- Codifying infrastructure with tools such as Packer, Terraform, Ansible. The infrastructure should be just as auditable, repeatable, and transparent as the code.
- Empowering and allowing the engineers who write code to deploy it.
- Immutable servers/containers.
- A solid suite of monitoring and alerting tools that auto-heal (preferred) or at least notify when things go wrong.
#1 is the technological combination of Dev and Ops
#3 is a corollary to #2
#4 is just good Ops
Devops is when those things cross over into normal developer work. The developers take on these "ops" duties themselves. It may or may not involve actual development. It always involves devs having to understand the infrastructure their code runs on.
Done right, it's a good thing. It makes devs sensitive to how their code is deployed and run in production, without being able to blame some other team for the application not performing well, not having adequate availability, etc.
Usually, it's not done right.
Devops is the act of codifying your operations. Automating deployments, ci/CD things like that. Putting alert and monitoring systems in place.
That's my understanding.