From Skills Back to Tools: Why Our Dashboard Assistant Moved Off the Claude Code (blog.promptlayer.com)

Wrangler AI, the in-app assistant, has transitioned from using the Claude Code SDK to a new architecture that employs simple prompts and explicit tools. This shift allows Wrangler AI to serve users more efficiently by streamlining how tasks, such as creating prompts and workflows, are executed directly within the dashboard. The Claude Code SDK, although powerful for autonomous coding tasks, led to significant latency and resource overhead, taking up to 3 minutes for simple queries, as shown in performance comparisons. The newly adopted tools architecture has demonstrated a marked improvement in speed, with responses now averaging between 24 to 39 seconds compared to several minutes previously. By reducing the number of LLM calls and eliminating unnecessary complexities, the new system enhances predictability, cost efficiency, and debuggability. Additionally, the explicit tool use minimizes errors and the potential security risks associated with autonomous script execution. Overall, this shift underscores the importance of tailored architectural designs in AI applications, as Wrangler AI focuses on interactive user assistance rather than autonomous background processes.