I’ve been a software engineering of varying degrees over the past 15 years or so, so I’ve seen some shit. The more senior you get the more shit you see.
What will follow will be a series of articles about situations I encountered during my work where I was able to introduce significant cost savings by doing like-for-like replacements of tech or infrastructure.
This ties in somewhat with my writing here about hosting K8S on KubeKitten.
I’ll start with some highlights on some infrastructural cost-saving measures that had low technical impact but a significant difference on runway for early-stage startups, covering these topics
Replacing Datadog with OpenTelemetry
Using Postgres instead of OpenSearch.
Switching from LangSmith to Langfuse.
Hosting on managed platforms like Fly vs. self-hosting
Not all of these costs are borne by the expense of the infra itself—Big Cloud ☁️ will offer significant startup credits—but the cost of the engineering involved to maintain it, the cost of rolling out CI or CD in-house, and the cost of running support on a system with heavy third party dependencies and poor observability.
This isn’t just a case-study in how to cut costs, it’s how to run infra without breaking the bank or defaulting to infra-creep on the cloud because you get it for free for a while.