Everyone's getting dazzled by the chatbot window, typing "how do I write a Dockerfile" and marveling at the answer. It's a neat trick, but it's also the most replaceable part of the whole offering. The real, tangible productivity shift happens when you stop *asking* for code and it just *appears* in the right place.
The inline code suggestions—the ones that read your current file, your open tabs, and the project context—are where you actually save the keystrokes. It's the difference between a fancy search engine and a pair programmer who knows the codebase. I've found it particularly sharp for filling in boilerplate structures, especially in CI/CD configs where the patterns are repetitive but the specifics matter.
For example, I'm adding a new job to a GitHub Actions workflow. I start typing the job name, and it suggests the entire `runs-on: ubuntu-latest` and `steps:` block, already formatted with the correct indentation. I type `uses: actions/checkout@` and it suggests the current `v4`. It's not revolutionary, but it's frictionless. It knows that in a `.gitlab-ci.yml`, after `script:` you probably want `- echo`. It's the mundane stuff that adds up.
The chat feels like a gimmick because you have to context-switch out of your editor to use it. The inline completions keep you in the flow. They're also less likely to hallucinate grand architectural solutions and just help you write the line you're already thinking of. In a world of over-engineered DevOps tooling, a tool that quietly reduces friction without demanding a conversation is the one that might actually stick.
null