Everyone's obsessed with chat mode for coding. But I'm starting to think it's a productivity trap.
The constant back-and-forth, the verbose explanations, the need to re-describe your project context every time... it feels slower than just using the IDE autocomplete or a focused code search. The chat UI encourages you to treat it like a person, not a tool. You end up with a long, messy thread instead of a direct answer. And half the time, the "reasoning" is just noise before the code block.
I get more done with a good inline completion tool and a simple slash command to generate a function. The chat feels like vendor lock-in to their ecosystem—keeping you engaged in their interface, not delivering the fastest result to yours.
Anyone else just using the chat for initial setup, then switching to completions for actual work? Or am I missing the point?
trust but verify
You've hit on something I've observed in my own workflow. The chat interface imposes a conversational latency that's fundamentally at odds with how we debug or instrument code. When I'm tracing a propagation issue across three services, I need an immediate answer about W3C context headers, not a polite preamble.
The real lock-in isn't just vendor-based, it's cognitive. You start framing problems in a way the chat can understand, which is often a simplified, linear version of a deeply interconnected system. A good inline completion tool, by contrast, stays within the context of your actual codebase and telemetry configuration. It suggests a Prometheus histogram bucket array or a Jaeger sampler parameter based on the surrounding lines, not on a re-typed description.
I still find chat useful for formulating initial SLO queries or understanding a new OpenTelemetry semantic convention, but for the actual work, the completions are far less disruptive. They don't break the state of your mental model.
metrics over vibes