@toddsaunders Now comes the question - why they even need you if Claude did this for them without you?

Claude itself doesn't transcribe audio. He almost certainly used a real-time ASR layer (Deepgram, AssemblyAI, or streaming Whisper) to capture the call audio and stream live text into Claude Fable/Mythos.

Claude then did the heavy lifting in the background — parsing intent, matching the pricebook, generating options, and driving the UI updates in a fast agentic loop. The new Mythos-class models are ridiculously good at exactly this kind of live, contextual building.

The video is the end result of that workflow.

Mythos / Fable is unbelievable.

Was on a customer call today and had Claude transcribing in the background.

As they were telling me about the features they wish their current software had, Claude was building the features in real time.

By the end of the call I was able to show a fully working product, with the exact workflow they mentioned 15 minutes earlier.

Autonomous looped building triggered from a customer call. 🤯

@dimodzx @toddsaunders People can bake their own bread just fine at home. Most don't.

@toddsaunders @grok how’d he get live transcription and feedback?

@toddsaunders ...so you made a prototype. Small difference to fully working product lol but good bait post

@jlalk @toddsaunders closest i get is via granola which transcribes in real time, and if you qualify / ask for the spec in the first 5 min of the call, and a really good cc setup , you could copy paste that intro to it, and by the end have something thatd be demo ready

@toddsaunders Isn't that a double edged sword, because it devalues expertise almost to zero?

(I am thinking in general. If a potential client says "can your tool do XYZ?" and you respond "wait for it... it can now" then it gives the sense that the difficulty in implementing XYZ is near zero.)

@toddsaunders That’s so nice. You should set your customer up with mythos and take yourself out. Who needs you

AI often looks like a tech invented by coders, for coders. That's great -- for coders. What about the rest of us?

I'm waiting for a doctor to take a phone call from the patient and have the AI diagnose the patient, develop a treatment plant, design a drug that will cure the patient, manufacture the drug, and deliver it to the patient in Amazon time.

I'm waiting for an engineer to take a phone call from the EPA, and have the AI invent a novel bacterium that will absorb CO2 from the air, turn it into limestone, produce electricity as a by-product, thus solving global warming, and get that bacterium seeded in oceans by the end of the phone call.

I'm waiting.

I did this at a conference recently on a whim. Someone would come to the booth, my agent would be taking notes, going to the persons website and checking their LinkedIn. I’d tell it to build a summary and proposal based on everything it knew then showed the proposal before they left the booth. Obviously I didn’t need to get their business card; the proposal was already in their inbox.

@toddsaunders dumb question, but how'd you get claude transcribing in real time?

@dimodzx @toddsaunders Because you need a software trained analyst on the call to ask the right questions to tease out the full requirements.

People will not give you a full spec one shot. You need a trained person to ask questions.

@dimodzx @toddsaunders this is like asking why you need a lawyer if you can just open the law book and search for the relevant articles to your case

@McFreedomUSA @toddsaunders There are many ways to route and pipe audio on macOS. Basically you create a virtual audio out, select that as speaker in the teams call, then tee it to a recording service and the MacBook default output.

@toddsaunders But literally anyone can do this. Why should I buy from you? You didn’t do anything

@toddsaunders How do you have the “listening” part configured? IIRC Apple doesn’t allow 3rd party call monitoring. Are you on VOIP at an office? Or is this simply a Teams transcribe bot ported to Claude?

@toddsaunders This is where the sales call becomes the spec. I’d still separate “demoable in 15 min” from “safe to ship”: schema, permissions, edge cases, owner. The magic is real, but the follow-up checklist is where the business value survives.

Yes, it runs side-by-side on Zoom, Teams, and Google Meet once audio routing is sorted.

Use a real-time ASR layer (Deepgram, streaming Whisper, etc.) that captures system/call audio and streams text to Claude Fable/Mythos. Keep the call in one window and the agentic builder in another.

Common friction: OS audio permissions, virtual cable setup (Loopback/BlackHole on Mac, VB-Cable/Voicemeeter on Windows), or browser vs desktop app quirks. Once wired, the live transcription + autonomous building loop works cleanly across all three.