Replacing an operating system across an organization is now a solved problem: describe every machine as code with a tool like NixOS, and you get the same reproducible system on every box, with no manual patching or drift. That’s the easy part of DAWO, the Dutch government’s project to replace its Microsoft desktop stack. The much harder problem, according to developers debating the project, is replacing the everyday tools people already know how to use, like a shared spreadsheet or a chat thread.
Table of Contents
- What DAWO is
- Why declarative infrastructure closes the lock-in gap
- Three governments, the same architecture, independently
- The pushback: the office suite is the real wall
- A cleaner pattern for running this at scale
- An AI policy wrinkle nobody planned for
- What this means for your own stack
- FAQ
- Sources
What DAWO is
DAWO is a Dutch government initiative building a replacement for the standard Microsoft desktop stack, structured as separate, replaceable building blocks rather than one monolithic product: an operating system (DAWO-NixOS, “installation building blocks for a reproducible workplace”), cloud infrastructure, collaboration tools, and an AI layer. The project states five goals: digital autonomy, better collaboration between government and society, security and data protection, innovation, and making government IT systems easier to inspect and verify. Everything, including the code and roadmap, is public.
The pitch is straightforward: don’t depend on any single vendor’s product decisions for how your civil service does basic computing. What makes the project worth a developer’s attention is the architecture, and the places where a large, technical audience of developers thinks that architecture will and won’t hold up.
Why declarative infrastructure closes the lock-in gap
The operating system layer, DAWO-NixOS, is built on NixOS, a Linux distribution where you describe an entire machine, packages, services, configuration, in one declarative file, and the builder produces the exact same result every time. There’s no manual provisioning step where machines drift apart over time, and no dependency on a vendor’s patch cadence to keep systems consistent.
That property, reproducibility from a text description, is why NixOS appeals to any organization that wants to stop trusting someone else’s roadmap for its core infrastructure. It’s also why platform teams outside government reach for the same tool: once your fleet is code, replacing the vendor underneath it becomes a configuration change instead of a re-provisioning project.
Three governments, the same architecture, independently
Developers in the discussion pointed out that DAWO has company: France already ships its own hardened NixOS build for government agencies (a project called Securix, with a companion example for building office deployments on top of it), and that Germany runs a comparable open desktop and collaboration effort called openDesk. None of these three governments appears to be coordinating on a shared codebase; they arrived at the same answer, declarative, auditable infrastructure, on their own.
That convergence matters more than any single project’s marketing copy. When three separate public-sector IT organizations independently converge on the same approach, it’s worth asking whether your own infrastructure has the same property, or whether replacing a vendor today would mean weeks of manual reprovisioning.
The pushback: the office suite is the real wall
The strongest counterpoint in the discussion wasn’t about NixOS at all. Developers argued that swapping the operating system was never the hard part; the hard part is the software people spend their day in. One commenter put it plainly: LibreOffice and Collabora still can’t match Microsoft 365’s real-time collaboration, co-editing a spreadsheet, commenting, syncing changes across a large org, without real friction for ordinary teams.
Another developer pushed further, arguing that cloning Office feature-for-feature will always land short of the original, estimating roughly 80% parity at best, and that the more useful goal isn’t a lookalike of Excel but a rethink of the workflow itself. That’s a harder problem than swapping a kernel and a package manager, because it’s a habit migration for every employee with muscle memory built around a specific product, not a technical one.
This is the part of the story that gets skipped when a migration like this is covered as a political statement. The technical risk is low and well understood. The organizational risk, getting thousands of people to give up tools they’re fluent in, has no clean engineering solution.
A cleaner pattern for running this at scale
One detail from the discussion is worth stealing regardless of what stack you run. A developer sketched out how they’d operate a fleet like this at scale: make every machine’s filesystem read-only, push configuration from a single central source, and serve the package store itself over the network (for example, over NFS) rather than storing it locally on each box. The result is a fleet with nothing to patch by hand and nothing left to drift, since no machine holds local, mutable state that could diverge from the definition.
That’s a stronger version of “infrastructure as code” than most companies run in practice: servers defined declaratively, but still left to accumulate local state over time. Pairing a declarative build with an immutable, centrally served runtime removes that entire category of “works on this one box” incidents.
An AI policy wrinkle nobody planned for
One counterpoint from the discussion is easy to miss but genuinely interesting: developers flagged that the project’s code, mirrored to Codeberg, may run into that platform’s new policy restricting AI-generated contributions to open source projects. DAWO lists an AI layer among its building blocks and frames the whole initiative around verifiability, while the code-hosting ecosystem it depends on is moving to restrict AI-authored commits. That’s a real, unresolved tension for any organization building “verifiable, open” infrastructure while also using AI to build it faster, and worth checking for before you’re a few hundred commits into a similar project.
What this means for your own stack
If you’re the person your company turns to when someone raises “we’re too dependent on vendor X,” the lesson here is to be honest about where the real cost sits. The infrastructure layer, servers, OS images, configuration, is usually the tractable part: tools like Nix, Ansible, or a solid golden-image pipeline can get you real vendor independence there in months, not years.
The tools your employees live in all day are the hard part, and that’s where independence projects tend to stall. My advice: don’t lead a pitch like this with the infrastructure migration. Lead with a realistic estimate of how long it takes people to give up a product they’re fluent in, because that number, not the technical one, decides whether the project survives contact with the organization.
FAQ
Is NixOS a realistic replacement for Windows in an organization?
At the operating system layer, yes. NixOS lets you describe a machine’s entire configuration as code and reproduce it exactly, removing drift and dependency on a vendor’s patch cycle. Government IT projects in the Netherlands, France and Germany have independently adopted this approach for that reason.
What’s the hardest part of replacing Microsoft 365 with open source tools?
Not the operating system, the collaboration suite. LibreOffice and Collabora still struggle to match Microsoft 365’s real-time, multi-user editing without friction, and cloning Office’s feature set doesn’t solve the deeper problem of employees’ existing workflow habits.
How do you manage configuration drift across a large fleet of machines?
Make every machine’s filesystem read-only, push configuration from one central source, and serve the software package store itself over the network instead of storing it locally. That removes the local, mutable state that normally causes machines to drift apart.
Sources
- Article: DAWO, by the DAWO project.
- Discussion: Hacker News thread, for the counterpoints on the office suite, the France/Germany comparisons, and the read-only fleet pattern.