Akshay Ramaswamy

posting about voice agents, AI products, and automating myself away

staff PM @elise_ai (200M+ ARR) · 2x founder (acq @Coinbase) · @Stanford / @YCombinator

[Akshay Ramaswamy @TheRealAk914]

Today, I’m excited to announce that the Omni team will be joining @coinbase in their incredible mission of democratizing economic freedom! Grateful for the support of all of our friends, family, and investors.

https://t.co/H6IrOJVT4z

[Akshay Ramaswamy @TheRealAk914]

@mjdouglas what are some examples?

[Akshay Ramaswamy @TheRealAk914]

consumer AI is so back

[Noah Shinn @noahrshinn]

Your Instinct can now handle phone calls.

Introducing Instinct Concierge – a white glove service meant to handle high-touch cases, such as making phone calls, high-end service booking, and more. You’ll be able to book the restaurant that doesn’t take online reservations, get on your dentist’s cancellation list, or have that cable bill sorted out.

We’re slowly rolling this out to our early access group today and will be expanding access soon.

[Akshay Ramaswamy @TheRealAk914]

The race to S2S is picking up!

@OfficialLoganK any plans to support the same client delegation style API as gpt-live?

[Logan Kilpatrick @OfficialLoganK]

Say hello (literally) to Gemini 3.8 Live and 3.8 Live Extended Thinking, our new SOTA live audio models, available with frontier price + performance.

3.8 Live supports 97 languages (can seamlessly switch), async tool calls, and more!

[Akshay Ramaswamy @TheRealAk914]

@BoBowchan cooked on this one!!

[AI Engineer 🔜 Paris 🇫🇷 @aiDotEngineer]

Live now: our Voice Track from AI Engineer World's Fair 2026.

Half a second that made a product feel slow, a trillion phone calls a year, and the case for agents that know when not to speak.

Thesis: voice agents do not fail at talking. They fail at listening.

https://t.co/xx1VOTKCDQ

- The New Primitives: @kwindla, Daily
- Speech-to-Speech Model Research at Google DeepMind: @valeriawu_ & Tom Ouyang
- Voice Agents Can Just Do Things: @charlierguo, OpenAI
- Your Voice Agent is Just a Walkie Talkie: @neilzegh, Gradium
- Voice-First AI Companion: @paularambles, Tolan
- 5 Voice Agent Failure Modes You'll Hit in Week One: @bevenky, Plivo
- I Monitored Crime Audio. Voice Agents Scare Me More. @sumanyu, Hamming AI
- Realtime Voice Agents with Frontier Intelligence: @bobowchan, EliseAI
- "My name is... my name is...": Midam Kim, ServiceNow
- Act, Confirm, or Stop? Amit Desai, Roku

[Akshay Ramaswamy @TheRealAk914]

interesting

[Shobhit Banga @shobhitbanga]

Introducing Jarvis Bench v0.5, @voicearena_ai's conversational agent benchmark.

We've been obsessed with one question at VoiceArena: why do voice agent demos sound incredible, benchmarks say models are near-perfect, and yet you probably didn't have a single real conversation with a voice agent in the last 24 hours?

Here's what's different about the Jarvis Bench. Real humans have live conversations with voice agents. A second group of humans blind-votes pairwise on two questions: which sounded more human, which got the job done. And one of the "agents" on the leaderboard is a human.

[Akshay Ramaswamy @TheRealAk914]

@elise_ai SF soon 👀

[Akshay Ramaswamy @TheRealAk914]

the team at @baseten is cracked, always a pleasure getting to jam with them!

some of the improvements we're seeing from posttraining have had a meaningful impact on latency, cost, and performance with our voice agent.

also scales well with where the space is headed with S2S models - client delegation means we can use any model for intelligence behind the hood. so exciting to see the convergence of these things coming together!

[Elise Labs @elise_ai]

Top leaders at EliseAI and @baseten across product, engineering and research met in NYC to discuss why we're betting on open-source models.

Watch the full discussion ↓

00:00 Why we’re betting on open source
05:37 Fine-tuning for better latency, quality and cost
16:04 Building evals that reflect real production behavior
27:51 Choosing the right model for the task
36:20 What open source means for the future of agents like Apollo, the one agent for property management

Featuring: Aaron Ellis-Bloor, Christopher Wirick, Mario Martone, @TheRealAk914 and @ShrutiGoli.

