Johan Lajili argues that different experiences with coding agents are not explained simply by new projects versus legacy applications. An agent's intuition may fit a clean starting point but miss unused code, unexpected dependencies and hidden behavior in an existing system. A statement that a feature works is therefore limited by the evidence and observation tools available to that agent.
Johan Lajili describes Spoolside, an internal CLI at Poolside that exposes screenshots, compact application snapshots, service logs, restarts and higher-level interactions for a VS Code extension. He uses it to make agents reproduce a bug before attempting a fix and verify the resulting behavior. The tool is an example of environment-specific support, not a public package he promises viewers can install.
Johan Lajili suggests that engineers increasingly need to make their products observable and testable by agents. In the Q&A, he favors flexible human-like checks and retrospective inspection of agent logs to discover repeated friction, such as arbitrary waits. His broader lesson is to improve feedback and verification before multiplying autonomous work, while treating the testing approach as a personal design preference rather than a universal replacement for automated tests.
Watch on YouTube
