AI
Can We Trust AI to Code Without Human Oversight?
- AI can create more code faster. There are more PRs and they are bigger. More WIP and context switching.
- See: Faros AI Engineering Report 2026
- Code review goals
- Correctness (does what it’s supposed to do)
- Shared learning (feedback from peers, learning what your peers and the system are doing)
- Alignment with organizational strategy and practices (standards, consistency, architecturally aligned with other systems this may interact with)
- Awareness of the codebase as a whole
- AI doesn’t reduce work; it intensifies it. (citation)
- To deal with the rate mismatch, companies are using AI to review AI’s code.
- Why are humans reviewing the code if AI is writing it, reviewing it, and modifying it?
- Chainguard says it’s no longer doing human reviews of code. (citation) This is akin to “dark factories” in automation.
- Your choice to move to this model depends on whether you have enough trust that your system will create code that’s production ready.
This video involves a seasoned developer explaining his experiences about using AI to write code, the challenges he’s faced, and his decision to stop using AI tools.
Software Engineering
Best practices for reliable Azure Functions
I’m doing client work that involves Azure Functions, so this Microsoft Learn doc provided a good overview of considerations such as hosting, storage, deployments, concurrency, security, availability, and monitoring.
Improve the performance and reliability of Azure Functions
Some greenfield client work brought up the question of whether to have multiple Azure Functions in a single repo versus deploying them separately. This Microsoft Learn doc not only discusses how to organize functions, but other best practices such as avoiding long-running functions, handling cross-function communication, writing stateless functions, and considering scalability options.