[Akshay Ramaswamy @TheRealAk914]

@bygregorr @elise_ai @OpenAIDevs it’s actually quite good with interruption handling, full duplex feels directionally much better

[Akshay Ramaswamy @TheRealAk914]

Been hard to keep this one a secret!

For the past couple of months, @elise_ai has been working with @OpenAIDevs as an early design partner for GPT‑Live-1.

I still remember my excitement back in 2024 when the first S2S models came out, but it took a long time to get to production - now it feels like we’re finally there.

There are lots of really interesting improvements to using GPT-Live-1 – some less obvious than others:

[Akshay Ramaswamy @TheRealAk914]

@cryptopunk7213 how does it compare to grok bot?

[Eric Bernhold @ebernhol]

Apollo is here: the one AI teammate for multifamily property management

[Akshay Ramaswamy @TheRealAk914]

@reach_vb big fans of it at @elise_ai!!
https://t.co/VvQ2f9zbHQ

[Akshay Ramaswamy @TheRealAk914]

Been hard to keep this one a secret!

For the past couple of months, @elise_ai has been working with @OpenAIDevs as an early design partner for GPT‑Live-1.

I still remember my excitement back in 2024 when the first S2S models came out, but it took a long time to get to production - now it feels like we’re finally there.

There are lots of really interesting improvements to using GPT-Live-1 – some less obvious than others:

[Akshay Ramaswamy @TheRealAk914]

@danshipper has been a huge unlock for us at @elise_ai!!

think we're going to see an explosion of new voice use cases enabled by full-duplex, would love to hear a podcast on @every on your take once you play around with it 👀

https://t.co/VvQ2f9zbHQ

[Akshay Ramaswamy @TheRealAk914]

Been hard to keep this one a secret!

For the past couple of months, @elise_ai has been working with @OpenAIDevs as an early design partner for GPT‑Live-1.

I still remember my excitement back in 2024 when the first S2S models came out, but it took a long time to get to production - now it feels like we’re finally there.

There are lots of really interesting improvements to using GPT-Live-1 – some less obvious than others:

[Akshay Ramaswamy @TheRealAk914]

@thsottiaux Voice will be the interface for AGI. Feels obvious to us at @elise_ai, and full-duplex is a big piece in getting there!!

https://t.co/VvQ2f9zbHQ

[Akshay Ramaswamy @TheRealAk914]

Been hard to keep this one a secret!

For the past couple of months, @elise_ai has been working with @OpenAIDevs as an early design partner for GPT‑Live-1.

I still remember my excitement back in 2024 when the first S2S models came out, but it took a long time to get to production - now it feels like we’re finally there.

There are lots of really interesting improvements to using GPT-Live-1 – some less obvious than others:

[Akshay Ramaswamy @TheRealAk914]

@palashshah @elise_ai @OpenAIDevs Likewise man!! Excited to keep working together on new stuff 🚀

[Akshay Ramaswamy @TheRealAk914]

@daidaijiayu been a blast working with you on this!!! voice team at OpenAI is cracked 🤝

[jiayu @daidaijiayu]

lfg!!!

[Elise Labs @elise_ai]

After getting early access to @OpenAI's GPT‑Live‑1, we’re bringing its natural, real-time voice capabilities to EliseAI, starting with healthcare.

Voice matters differently in healthcare. When someone calls, they’re often trying to schedule an appointment, get an answer about a bill, or figure out what to do next. A missed call or frustrating conversation can mean they simply don’t get the help they were looking for.

GPT-Live-1 makes those interactions faster, more fluid and more human, with speech-to-speech conversations that can respond naturally to pauses, interruptions, and the way people actually talk.

Here’s an early look at it running on EliseAI.

[Akshay Ramaswamy @TheRealAk914]

More info here!! https://t.co/W4I6woOJkK

[Akshay Ramaswamy @TheRealAk914]

We believe these more natural conversations will drive better call outcomes. In healthcare, that can mean more patients successfully scheduling appointments, resolving billing questions and getting the help they need.

The future of voice AI is bright!

[Shruti Goli @ShrutiGoli]

excited to finally share this. for the past couple months @elise_ai has been an early access partner with @OpenAIDevs on gpt-live-1.

we've brought it into production starting with healthcare. when patients call, they're usually trying to book an appointment or sort out a billing question. a bad call means they just don't get help.

full-duplex, handles interruptions, stays engaged mid-task. cascaded pipelines can't do this. patients get what they called for.

it's been great collab-ing with you @theteriyu @pbbakkum @SaiMandhan and the rest of the openai team!

shout out to the entire voiceAI team including @BoBowchan @TheRealAk914!