Once a year we disappear for three days

Once a year, we pack up our laptops and disappear. No client projects. No tickets. No utilization targets. No deadlines. Instead, three days for the things that usually don't fit into everyday work.
We call it Code Retreat. Although by now, the name does a pretty poor job of describing what actually happens there.
It all started with talks

In 2019, we met for the first time in a remote village in the Verzasca Valley. Back then, the format was still fairly traditional: someone knew something exciting and told everyone else about it.
Design systems. Accessibility. Serverless. Frontend performance. Web Components. Text mining. Personality psychology. Metaphors in user interfaces.
In 2020, in the Bernese Oberland, the list of topics got even wilder: Next.js, machine learning, CRDTs, state machines, GitOps, infrastructure as code, D3, CSS, performance monitoring — and a workshop on AI and big data.
- AI.
Three years before it felt like every other LinkedIn post started with “AI is changing everything”. Back then, though, it sat somewhere between “How is beer brewed?” and “Sun & Stars”. Which, come to think of it, is a pretty good snapshot of our Code Retreat.

In 2021, we headed to Amden above Lake Walen. Once again, there was Flutter, Ethereum smart contracts, Web Components, Terraform, module bundlers, visual perception and linguistics. But after three years, we realized that just sharing knowledge wasn't enough for us anymore. We wanted to try things out. And above all, we didn't want to show twelve more slides about an idea on Friday evening. We wanted to build the idea.
So we turned it into a hackathon

In 2022, we changed the format. No more ten individual talks. Instead: small teams and two days.
At Wirkstatt Auboden in the Neckertal, one team tried out new web frameworks like Remix, Solid, Astro and Qwik, and had to answer the question of whether Next.js was still our best choice. Another dug into New Work and our own organization. And because apparently no headless CMS was quite good enough for us, one team simply decided to build its own. Of course.
Why use an existing solution when you can just hole up for two days and build your own CMS?
Since then, that's been more or less the rule: it starts with an idea. Two days later, something has to exist. Not necessarily finished. Not necessarily sensible. But concrete enough that we know more afterward than we did before.
Our sales pipeline was born in a forest hut

In 2023, we were in Laax. Part of the retreat actually took place in a forest hut. At some point, there were Post-its stuck to a glass wall, and we were discussing how our sales pipeline and quoting phase should work going forward. We still do a lot of it that way today.
At the same time, SMARTA-AI came to life. The idea: what happens if we connect an AI model to our internal knowledge from Notion and Slack? So pretty much exactly the question everyone is discussing today under the labels RAG, company AI or enterprise search.
Alongside that, we built an AR experiment, a WebGL visualization of CAD models and improvements to our internal apps. Not all of it lived on. It doesn't have to. Sometimes the most important outcome of a Code Retreat is: that was fun, let's never do it again. Also valuable.
And sometimes a product suddenly appears
In 2024, we locked ourselves in a schnapps distillery in Thurgau for three days. What could possibly go wrong?
There, one team tackled a pretty mundane problem. We host summer parties, apéros and other events. Every time, we need sign-ups. Every time, we cobble together some kind of solution. Google Forms doesn't feel particularly sexy when you send it to clients. Event platforms don't really fit either. And the same problem kept cropping up with our clients.
So the team built a proof of concept for a headless event management system. The prototype later got a name: Eventless. Today, its first clients use Eventless in place of exactly the workarounds that used to annoy us.

So the thing that came out of a distillery, a pile of laptops and probably not exclusively mineral water actually became a product. A real one. For clients. Not a “product” in the sense of three Figma screens and a domain that expires six months later.
In 2025, AI went from topic to tool

In 2025, we went to St. Anton in Appenzell. With a view over the Rhine Valley, one team tested Figma MCP together with shadcn and a custom Next.js setup. The goal: create a design in Figma and generate working code from it as directly as possible. Without writing a single line by hand.
Another team built an internal smartive GPT with access to our own resources. Another experimented with an AI image generator that understands our visual language. Others built a Flutter app for OpenAir St. Gallen and a digital Advent calendar.
And then there was the team that wrote practically no code at all. They wrote a book. AN ENTIRE BOOK. Not “we exported a PDF and now we're calling it a book”. The thing actually got printed. Not on A4 paper from the office printer with a staple in the corner. As a proper paperback. With a spine and everything.
About our culture, self-organization, roles, decision-making processes, rituals and all the other things people keep asking us about: how does this company actually work? To be fair, we ask ourselves that regularly too.
This year, AI goes one level deeper
In 2026, we're heading to Thusis. And if you look at the list of topics, something has changed.
An AI agent is supposed to understand support requests from email, Slack and Notion, and maybe prepare the matching merge request right away. One team is building a possible Harvest replacement. Another is experimenting with local LLMs that first scrub sensitive data from production logs and then send a coding agent off to hunt for bugs.
We're looking at what project teams look like when designers produce code, clients produce code and individual developers suddenly work with a small army of AI colleagues. We're building an AI development platform, an internal stock marketplace and a generator for smartive slides. And we're testing whether AI-simulated users with Playwright and vision models can actually find useful UX problems.
Put the topics from 2019 next to these, and the progression is pretty absurd.
- 2020: What can AI do?
- 2023: Can AI understand our knowledge?
- 2025: Can AI speed up our work?
- 2026: Okay. And do we actually still need teams this big?
That escalated quickly.
But that's only half the story

Whenever we talk about the Code Retreat, we quickly end up talking about the things that came out of it. Eventless. The book. AI experiments. All sorts of stuff that later actually became part of our daily work.
Yet the other half is at least as important.
Seeing everyone again. Not as a small rectangle in Google Meet. Not for 25 minutes between two other meetings. But properly. Being somewhere together for three days. Working with people you hardly ever share a project with. Eating together. Staying up late. At some point, closing the laptop. Cracking open a beer.
And then those conversations happen. About work. About the company. About things on our minds. About families and life. About decisions we probably should have discussed a lot earlier. Some conversations go pretty deep. Others not at all. Some evenings are also just a whole lot of fun. And yes: occasionally rather boozy. Maybe that's part of it, too.
A hackathon at the office wouldn't be the same

Technically, we could of course do all of this at the office. Block two days. Form teams. Order pizza. Go. It would be cheaper. It would be easier. But it wouldn't be the same.
It makes a difference whether everyone heads home after the workshop or whether you're sitting around a campfire with the same people two hours later. Whether you jump straight back into your usual meeting the next morning or have breakfast together and then keep building. Whether you're sitting in a meeting room or in a forest hut in Laax. Or a schnapps distillery in Thurgau.
Maybe it's no coincidence that a few of our best ideas were born in exactly those places.
Eight years later
In 2019, we went to the Verzasca Valley and told each other about design systems, serverless and accessibility. In 2026, we're heading to Thusis to discuss how AI is changing project teams, support processes and software development.
In between, there have been quite a lot of prototypes. One real product. AN ENTIRE DAMN PAPERBACK. Countless technical experiments. A few ideas that vanished completely. Post-its on a great many walls. Deep conversations around the campfire. Lots of wonderful moments. A few evenings that were probably less productive toward the end.
And above all, eight years in which we've taken a few days every year to think together about the things that always get pushed aside by something more important.
What will come out of Thusis this year? No idea. That's exactly why we're going.

Written by
Josh Wirth
Got a similar challenge on your plate? Let's talk.





