Cutting cloud cost without slowing delivery

Cloud bills creep up quietly. The fix is rarely one big switch — it’s a few boring habits done consistently.

What worked

1. Rightsizing first — compare actual CPU/memory to instance size before buying Reserved Instances.
2. Kill idle — old load balancers, unused EBS volumes, forgotten NAT gateways.
3. Visibility — budgets + anomaly alerts so finance and engineering see the same numbers.

What to avoid

Don’t lock into long commitments before you understand usage. Optimize the noisy 20% of spend first.

Takeaway

Cost work is platform work. Treat it like reliability: measure, alert, automate, review.