best of the day in this Recap.
AI is changing the foundations of business – and agentic AI is changing the game even faster. Speed is of the essence, so AWS Summit Zurich focused on showing how you can move AI transformations out of pilots and into scaled-up production – without losing control over security, compliance and effective data management.
Our morning keynote described how Geneva-based investment firm Pictet, founded in 1805, protects its "digital fortress" while still innovating. And there’s rich, actionable insight from other industry leaders, including Roche Diagnostics (taking agents from prototype to production in under six months), Zurich Insurance, Syngenta and Swiss Post.
Read on to see how, and why, Swiss businesses are building their AI on AWS
You’ll hear from customers achieving breakthroughs by migrating workloads to the cloud and integrating AI into their operations. Pictet's group CTO, for example, explains how a 220-year-old bank is successfully adopting AI without compromising client confidentiality or operational security.
And you'll see how AWS is investing long-term in Switzerland, with sovereignty and AI at the heart of how we work. Discover where agents are heading, and how AWS is removing every barrier between your imagination and reality.
At AWS, we believe builders should have the freedom to invent anything.
And today, AI agents are transforming what's possible. This keynote will show you how AWS makes it easier than ever to build, deploy, and scale agents; and where agentic AI is delivering results today.
You'll see how agents can change the way your teams work, build software and stay secure, and what it takes to move your own agents from prototype to production.
“Agents are already building applications, answering customers and making decisions. Compounding momentum is one of the generational shifts of my lifetime."
Rory Richardson: Director, Next Generation Developer Experience, AWS
Register to watch the full session
Getting agents from pilot to production is now the priority.
Demos are simple; compliant rollouts at scale are not. Roche Diagnostics took its software-lifecycle agents from prototype to production in under six months by building in observability, evaluation and strict human approval steps.
Build one shared, governed platform to let teams move fast.
Syngenta built its agent platform once, sharing identity, integration and governance. Employees have since created over 800 knowledge bases, and a new agent can be deployed in under thirty minutes.
Security, compliance and Swiss data residency come first.
Pictet will not trade safety for speed. No compromise rules include: inference stays in Europe (Switzerland for critical workloads); data never trains anyone else's model; every call traces to a person.
Business experts can work alongside developers.
Kuehne+Nagel put Amazon Quick in front of 21 experts and managers. A first working prototype appeared within four hours, and became the precise specification that pro developers then scaled. The future is collaborative.
AI agents can modernize your legacy systems and run operations.
The Swiss Federal Office of Energy is using Kiro to refactor 200,000 lines of legacy pipeline configuration into tested Python, while Swissport uses AWS DevOps Agent to investigate alarms automatically.
"At AWS, we measure our success on what you build and how this collectively shapes the future of Switzerland."
Felix Schoenherr, Country General Manager, AWS, Switzerland
For the Builders
Industrial & Manufacturing
Where AI meets the physical world: reliable quality, smoother logistics and faster decisions, from the factory floor to the supply chain.
Data & Intelligence
Your data already holds the answers. See how to make it AI-ready, keep it governed and put transformative insight into everyone's hands.
AI & Agentic
Move from promising pilots to agents your organization can rely on: the platforms, models and guardrails that make AI work at scale.
Explore a new world of creativity and productivity – and learn from people already driving change.
Learn more
Discover the tooling and services which will help you build agents properly, modernize legacy systems and unlock your imagination.
Building an agent that works in a demo is easy – part of the value is that we can all apply agents to real-world challenges with minimal technical knowledge.
Making it dependable for thousands of people is much harder. Between pilot and production sits real complexity: identity and access, governance, memory use, and the inevitable issue of cost and connections to the systems and knowledge your business already runs on. Our AI & Agentic sessions will show you how to cross that gap. Start small, build on a shared platform, choose and tune the right models with freedom, and give every team a safe route from first idea to live agent.
Key takeaways
Scaling any product is not just about repetition – it’s about security, assurance and confidence in different situations.
This session sets out practical rules for running agents across a large organization, from scoping and observability to multi-agent design and testing. Zurich Insurance then demonstrates the group-wide platform that lets its teams build agents quickly, with governance built in.
"When governance is built in, teams never have to follow a separate process."
Maria Apazoglou, Group Head of AI Engineering and Platforms, Zurich Insurance
Who should watch?Architects and platform teams scaling agents across an enterprise.
What you'll get by the endA set of working rules for building, securing and scaling agents.
Work backwards from the user and give each agent a narrow, defined job.
Split any overloaded agent into specialists once tools and tasks multiply.
Isolate every session, then add memory so the experience stays personal.
A platform team can carry the shared load so that use-case teams move faster.
AI is moving from answering questions to carrying out complex, multi-step work. Anthropic and AWS explain what that shift means for enterprises.
They then set out the routes to using Claude on AWS, including options that keep data inside your AWS boundary, and close with results from BMW and Novo Nordisk.
"Choosing where inference runs has become a strategy decision."
