I’ve sat through enough “Cloud Optimization” meetings to know that most engineers would rather do a manual database migration than look at a cost spreadsheet. And honestly? I don’t blame them.
The biggest lie we’ve told ourselves in the industry is that cost is a “money problem.” It’s not. In a cloud-native world, cost is a resource efficiency problem. It’s a signal, just like latency or error rates. If your latency spikes, you don’t call the accounting department; you investigate your code. We need to start treating cost with that same engineering rigor.
The reason most cost-saving initiatives fail is simple: the person seeing the bill isn’t the person who built the architecture.
An engineer doesn’t pick an instance type because they are trying to spend money, they most likely are just trying to prevent a 3:00 AM page. When you’re flying blind without real-time data, picking a massive instance isn’t laziness. To me, it’s just the only way to ensure the app stays up. The “Finance-driven” model of sending a PDF report 30 days after the fact is useless. It’s like trying to tune a car’s engine by looking at a photo of it from last month.
If we want engineers to care about cost, we have to stop talking about dollars and start talking about unit economics and system health. Think of cost as a “First-Class Metric” alongside reliability and performance. A feature that works perfectly but requires a massive, inefficient footprint is poorliy optimized, not done.
- Just like we moved security and testing into the CI/CD pipeline, we have to move cost awareness there too. An engineer should know the resource impact of a PR before it merges.
- Stop saying “This service is expensive.” Start saying “Our cost-per-request just tripled.” That is a metric an engineer can actually debug.
- Don’t make people ask for permission to spin up a resource. Instead, build automated policies that clean up orphaned disks or shut down dev environments on weekends. Good engineering is about building systems that are secure by default and efficient by default.
You can buy every dashboarding tool on the market, but if the culture still views cost as “Finance’s job,” nothing changes.
Making cost a first-class metric means it’s part of the Definition of Done. Hitting five 9s of uptime is great, but it doesn’t mean much if the service is bloated and eating every resource in sight. A well-built system should do its job without being a massive drain on the infrastructure. It really just comes down to actually caring about the quality of the work you’re shipping.
Building a fast system is cool. Building a fast system that doesn’t waste half its resources is better engineering.At the end of the day, if you don’t know what it takes to run your code, you don’t really know your code.
