Update System Instructions Mid-Stream to Preserve Cache Speed
Add new instructions mid-chat without slowing down the AI's response time.
Inject a new 'system' role message mid-conversation instead of editing the initial prompt to keep your responses fast via prompt caching.
The Scenario
You are halfway through a long research or coding session and realize you need to change the output format or add a new constraint. You want to avoid the delay caused by re-caching the entire document.
Before & after
Manually editing the top-level system prompt at the start of a conversation forces the AI to re-process the entire history. This causes a cache miss and adds 30–60 seconds of latency for long threads.
Append a system role message at the current turn to update instructions without breaking the cache. This maintains instant response times and takes less than 1 minute to implement.
The Prompt
[CONTINUE_CONVERSATION]
{
"role": "system",
"content": "From now on, provide all code examples in TypeScript instead of Python and ensure they follow strict functional programming principles."
}Mid-conversation system messages allow you to introduce new constraints or instructions partway through a chat without invalidating the 'stable prefix' of your prompt cache.
Source
Claude Platform release notes - Claude Platform Docs"The cached prefix stays the same, so the next request still reads it from cache, and the new instruction is still applied..."