Ten Years of Deploying to Production

3 days ago

In 2018, the company had an operations team (Ops) managing production deployments, which were infrequent (once every two weeks).
The Ops team had a tool for spinning up VMs, which was crucial for tasks requiring heavy resources like GPU training.
Deployments were rigid; errors could delay deployments by weeks unless Ops was accommodating.
The data science team faced challenges with misbehaving models in production, leading to customer complaints.
There was no formal PR review process; code was often edited directly on VMs and pushed to GitHub haphazardly.
The author took initiative to improve DevOps practices: setting up an internal PyPi, using git tags, creating Chef recipes, and establishing PR reviews.
By 2026, the focus shifted to platform engineering, emphasizing developer experience, quick CI/CD, and resilient production systems.
The contrast between 2018 (rigid, ops-centric) and 2026 (flexible, developer-centric) highlights the evolution in deployment and operations culture.

Hasty Briefsbeta