On this episode, he explains how VAPI aims to create a voice-default future where we talk to all our computers—and goes deep into every step of VAPI’s voice pipelines and the technical challenges along the way.

Some highlights:

☎️ Massive scale: VAPI has processed 44 million voice calls on their platform, growing from a COVID-era one-click Zoom meeting button to a full voice infrastructure company used by thousands of developers.

⚡️ Latency matters: Voice AI needs to respond within 500 milliseconds to feel natural to humans. That means cleaning audio, detecting when users are done speaking, transcribing text, generating responses, converting to speech, and handling interruptions—all within a fraction of a second.

🗣️Voice-first future: Nikhil is betting a future where voice becomes our default interface with all computing systems.

If you’ve ever wondered how voice API actually works—this is the episode for you.

Chapters

00:00 - Introducing VAPI

04:42 - Pivoting through COVID

05:42 - ChatGPT existential crisis

08:33 - Technical challenges of voice

12:42 - Anatomy of a voice call

14:46 - Knowing when someone is done speaking

18:37 - Routing to the fastest model

22:07 - Knowledge and context injection

26:47 - The text-to-speech bottleneck

31:14 - Handling interruptions gracefully

33:43 - The 500-millisecond barrier

36:56 - The DNS latency discovery

39:25 - Scaling the team and what's next

Links

• VAPI

• Nikhil Gupta on LinkedIn

• VAPI on Product Hunt

• Stratechery by Ben Thompson - Recommended reading from Nikhil

• 💬 How Wayfaster started: “Wayfaster started out as a way to interview your entire pipeline in a couple seconds... the only way you can really do that is using voice AI."

• ⛔️Why AI video avatars are a no-go: "We tried avatars and every candidate hated it... I think it just makes candidates feel dumb. It's just like, why am I talking to an avatar? Like how stupid do you think I am?"

• 📑How AI interviews give candidates a chance: "Candidates have just gotten really accustomed to getting this default auto-reject email... At least here, I know that if I do a good job on the interview, I have a shot at this opportunity."

• 🐻Why he’s bearish consumer voice AI: "I'm actually a little more bearish on the consumer voice AI use cases. I like voice AI in constrained environments, like B2B, there's an amount of intent... I don't think people are going to want to be talking to their phones all the time."

Hope you enjoy the conversation! As always you can subscribe at https://smartspeakers.fm.

Chapters

0:00 Welcome and intro

1:32 First big AI aha moment

5:08 Career background

7:40 Exploring startup ideas

10:34 How WayFaster was born

14:21 Initial target markets

16:45 Candidate response to AI interviews

19:07 Benefits for recruiters and candidates

23:30 How AI is transforming recruiting

26:51 The two-sided recruiting game

31:05 Why Varun is bearish on consumer voice AI

35:09 Future generations and AI adoption

38:09 Content recommendations and closing

Links

• WayFaster Website

• Varun Khurana on LinkedIn

• South Park Commons

• Deepgram

• Charles Rubenfeld's Newsletter

Kwin Kramer from Daily remembers that time, too—and says today's voice AI moment feels just like those early internet days. That sense of endless possibilities is back.

His open-source project PipeCat has become the standard toolkit for voice agents. What began as an experiment now powers voice AI for OpenAI, Google DeepMind, and countless startups, making conversations feel natural and responsive.

Some highlights:

• That early internet feeling is back: "1995 to 1999 felt a certain way. It never felt that way again until 2023 to 2025."

• GPT-4 transformed Daily's business by removing a key bottleneck: "Previously you needed two humans for a conversation. Now you only need one, maybe not even that."

• Voice AI's killer feature? Latency matters: "If response times are long, you're in that uncanny valley where people get uncomfortable."

• Kwin's bold prediction: "We're all going to have friends in our group chats that aren't human because LLMs are actually really entertaining."

Hope you enjoy it as much as we did.

Links

Daily: https://daily.co/
PipeCat: https://pipecat.ai/
Kwin on Twitter: https://twitter.com/kwindla

Chapters

0:00 Intro

2:02 First AI aha moment

5:43 MIT Media Lab beginnings

9:05 BBS and door games

15:26 The AllAfrica journey

18:54 Starting Daily

21:13 COVID's impact on WebRTC

22:36 GPT-4 transformation

31:26 Building voice for LLMs

35:17 PipeCat's key challenges

44:10 The future of speech-to-speech

47:10 Voice AI adoption trends

52:34 Vibe coding revolution

56:11 What's next for PipeCat

Tom has deep experience in the voice world—we loved this conversation and we’re sure you will too!

