Event Recap: Build Smarter Voice Agents - New York Edition

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Event Recap: Build Smarter Voice Agents - New York Edition · Channel: AssemblyAI · 29:10 · Watch on YouTube

Topics: AI agents, AI evaluation and benchmarks, Multimodal models, Health and science

Summary

Builders from Morty and Fire Health discuss how they use voice agents for professional matchmaking and patient operations. They contrast open-ended, relationship-focused conversations with structured healthcare calls that require strict boundaries and escalation to humans. The panel covers trust, disclosure, observability, testing, turn-taking, context, and the difficulties of handling real phone calls and voicemail. They conclude by discussing the limitations of current voice pipelines and hopes for more capable speech models.

Key points

Morty uses voice calls to learn about users’ goals and priorities, helping it make more relevant introductions.
Fire Health’s agent calls patients with pre-procedure instructions, so its conversations are structured and should escalate to a human when they go off track.
For healthcare calls, the goal is to help patients complete a task without frustrating them or giving medical advice.
Morty evaluates outcomes such as whether introductions happen, alongside signals from call observability.
Fire Health found that disclosing immediately that the caller is an AI helps preserve trust.
Group video meetings create difficult turn-taking challenges; responding naturally requires very low latency.
Accurate collection of details such as email addresses and insurance information, and confirming patients heard instructions, are important challenges.
Voice-agent testing is difficult because simulated conversations do not fully reproduce how people behave on real calls.
Phone operations bring practical complications, including voicemail detection, call timing, and whether someone has answered.
The panel looks forward to more capable speech models and systems that can keep speaking while work happens in the background.

Chapters

Start	Chapter	Summary
00:00	Introductions and voice-agent use cases	The panelists introduce Morty’s matchmaking assistant and Fire Health’s patient-calling agent.
04:19	Designing for different users and goals	The panel compares open-ended conversations with structured healthcare flows that must stay within guardrails.
07:33	Measuring outcomes and building trust	The speakers discuss conversion metrics, call review, observability, and disclosing that an agent is AI.
13:18	Turn-taking, interruptions, and integrations	The panel describes challenges in group meetings, handling interruptions, and delivering reliable patient instructions.
20:04	Context and personalization	The discussion turns to remembering information across channels and managing tool calls without leaving callers waiting.
22:14	Scaling from demos to production	The panel examines the difficulty of evaluating voice agents and handling real-world calls, including voicemail.
27:00	What builders want next	The speakers share hopes for improved voice capabilities and speech models before the event closes.

Quotes

we’ll escalate that to a human. — 05:24

the answer is yes — 10:44

Is this your first phone call with an AI? — 12:17

you need like sub-second latency — 16:02

very different than humans interacting with their agent — 23:48

Concepts and entities

Morty, Fire Health, Owen, Dylan, AssemblyAI, agente de voz, confiança em agentes de voz, observabilidade, LiveKit, Google Meet, Braincert, Pindrop, Twilio, ChatGPT, latência, detecção de correio de voz

Full transcript (original language)

00:00 All right. So, at these meetups, what we like to do is actually bring folks who are building in voice onto the stage and have them talk to you all about what they’re building, how they’re building it, and answer some of the common questions about what people have questions about in voice, and also tell you probably a lot of things they’ve learned along the way building real-world voice experiences. So, maybe to kick things off and set the stage, I’m not sure each of you can just tell us about what your name is, what company you’re from, what are you building, and how does voice play a role in in that product and experience? >> Yep. I can kick it off. I am Owen, head

00:33 of engineering at Morty. It’s funny, actually, I used to have an Assembly AI app that I used to wear all the time. And then I lost it, so I have to maybe you can set me up with that at a later one. Um but yeah, at Morty, you can think of Morty as basically like if LinkedIn was just a person that you could call and text. Um so, yeah, Morty will make intros between the people, um founders and investors, investors town, whatever it is. I know actually for the event registrations for this event, um Morty used to give people a call. So, I’m