Lingli Zheng, Sr. AI GTM Lead, AWS
Who should watch?Leaders choosing how to bring Claude into their organization.
What you'll get by the endA clear map through the main ways to deploy and use Claude on AWS.
Agents are moving from coding help to wider knowledge work.
Back the hard use cases; the gains grow as models improve.
Choose your route according to where data must stay and who processes it.
Where you run inference is now a business decision as well as an infrastructure choice.Watch the full session
An agent is only useful once it can reach your tools, your knowledge and the web, with the right permissions.
This session shows how AgentCore Gateway makes those connections, and introduces managed knowledge bases and web search. Syngenta then explains the shared platform behind more than 800 employee-built knowledge bases.
"Producing an answer is not enough: AI has to find, reason and then act."
Stefano Pecoraro, Head of R&D IT Engineering Technology, Syngenta Group
Who should watch?Architects connecting agents to enterprise systems and knowledge.
What you'll get by the endPractical approaches for giving agents secure access to tools, data and the web.
Prompts are easy to produce. The hard part of enterprise agent deployment is access and identity.
Route every tool through one gateway so that policy is applied consistently.
Carry the user's identity through every step to maintain permissions.
Centralize governance, then let domain teams build on shared foundations.
Swisscom's contact-center voice bots need answers fast to deliver to customers, so the team fine-tunes small models of its own. But as more business units came on board, costs and release times climbed.
See how sharing one base model across many lightweight adapters cut GPU instances by about 60% and made deployments about 30 times faster.
"Nobody on a phone call should wait minutes for a reply."
Basil Fabris, Data Scientist, Swisscom
Who should watch?ML engineers and platform teams running fine-tuned models.
What you'll get by the endA proven pattern for serving many fine-tuned models on shared infrastructure.
Fine-tune small adapters, not whole models: there’s far less information to train on and store.
Serve many adapters from one base model on a shared endpoint.
Separate model releases from infrastructure releases so data scientists can deploy.
Keep each tenant's models, storage and permissions isolated as you scale.
AI is only as trustworthy as the data it ingests, and in Switzerland that means control is a priority.
Where is the data held? Who may see it? Where does inference run, and can every answer be traced and understood?
Our Data & Intelligence sessions tackle the sovereignty, governance and quality challenges at the heart of scaled AI functionality. And then you’ll see the payoff: data that is easy to find, safe to share and ready for AI, so people across the business can have unrivaled access to actionable commercial insight.
How do you transform a data platform into a foundation for a data-driven business?
Swiss Post's logistics data warehouse steers its sorting centers every day. See how the team modernized it on AWS without switching it off, then added shared AI building blocks so data science and agent projects start faster.
"Data, like a pit stop, is not the whole race, but it can decide the result."
Marcel Neidinger, Solutions Architect, AWS
Who should watch?Data, analytics and platform leaders modernizing critical systems.
What you'll get by the endA practical blueprint for turning a legacy data warehouse into an AI-ready platform.
Measure “time to data”: it’s slow access, not slow technologies, that delay projects the most.
Simplify your tool stack. Meet developers where they already work with the tools they’re familiar with.
Build data quality checks around what the business already calls good data – so you can consistently aim higher.
Be ambitious: plan data and AI projects to deliver in months, not years.
In a regulated business, every new AI idea raises the same questions about risk, data and control. Pictet Asset Management answers them once, at platform level.
See how shared building blocks give developers and business users the freedom to create their own agents, while inference stays in Europe and every call is securely logged.
"Each project, whether it worked or not, funded part of the platform."
Edilton Dos Santos, Head of Platforms, Cloud and Tech Services, Pictet Asset Management
Who should watch?Business, technology and risk leaders in regulated industries.
What you'll get by the endA model for scaling agents safely by governing the platform as a whole, freeing project developers to be creative.
Use each early project to pay for a piece of the platform.
Validate the platform once so later solutions reach production much faster.
Expose business capabilities as reusable building blocks that any agent can call.
Balance speed with safety: aim to grow value faster than cost.
On the factory floor and across the supply chain, small gains matter: less scrap and waste, predictive maintenance for fewer delays, faster decisions on both the shop floor and in the management suite.
But machines speak different languages, costs are rising and experienced people are retiring. These sessions will show you AI at work in real operations, from predicting the quality of every part to automating the paperwork that slows logistics. The common thread is practical: get the data foundation right and keep experts in the loop who can shape tomorrow’s tools from hard experience, and you’ll transition to AI-powered working with confidence.
Teams want AI that knows their business in depth (and doesn’t get sidetracked on the rest of the internet). IT wants AI it can govern. Amazon Quick aims to give you both: one workspace that connects all your data sources, answers questions and takes action.
In this session, you’ll get live logistics demos, then hear how Kuehne+Nagel's own experts built a working procurement app without writing any code.
"Quick takes you from a question to an answer, then on to an action and a result."
Jerome Bachelet, Senior Solutions Architect, AWS
Who should watch?Business and operations leaders who want AI without effort.
What you'll get by the endA clear view of what Quick does and how to start building.
