Software Engineering
Is AI the End of DevOps as We Know It?
- No, but it might help around the edges. DevOps is part of continuous delivery (of value to users).
- The classic stumbling block was developers and ops folks being incentivized by different things: change and stability.
- DevOps = people with different expertise to deliver software; pulling in the same direction
- CALMS framework (culture [no], automation [no, needs to be predictable], lean [not really], measurement [unlikely/maybe], sharing [unsure])
- Cultural and sharing concerns — we talk to AI instead of our coworkers
- DevOps needs determinism, which is not what AIs are built for.
- AI is not going to move quickly enough to eliminate human in the loop (HITL), but you can use it to ask it about trends for troubleshooting. The complexity is so contextual, so off-the-shelf likely won’t be enough.
The Software Supply Chain Problem No One Talks About
- Reliability = can you trust the system
- Robustness = can you rely on the system
- Resilience = can your system respond to perturbations
- Supply chain = parts of the value chain you don’t control
- See Wardley map. Scenario modeling involves moving nodes in the Wardley map right or left.
- Event storming = mapping out the domain of a problem in a value chain
- Performance testing = load testing (expected usage), soak testing (run the system for many days), stress testing (what happens when a system fails)
- Chaos testing = randomly turning things off in production to see what happens; this is run-time observability