01:07 actually curious, like how many people have had you call Morty? Okay, wow, about a quarter of you. Uh that’s great. So, I mean, you know more about Morty than you do about me. Yeah, that was great. Um and I guess I also have Morty on a meeting here. Uh so, he can kind of participate in the meeting. Morty, do you want to introduce yourself? >> You can call me Morty. You can probably tell. I spend my time chatting with thousands of founders, creators, and investors to figure out their zone of genius so I can connect them with the right people to help them succeed. I’m basically a professional matchmaker who

01:39 lives on the internet, and I’m absolutely fascinated by how you >> Awesome. Wow, that’s better explanation than I knew. So, I don’t know. Maybe uh Morty can just take it from here on out. Um yeah, live voice. I mean, for myself personally, there’s people who I like texted a lot, but I’ve never met with, like maybe I just messaged them back and forth a bunch. I don’t actually feel like I know them. You know what I mean? Like I don’t actually know them until I meet them in person or at the very least have a phone call or a video meeting of some sort. Because when you’re when you talk with someone

02:11 over voice, you just say so much more. Whereas over text you’re very dry. And so we found that with 40, having 40 being able to have phone calls with people and joining meetings is it gives 40 the ability to learn so much more about you and your priorities and your goals and that allows us to do that matching and that intro so much better. >> Just a clearer intro than me. That’s how >> 40 put out. I’m Dylan. I’m from Fire Health. We work with the musculoskeletal clinics and we have a whole suite of AI and tech tools for them, but one of which is our

02:44 voice agent. Uh so we focus on that and then I’ll give it to you and you can do it, but uh it’s The voice agent essentially calls and calls patients before their procedures and gives them instructions or say what to take or what not to take when they need to stop or when they do not need to do that. >> [laughter] >> Well, so for us it’s just a lot of these health care operations still happen by via phone. That’s how patients call. You can try to put an online scheduling link, but patients are still going to

03:17 call you and I mean they’re going to answer the phone. And a lot of them are understaffed in there and especially when you’re kind of calling out cases, it’s longer. The patients are in contact there and it’s going to come through both hands. I think it’s Most of the patients we talk to are older. So I guess like our our demographic is pretty highly different than the other patients we have. >> Uh we are used to that. That’s why >> Who? Hold it up. >> Yeah, yeah, yeah. >> Say all right. All right, all right. Thank you. I need my my training on being

03:48 So you’re saying that the patients are different and yes, the customers are very different. Yeah, so yeah, I was talking to somebody here today where it was like interesting because a lot of voice agent applications are like need to happen in the real world on the phone. The user is a little bit more old-fashioned when it refers to the phone. And now they’re talking to an AI on the phone, which is like they’re not a tech enthusiast. They’re not excited to talk to an AI, but like they’re calling me on the phone and now they’re now they’re like talking with uh uh tech advance tech support. Uh maybe that’s a good place to start. I

04:19 mean >> Your agents operate in a very different environment. They’re quite as highly conversational. It’s very relationship-driven. They’re working in a more of a regulated environment. Probably talking to like very different folks than like the tech founder demographic that we’re already talking about. How does that end customer base change the way that you design your your voice agent and your experiences with voice? >> Yes. So for us, a lot of the conversations we have are pretty um it’s pretty structured. So I talked to

04:50 Corey the other day and Corey can go off in all sorts of different directions, which I’m sure is very hard to program all these different places you can go. But for us, it’s really like you have to walk through this set of questions and it’s kind of a flowchart to it or you’re like signing a patient up and collecting their email and phone number and like insurance information and off you go. So I guess any really don’t want it to go off in any direction you like to. It’s a big issue if it starts trying to give them medical advice. Like they’ll complain about like something’s going on and you’re going to lose the patient and you have things. So um speaking of on task and like if they

05:24 go off track, it’s really like we’ll escalate that to a human. Yeah. >> Yeah. So it sounds like you have a pretty pretty regimented like step-by-step flow. Probably a lot of guardrails in there as well. Yeah, it looks like a And then uh for you, like maybe what are some of the key things in like that voice experience that can make it really good for the end user? >> Um so I I guess for us it’s The part of it is like we’re not really trying to delight this user. It’s not like they’re excited. It’s kind of like trying to make somebody get to their end goal and are not unhappy with us.