Links

https://canonical.chat
https://x.com/Tom_Shapland
https://www.linkedin.com/in/tom-shapland-b4494212/

Chapters

0:00 - Intro

1:05 - Welcome and first AI moment

3:03 - Computer vision for thirsty plants

5:50 - Explaining AI to farmers

7:52 - Hardware is hard

8:31 - Non-VC shaped business struggles

16:51 - Starting a new company

19:41 - From metrics to conversation stages

22:42 - Voice AI evolution

26:40 - Balancing determinism and freedom

29:53 - Who uses Canonical and when

32:39 - Don't make your agent spell anything

35:52 - Zero to 80% is easy, production is hard

39:28 - Future of voice AI adoption

42:21 - Book recommendations and closing

Olivia Moore is the author of a16z’s market report on voice AI agents, which is a must-read if you’re interested in the space.

She’s an amazing thinker and we cover a lot in this conversation, including:

• Olivia's thesis that Voice AI is in it’s “internet in 1999” moment

• The "wedge" strategy: using Voice AI to get in the door before transforming broader business operations

• How recruiting became a surprising early adopter of Voice AI (candidates actually prefer it)

• Voice AI's unique ability to handle compliance and stay on-script in regulated industries

• Shifting from per-minute pricing to transaction-based fees or success metrics

• Why big tech platforms (OpenAI, Google) won't dominate vertical-specific voice applications

Links

• Olivia on X: https://x.com/omooretweets (must-follow!)

• Her market report on voice AI agents

• Dr Sbaitso emulator - this was packaged with new Sound Blaster sound cards! Strange stuff.

Chapters
0:00 Intro and welcome
1:24 Olivia's first AI "aha" moments
3:42 Coolest non-voice AI tools
6:29 How Olivia got into the Voice AI space
10:03 From consumer to enterprise
13:50 DMV-bot
16:22 Voice AI as a wedge
20:57 Traction in the voice market
24:24 Voice AI for recruiting
28:59 Why AI doesn't go off-script
31:03 Better experiences for candidates and customers
33:24 The future of call centers
37:02 Bear cases for Voice AI
41:25 Open problems
44:29 Pricing
47:12 Working with twin sister Justine
50:11 Three things Olivia's excited about

Subscribe and listen:
https://smartspeakers.fm

Phil shares how Standard Practice evolved from a healthcare payment startup to building an AI assistant that handles complex insurance calls. Along the way, we learn about Phil's life in Japan, his Oreo obsession, and why he believes natural conversation is as hard as self-driving cars.

I loved Phil’s concept of “success, but doof” — it neatly encapsulates just how hard it is to measure the success of AI voice calls.

Stay tuned next week for our chat with Olivia Moore from a16z.

See you soon—
dave & harish

Follow Phil:
https://www.linkedin.com/in/phil-markunas
https://philm.io/
https://x.com/philmarkunas

Standard Practice:
https://standardpractice.ai

Chapters
0:00 - Introduction to Phil Markunas, CTO of Standard Practice
1:34 - Emotional responses to voice AI
5:17 - How Army service shaped Phil's leadership
8:07 - Living in Japan
13:04 - The Nibble Health origin story and creating a medical bill payment card
17:59 - Facing challenges and developing the SimpleBill medical analysis tool
22:07 - Pivoting to voice AI
26:25 - Why voice conversation is an AGI-level problem
33:21 - Beyond "just prompt it up" to building sophisticated voice architecture
40:45 - "Success, but doof" moments; and the future of Standard Practice

Subscribe — and let us know what you think!
https://smartspeakers.fm

We talk about:

• Our journey since selling Chartable to Spotify in 2022: life inside Spotify, and Harish’s explorations outside corporate life in construction and factory tech

• Starting to work together again last July

• Exploring projects: a Chartable-like analytics platform for audiobooks, a text-to-speech bookmarking tool called Earmark, and finally experiments with voice AI

• We interview ChatGPT, which is surprisingly good at podcast ad reads!

• Harish thinks ChatGPT lied to us at least twice. What do you think?

Stay tuned next week: Phil Markunas from Standard Practice discusses voice AI in healthcare billing.

Thanks for listening, reading, and watching. We’d love to hear your feedback! Subscribe and comment at https://smartspeakers.fm

You can subscribe at smartspeakers.fm, on YouTube, or in your fave podcast app.

See you soon!

—Dave and Harish