Building voice AI? I work in DevRel at @Speechmatics. I share runnable demos, integration notes and tests with real-world audio from voice agents to talking games. Start here: github.com/speechmatics/spee… What are you building, and where does speech fail?

Building Voice AI for healthcare? Here is a challenge worth thinking about. What happens when a doctor mentions a complex medication, a patient switches languages mid-sentence, and multiple people are speaking? Your STT needs to keep up. I have been looking at our new Oak model at @Speechmatics , and the benchmarks are interesting: 91% medical term recall, ranking #1 across all 11 languages tested against 8 other providers. If you are curious to experiment, we have released an Oak Consultation Studio example on GitHub: github.com/speechmatics/spee… Would love to hear from developers building medical scribes or healthcare agents. What's been your biggest challenge with speech recognition? #VoiceAI #HealthcareAI #DevRel

12k for 5090 6k for standard... Ok this is slightly out of control now... en.gamegpu.com/news/zhelezo/…

After a week with @livekit @fin_ai , @aquavoice ,@livekit, @cartesia , @AgoraIO, @Vapi_AI and many other builders across SF, one thing is clear: Voice AI has moved past can it talk? The real question is whether it understands well enough to act. Try @Speechmatics. Stress it. Break it. Show us what you build. Here is my summary below 👇 #SFTechWeek #VoiceAI

The more powerful voice agents become, the less room there is for close enough. Great session tonight at #SFTechWeek with @aquavoice , @livekit and @Speechmatics . The demos were fun, but the bigger point was serious: If a voice agent can change your calendar, update systems, or cancel a flight, one bad transcription can have real consequences. That is why accuracy, context, turn-taking and diarization matter just as much as latency. Voice is becoming an interface. Now it has to become one we can trust.

If voice becomes the default customer interface, speech-to-text stops being a feature. It becomes infrastructure. Great session from @rati_zv and @gavinjoyce at @fin_ai today. The more capable voice agents become at taking real actions, the more important the first layer gets: Did the agent actually understand the customer correctly? That’s exactly the problem we’re focused on with Agent STT at @Speechmatics . Accuracy, turn-taking, interruptions and latency suddenly matter a lot more when the AI can actually do something. #VoiceAI #SFTechWeek #Speechmatics

Gaming is one of the best stress tests for Voice AI. Players interrupt. Go off script. Say weird things. Change direction constantly. And your agent still has to keep up. At the Conversational AI x Gaming Hackathon at #SFTechWeek today with @playmanogames and Cartorga. Really interesting seeing how both teams are pushing voice and AI into gaming, while from the @Speechmatics side we get to think about the part underneath it all: Can the system actually understand the player? This is where Voice AI gets fun. TIME TO HACK

Voice AI is escaping the demo. Stopped by Coffee @cartesia at #SFTechWeek today. Humans still make the coffee. But the voice tech is the interesting part. From @Speechmatics side I think. Great voice AI isn’t just natural TTS. It has to hear accurately, handle noise, interruptions, accents and know when you have actually finished speaking. That whole loop has to disappear. That’s when voice becomes an interface. In SF building Voice AI this week? Come find me. I want to see what you are building and, even better, what still breaks.

If you are building Voice AI in SF this week, I want to see it. Not just the polished demo. Show me the edge cases. The interruptions. The weird names. The latency problem. The thing that breaks 1 time in 10. I am here for #SFTechWeek with @Speechmatics and made a quick video on why we built Agent STT specifically for voice agents. If you are building with @livekit , @pipecat_ai , or your own stack, DM me. Let’s grab a coffee and compare notes. #VoiceAI

Your voice agent responds instantly. Cool. But did it actually understand the person? I attended Agora’s From Prototype to Production: Scaling Voice AI at #SFTechWeek. The panel got into what happens AFTER the demo. Speed isn’t everything. Customer service and coding agents have different expectations. Acknowledge quickly, but take the time to do the work properly. More context isn’t always better. Give the model what it needs, when it needs it. Keep conversations responsive while heavier work happens in the backend. Stop asking one model to do every job. Route simple tasks to faster models. Bring in deeper reasoning when it matters. Test beyond the happy path. Right tool? Right information? Correct permissions? Trace failures and evaluate audio, not just transcripts. Your tests need to reflect your business. The panel discussed smart glasses combining voice, gaze and spatial context. “Remember that door design” hits differently when the agent knows which door you’re looking at. My takeaway, working at @Speechmatics getting to the wrong answer faster is still getting it wrong. Huge thanks to @AgoraIO for an awesome event! Next, will be at Speechmatics × Aqua Voice × LiveKit for Solving Voice as an Interface. partiful.com/e/Qfb44oJOo4cYv… #VoiceAI #AIAgents

Your voice agent works perfectly. Until a real human uses it. Last night at the @Speechmatics x Tuner Voice AI (usetuner.ai/) Meetup, we talked about what happens AFTER the demo. Background noise. Multiple speakers. Alphanumerics. Unexpected workflows. Real production calls. The big takeaway? Your #STT can work. Your #LLM can work. Your tool call can work. And the user can still fail to achieve what they wanted. Great panel, great demos, and lots to think about. Voice AI gets really interesting when it leaves the demo room. #VoiceAI

"Sorry, I didn't catch that." Nobody puts that in the demo. Your voice agent works now put 1,000 real users in front of it. @Speechmatics x Tuner are hosting a London meetup on where voice AI actually breaks in production (and how to fix it). 📅 Thu Oct 1, 6–8:30pm 📍 London 🔗 luma.com/pyhvutqe #AI #ConversationalAI #TechEvents

David Heinemeier Hansson, the creator of #Ruby on Rails, said something in his 2026 keynote that I haven't stopped thinking about. He is retired from being a professional programmer. Not because he burned out or because AI pushed him out reluctantly but because he's chosen to see it with joy instead of regret. His argument: writing code by hand, chiseling it out line by line, was a wonderful and productive era. A quarter century of flow states, learning, satisfaction. But that era is ending. Today, hand-written code is no longer the economically productive way to build software for most programmers at most companies. By the end of this year, he says, it'll be true almost everywhere. His framing for what comes next is the part that struck me most: this isn't a loss, it's a new career. Not chiseling code, but making things. Not typing every instruction, but steering intelligence that a few years ago only existed in science fiction. A privilege, he says, to have lived on both sides of that chasm to have done it by hand, and to be there at the exact moment the switch happened. I agree with almost all of it. And I think there's a next step he didn't quite name. If typing code is no longer the bottleneck, then typing itself becomes the bottleneck. Keyboards were never the most natural interface for human intent they were just the only precise one we had. Voice is faster to think in, faster to correct, faster to refine. We explain things out loud better than we type them. That's the shift we're building for at @Speechmatics . As intelligence becomes the thing you steer rather than the thing you hand-code, voice becomes the input layer for steering it. For developers directing AI systems. For teams working hands-free. For anyone who'd rather describe what they want than encode it character by character. DHH lived through the golden age of hand-written code and is walking into what's next with excitement, not nostalgia. That's the right posture. The next chapter won't be won by whoever types fastest. It will be won by whoever can speak their intent most clearly to the systems doing the building. That is the future we are building toward. #voiceAI