The more things change, the more they stay the same
Description
A screenshot of a tweet from user Lili Cosic (@LiliCosic). The tweet, in black text on a plain white background, reads: '"Works on my cluster" is the new "Works on my laptop"'. This is a modern take on the classic developer excuse, 'It works on my machine.' The original phrase was used when code worked in a local development environment but failed in testing or production. This updated version reflects the shift to distributed systems and container orchestration. The joke is that even with complex, supposedly reproducible environments like Kubernetes clusters, the fundamental problem of environment-specific bugs and configuration drift persists. It's humorous to senior engineers because it shows that despite decades of technological advancement, from standalone laptops to massive distributed clusters, developers still face the same core challenge of making their code work reliably everywhere
Comments
7Comment deleted
We added so many layers of abstraction - containers, orchestrators, service meshes - that we can no longer blame the hardware. So now we just blame the YAML
We finally containerized the blame: now it’s “works on my 40-node Kubernetes cluster, behind two service meshes and a thousand-line Helm chart” - if prod can’t recreate that, clearly ops chose the wrong YAML whitespace
Remember when we thought Docker would solve 'works on my machine'? Now we've just scaled the problem horizontally across 47 nodes, 3 availability zones, and a service mesh that nobody fully understands
The progression from 'works on my machine' to 'works on my cluster' perfectly captures how we've successfully distributed not just our applications, but also our excuses across multiple nodes. Now when production breaks, we can blame an entire orchestrated fleet of machines instead of just one laptop - truly achieving horizontal scaling of accountability
“Works on my cluster” translates to: I had cluster-admin, zero resource limits, no NetworkPolicy, and egress to the world - so of course it worked
Laptop: 10 req/s. Cluster: 10 req/s over 10 nodes. Prod: 'Why so slow?' - the eternal CAP theorem troll
Translation: “Works on my cluster” = minikube + cluster‑admin + :latest + imagePullPolicy=IfNotPresent; prod with Gatekeeper, quotas, and NetworkPolicy replies, “not in this namespace.”