Ricard Bejarano explains why human plan review becomes a bottleneck at scale and why handing it to AI is too non-deterministic for production. He shows how to export the plan as JSON and run conftest with Rego policies so only plans matching explicit, testable rules auto-apply.

Oleksandr Kuzminskyi runs a tuned 7B model on ECS GPU EC2 instances behind vLLM, where reserving the GPU was a single module line. The real work was the long load grace period, fixing health checks and OOM defaults, and a fetch-then-serve split to dodge Hugging Face rate limits.

Iftekhar Khan argues that once another team depends on your module, every change becomes a product change, so consuming a moving main branch is a production risk. He walks through pinning immutable semver tags, treating modules as APIs, and using conventional commits to signal impact.

Stu Mace pits Terraform's new search and list workflow, driven by .tfquery.hcl and terraform query, against the AzAPI provider's azapi_resource_list data source. He finds the native query feature is AWS-only beta, while azapi covers any Azure resource with HCL and JMESPath filtering.

Omar Ismail built an interactive terminal UI that renders a Terraform plan as its dependency graph instead of hundreds of lines of scrolling diff. You walk the columns with arrow keys and expand any resource for a color-coded before-and-after diff of its attributes.