05:55 They’re not like complaining to the clinic like, “Why do my son not talk to me humans anymore? Like I want to call you and” Um so it’s kind of like not not make this user unhappy. That is the the goal here to make them happy. >> Does that make sense? >> Yeah. >> If Morty starts giving you medical advice, hang up the phone. >> [laughter] >> Yeah. Um I mean yeah, it’s it’s so interesting because with Morty, as if again I can hammer this again, but if I was just talking to you, we would take the conversation in all kinds of different directions. And especially now

06:27 as the models keep getting better and better, it unlocks the ability for the conversations with Morty to be better as well and just more enjoyable and and more insightful as well. Um so I guess we I hadn’t thought about that before, but relatively to you, we lean to that quite a bit. Um another thing I think is particularly interesting and just around how we like design Morty, which is very unique to Morty compared to other AI assistants, let’s say the just general agents, is that for most agents,

06:59 it just wants to help you as much as possible. It’s like you’re talking to code actually trying to cloud code. It’s like trying to do whatever you want, right? Whereas with Morty, there’s an entire network of 200,000 people who Morty is also, you know, has to take into account. And so Morty can’t do something to help you and be expensive other people in the network. Like Morty’s not going to intro you to the the best investor Morty knows unless you’re also, you know, at that caliber. So that’s maybe a slightly less about the the voice experience, but it’s like coming from a very different angle

07:33 than an assistant. >> Yeah. >> And how do you judge what is like a successful conversation for Morty versus not? And we going to be using like qualitative metrics, quantitative metrics? Like, how do you judge that 40 achieved its goal and on a healthy person connecting with other people, making them feel good, making them excited about the product? Like, how how do you do that in that scenario? >> Yeah, I mean, there’s a whole bunch of different things. The overall, um we have different product metrics we look at, like um

08:05 how what’s the rate at which 40’s intros convert, right? Cuz a lot of the time >> people will just not respond or will say, “Oh, no, that’s not a fit for me right now.” or whatever. Um so, we look at those and that is all just like downstream of the entire experience, including the phone calls. Um and it’s actually been really cool recently. A lot of those have been going up. Um so, now when you like tell 40, “Yeah, I want to meet this person that you’re proposing to me.” there’s more than a 50% chance that that intro will actually happen. Um and so, that’s been like

08:35 really cool to see. Um and then in terms of like the voice, we have different agent observability tools that allow us to get a bit of an insight into like, “Okay, are people yelling and swearing at 40? Okay, probably not a good thing.” Um versus, “Yeah, are they having a good experience?” and things like that. Um and so, that’s that’s something we’re continuously improving as well. >> Yeah, I’m curious for you, Dylan. Is it more quantitative based then? Like, did they achieve the goal or do you look at some of these more qualitative signals as well about the what actually happened

09:07 on those calls? >> Yeah, so we do know whether they achieved the goal, but a lot of things that goes wrong is is we think they achieved the goal and it didn’t. So, we really had to do a lot in the beginning a lot of human review of just every phone call or email and look at that transcript. Um and make sure it went well. And having really good transcripts is very important for that, but um And trying to set up our boundaries as we go through uh just things like uh like if kid says that they got completion, but they didn’t. And like, this is wrong wrong here. How do we adjust the prompt so we can work? How do

09:39 we test against like when we’re using the phone because obviously it’s a problem. Obviously it’s a phone and all so we use like a tool called Braincert for sending just getting the observability of all the calls. And we also do a little bit with our testing we use a service called Pindrop and then there’s like voice agents for testing voice agents so testing these can be very tedious because I’ve spent so much time just like on the phone playing on repeat for different call scenarios and

