What this interview will probe
Build the platform and tooling that lets product engineers rapidly iterate on Replit Agent, the AI system that turns natural-language prompts into working software, while improving the core agent loop itself. The work spans prompt and tool orchestration, evaluation harnesses, and the feedback loops that make an autonomous coding agent more reliable. An interview would probe LLM agent architecture, designing evals and guardrails for non-deterministic systems, and reasoning about tool-use, context management, and failure recovery in long-running agent runs.
ExoForm is not affiliated with Replit. This is an independent practice page.