Todoist Ramble: AI Builds Tasks As You Speak, Bypassing Text Transcription

Todoist launches Ramble: ramble into your phone, and AI instantly breaks it into a task list — voice interaction finally bypasses the detour of "transcribe to text first, then understand."

What this is

Doist (Todoist's parent company, founded in 2007) has introduced Ramble, inspired by the rapid-fire task-delegating boss in "The Devil Wears Prada." Users speak directly, and the AI creates, edits, and deletes tasks as it listens. Under the hood, it uses the Google Gemini Flash model's Live API; audio goes straight to the model without pre-transcription. Language detection, speech recognition, and semantic understanding are completed in one pass. If you switch apps and return, you resume the chat using a resumption token.

Industry view

We see a clear trend: AI is shifting from "Q&A mode" to "streaming execution." It no longer waits for you to organize your words but acts while listening. Google's Agent Platform (enterprise-grade AI Agent development platform) has demonstrated capabilities in real-time audio streaming and proactive tool calling. But the risks are clear: Doist admits that traditional assertive testing completely fails for this non-deterministic output, forcing a switch to semantic verification — as AI autonomously decides when to add or delete tasks, the probability of errors and the difficulty of troubleshooting are both high. Additionally, Doist has left interfaces open for other models, indicating their own concerns about relying solely on Google as a single supplier.

Impact on regular people

For enterprise IT: The task entry point shifts from keyboard to voice; mobile scenarios (store walks, inventory checks) will benefit first.

For individual professionals: Voice memos now have a structured output, but there remains a gap between "speaking clearly" and "AI understanding correctly." In the short term, it is better suited as a drafting tool.

For the consumer market: Todoist has a globally leading user base; if this feature proves stable, it will accelerate the popularization of "voice-first" in productivity tools — provided the AI doesn't take unsolicited actions.