There is nothing especially new about talking into a machine and expecting typed words to come out the other side. We have had dictaphones, transcription workflows, and voice-driven document creation for years. What has changed is the quality of the tools around that basic idea.
What interests me is not speech-to-text on its own, but what happens when you combine it with a large language model, or LLM. An LLM is an AI system that can work with language: it can summarise, reorganise, rewrite, and help turn rough thinking into a more coherent draft. That combination creates a practical workflow for people who think better by speaking than by typing.
The basic idea
In my own case, I have been experimenting with this inside apLabs, which is my personal productivity AI toolkit. The use case is straightforward: I record an informal voice note, upload the audio file into apLabs as a document, and let the system recognise that the attachment is audio rather than a PDF or a text file.
From there, the workflow is handled in the background. The system sends the audio through a speech-to-text API call to an OpenAI or Gemini TTS conversion tool, stores the resulting text, and makes it available in the prompt I am working with. In simple terms, my spoken notes become source material for the AI to work on.
The important point is that this was not built as a completely separate feature. I already had a document-processing pattern in place for other file types. PDF files, for example, were already being converted into Markdown in the background. So rather than inventing a whole new architecture, I followed the same pattern for audio.
That matters because useful AI systems are often less about clever demos and more about disciplined reuse of sound design.
Why this helps
The real value is not that the machine transcribes my words perfectly. I dont have to “train” a dictaphone software tool to understand my accent. The value is that I can speak naturally, capture rough ideas quickly, and then ask the LLM to impose some structure.
A rough spoken draft is often full of repetition, unfinished thoughts, and side tracks. That is normal. Human speech is not usually organised like an article. But an LLM is quite good at taking that material and reshaping it into a logical flow.
In practice, that means a blog post that might previously have taken several hours to wrestle into existence can move to a usable first draft in perhaps 15 to 20 minutes. That does not mean the article is finished. It still needs judgement, editing, and rewriting. But it is a much better starting point than a blank page.
For anyone who finds long-form typing slow, tiring, or simply unnatural, this is a very practical improvement.
The modern dictaphone
In one sense, this is just a modern version of an old idea. The dictaphone has not disappeared; it has evolved. Instead of handing the recording to a typist, you hand it to a system that can both transcribe and help shape the result.
Of course, it is not magic. The draft can still be weak. The structure may need work. The tone may need adjustment. Facts still need checking. But the workflow is useful because it helps convert loose verbal thinking into something editable and concrete.
I find that particularly valuable for early-stage writing, where the hardest part is often not polishing the words but getting the ideas out.
Practical lessons for technical leaders
If you are considering a similar workflow, a few points stand out:
- Start with a real use case, not the technology. In my case, it was drafting articles from spoken notes.
- Reuse existing architecture where possible. New AI features are easier to manage when they fit established document patterns.
- Treat AI output as a first draft, not a finished product.
- Optimise for friction reduction. The gain comes from making it easier to capture and shape thought.
For consultants, engineers, and technical leaders, that may be the real opportunity. AI does not replace thinking. But it can reduce the effort required to turn thinking into a usable artifact.
Closing thoughts
This article itself reflects that workflow. It began as spoken notes, not typed prose. That is why I think this use case matters. It is about creating a more natural path from idea to draft by removing friction. Eventually the keyboard might dissapear from our lives in favour of a far more natural and powerful interface.
If you do a lot of knowledge work and you often think better aloud than at a keyboard, it may be worth exploring what a speech-to-text plus LLM workflow could do for you.
Call to Action
If this way of working resonates with you, explore how voice capture, transcription, and LLM-assisted drafting could fit into your own writing or technical knowledge workflow.