Start with your most painful process, not the most impressive use of technology.
Let the people who run a process build the first version to improve that process.
Treat the prototype as a precise specification for developers to scale.
One connector, each person's own credentials: governance carries through to every app.
Swiss manufacturers face rising costs, demand to use recycled material and a wave of retiring experts.
OST and AWS demo a working solution mitigating these challenges on real injection-molding machines: featuring live quality data for every part, with AI suggesting the cause when something goes wrong.
Three engineers new to AWS built it in 12 weeks.
"The real value is getting from data to decisions."
Oliver Weissenborn, Team Leader Injection Moulding, OST
Who should watch?Manufacturing leaders and engineers exploring AI on the shop floor with a focus on value.
What you'll get by the endA reusable blueprint for connecting machines, data and AI in production.
Build a reliable data foundation before scaling any AI use case.
Give every machine message context: unit, timestamp, source and a part's serial number.
Keep pass or fail decisions based on hard rules; use AI to suggest root causes.
Treat AI as a boost to engineering expertise, never a substitute for it.
Our Builder sessions are for the people who build and run software or maintain the infrastructure that helps you do so. We follow the whole lifecycle: see how agents can untangle undocumented legacy systems, turn specifications into tested code and investigate incidents before you've opened your laptop.
Look inside an agent to see how runtime, tools, memory and policy fit together, and find out when plain, predictable code is still the better choice. The way we work is changing – this track is your perfect primer.
Most agent projects stall between prototype and production. This session works through the four things that decide whether yours ships successfully: operations, data and context, reliability, and trust.
Roche Diagnostics then show how they took a nine-agent software-lifecycle system into production in under six months, in a typically highly regulated environment.
"A single, centrally governed definition of the process reaches every agent at once."
Alberto Antoñanzas, Senior Architect, Roche Diagnostics
Who should watch?Builders and architects taking agents beyond the prototype.
What you'll get by the endA four-pillar checklist for getting an agent safely from proof of concept to production.
First, decide whether a ready-made agent will do 90% of the job before building your own.
Focus on observability from day one: you cannot fix what you cannot trace.
When an agent gets something wrong, check its data before blaming the model.
Use easily comprehensible code for predictable steps and keep people in the approval loop.
The Swiss Federal Office of Energy had years of business logic locked inside visual data pipelines that few people understood.
This session explores how a small team used Kiro to reverse-engineer those legacy systems, rebuild them as tested Python and keep national geodata flowing throughout. Includes a practical guide to modernization patterns.
"Speed was only part of it; the aim was a system the team truly understands."
Olivier Monod, Geoinformation Specialist, Swiss Federal Office of Energy
Who should watch?Developers and data engineers facing a legacy modernization.
What you'll get by the endA repeatable method for modernizing legacy pipelines with AI, without stopping your services.
Understand why you are modernizing before choosing whether to rehost, replatform, refactor or reimagine.
Let AI read your legacy system first and write down what it does.
Replace one piece at a time and take the time to compare new output against old.
Automate documentation and specifications so they stay current with every change.
What actually happens inside an AI agent? This deep dive walks through the loop of reasoning and tool calls, then the parts needed to run it safely in production.
Hilti shows us the principles in practice: an analytics agent for its ON!Track asset software that writes code to answer customers' questions.
"Doing the sums is a job for code, not for a language model."
Brijesh Modasara, Lead AI Solutions Engineer, Hilti
Who should watch?Business leaders and engineers who want to understand how agents really work.
What you'll get by the endA working mental model of an agent and the controls around it.
An agent loops between a model and tools until the goal is met.
Keep raw data out of the model; let generated code do the analysis.
Guardrails govern what agents say; policy governs what they can do.
Give each application its own gateway and policy, rather than one per tool.
It’s 3am and you get paged with a problem. AWS DevOps Agent investigates the alarm for you, so a likely root cause and a fix can be waiting as soon as you open your laptop.
Watch a full incident demo, hear how Swissport wired the agent into its ticketing across its AWS accounts, and see what else is new in DevOps process optimization.
"Engineers can now begin with context instead of an empty ticket. Those engineers still make the call; the agent speeds up the analysis."
John Windsor, Head of Cloud Infrastructure and Architecture, Swissport
Who should watch?DevOps, SRE and cloud operations teams on call.
What you'll get by the endA clear picture of how DevOps Agent investigates incidents and where it fits.
The agent investigates in parallel across logs, deployments and infrastructure changes.
Bring findings into the tickets where support teams already work.
Correlate repeat alarms centrally to cut duplicate tickets and alert fatigue.
AWS DevOps Agent also helps with migration planning, code review and cost reports.
Watch the full sessions from AWS Summit Zurich, now available on-demand. Go beyond the highlights with 12 sessions featuring insights and perspectives from AWS leaders.
Watch on demand
We'll see you next year when AWS Summit Zurich returns on 8 September 2027. Save the date and join us for another day of ideas, insight, and inspiration from the people building what comes next.
