Published September 6, 2026 · facts checked on that date
Talking to your AI on Windows: what voice dictation doesn’t do
Voice dictation is a solved problem. One shortcut, you talk, the text lands at the cursor, and it lands well. What is not solved is what happens after the text, when the one you are talking to is an AI that works, and answers on its own schedule.
All these apps stop at the text
That is not a criticism. It is their promise, and they keep it. Superwhisper describes a dictation that works in any app on Windows 10 and 11, with automatic pasting of the transcribed text. Spokenly sums up its own in one sentence: a global shortcut, and speech becomes text at the cursor. Whispering, a free open-source app, puts it even shorter: press, talk, get text.
The trip ends where yours begins. You dictated your question, and the AI answers with forty lines. Then another task finishes a job you started ten minutes ago, and a third one is waiting for your approval. None of that comes through the keyboard, and none of it is a transcription problem.
The industry has started to notice, and each player handles it differently
Superwhisper explicitly targets developers: its Windows page highlights talking to Cursor, VS Code and Claude Code instead of typing. Voice stays the input, but the target is clear.
Spokenly goes a step further and even reverses the initiative: its local server lets Claude Code, Codex or Cursor ask you a question, which you answer by speaking. One option reads those questions aloud. Three limits, written in their own documentation: these are the agent’s questions, not its answers, the option is off by default, and it is documented for macOS only. Their Windows page mentions no integration of this kind.
Whispering chooses to remain a building block: free transcription, the model of your choice, plain files on your disk. The project joined a larger ecosystem and its original repository has been archived since February 2026. Its latest published release dates from December 27, 2025.
The three questions text doesn’t settle
- What did it answer? Reading forty lines on screen means stopping, scrolling back, and picking the thread up again. Listening only means listening.
- Where are the others? When three tasks are running, the useful information is not their content. It is which one has finished.
- How do I get back to it? Finding the right window, the right conversation, the right project: that is where the time you saved dictating gets lost again.
What SpeakApp does with that trip
With SpeakApp, your AI’s replies are read to you and you answer out loud, straight into the right task. It queues the replies from your AI tasks, reads them and keeps their source visible. You answer without going back to the keyboard, and you get back to the right task. Dictation writes at your cursor, in 40 languages, and you can dictate in one language so that SpeakApp writes in another. It is available on Windows, coming soon to Mac.
The exact path your data takes, what stays on the PC, what leaves it, and who receives it, is documented on the security page.
Sources and method
Every fact about a third-party app comes from an official page, opened and checked on September 6, 2026. These products change fast, so this article carries its date to tell you what it is worth. “Not documented” means the capability does not appear in the official offer we consulted, not that it does not exist. This article does not rank raw transcription accuracy: we have no shared measurement solid enough to claim that one engine transcribes better than another.
Available on Windows, coming soon to Mac.