There Is No Repro for a Phone Call
Every other part of my stack comes with a safety net I never asked for. TypeScript refuses to compile. A test goes red. A pull request makes somebody ask why I touched that line. Then there is the prompt. A client sends a note saying the agent came across as a bit pushy on that last call. I open the prompt, soften one sentence, hit publish. No build. No diff. No test suite. The next caller gets…
There is no safety net for changes made to prompts in a phone call system. Unlike other parts of the stack, such as TypeScript or tests, prompts do not have compile checks, red alerts, or pull request reviews. When a change is made to a prompt, it is immediately deployed across all branches of the conversation, including scenarios the developer may not have considered. This global state is written in prose, with no scoping or privacy.
This lack of control leads to unexpected regressions. For example, a change aimed at making an agent more forthcoming about pricing may cause the agent to answer unrelated questions or become slower and more repetitive. These issues can go unnoticed until they surface as callers hanging up mid-booking. The lack of deterministic replay means a previous call cannot be replayed against the new prompt to verify the change. The output is a conversation, not a value, making it difficult to assert exact wording.
Additionally, production config is edited live in a browser, with no review step or change history. The platform updates the model version, voice, and recognition behavior, potentially changing the agent without explicit notification. Therefore, a small regression set of real calls is essential for identifying potential issues. These scenarios should include high-volume intents, refusals, awkward situations, and out-of-scope questions.
By asserting outcomes rather than phrasing, and automating the structure while listening to the human aspect, regressions can be caught more effectively. Finally, keeping the change reversible through version control ensures a clear audit trail and allows for quick rollbacks if necessary.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