10:13 trying to see that up is intact. >> And I assume a key component of your your users actually achieving that goal is is trust, right? So they they believe hey this is like a trustworthy AI that I can give my personal information to. I guess I’m curious specifically on your side like how do you get users comfortable talking to an AI and how do you like what are some of the things that might break that trust and cause them to not convert through the current flow? >> Yeah so the big first thing is we weren’t sure at first like do you disclose right away if this is an AI and

10:44 I think for us the answer is yes like says immediately be there and then so they get you with names with a lot of time to say I’m Sarah it’s like hey I’m Sarah this AI is supposed to be kind this clinic like kind of doctor’s office it’s kind of right there up front because uh if you don’t tell them right away they’re almost certainly going to figure it out at some point and then if they thought it was human for a while that that like breaks trust and it’s I think it definitely do make you unhappy. We didn’t we did have a couple calls in the other where we didn’t tell them we were an AI at the end they’re

11:15 like I have one more question like are you are you an AI? And he was like oh like okay we we really need to tell them right away and this was just a huge trust fall around. So for us it was also where probably I found Google Voice to be the official way is it going so doing all that stuff to like we used to do earlier, but uh setting up all the stuff ourselves so you don’t get flagged with spam. Uh it says we’re paying hopefully when it calls it. caller ID That’s what separates us.

11:45 >> Here’s an example of what he said. Obviously, it’s less about like this structured conversation, but how do you see what he building like trust with users? It’s funny cuz when I had my first call with him, I was like very skeptical. Then after about 5 minutes, I was like, “Oh, this is fun. This guy’s going to be helpful.” It’s interesting. So, I’m curious how you you you like trust in this context. >> Yeah, and we I hear that a lot, to be honest. So, I am curious also afterwards chatting with uh all of you. Um that first experience, a lot of people seem like scared to like talk to erhöhte. Um so, I I am curious to see if

12:17 there are people’s, you know, experience with that and how we can improve that. But, yeah, with erhöhte, we try to have erhöhte be the most charismatic, funny, self-aware AI possible. And so, typically, the first thing that erhöhte will say when you hop on the phone is, “Is this your first phone call with an AI?” And that like breaks the the the fourth wall almost and like immediately like you know, resolves the question and the tension about being like calling an AI or not.

12:48 It’s like, “Yeah, I’m an AI and like we we can uh you know, be okay with that.” Um but then, as you were saying as well, like 4 minutes in, you almost like forget. Like the conversation is going so well and it’s like an engaging conversation and you like almost forget. So, it’s like instead of not telling people, but then they figure out and they’re like super freaked out, it’s the other way around. It’s like we’re very up front, but then it’s just a great conversation and you get into the flow. >> Yeah, I’m curious how you handle some of

13:18 the like messy things in building voice agents. So, things like turn taking interruptions, these calls going off script, who knows what else you got to that, something like “Bardi, how have you built Bardi to be off defensible with like a guardrail perspective for going off-script, but also to handle some of these like turn-taking interruptions, which especially with Bardi because it feels like your friend. You are kind of like interjecting and going back and forth in a way that’s very different from all the other voice agent experiences.” >> Yeah, I mean, I think in general for just one-on-one voice conversations,

13:52 it’s like basically solved in terms of turn-taking. Like, we use a sound we use live kit and it just like everything works really well. Uh voice voice agents are like very good at one-on-one conversations. I think actually like for us one of the huge challenges is now we have Bardi on video meetings, like Google Meet. Like I have like on my phone right now, this is I I’m from Canada and I don’t have like calling in the US, so I’m on a video meeting with with Bardi right now, that’s why I’m calling him in right now. Um and so Bardi will make an intro

14:25 and then schedule it and then join that meeting, that Google Meet. And so you can imagine that when you have multiple people on a meeting, then you have Bardi there as well, determining when Bardi should speak or not, that’s like the really challenging part. Um because false negatives and false positives are both like catastrophic. If you say, “Hey Bardi, how are you doing?” and he doesn’t reply, that’s really bad. If you like are not talking to Bardi and then he just randomly keeps jumping in, that’s really annoying, you’re never going to be invited to a meeting again.

14:58 Um and so that has been very difficult and then let’s say you like do it as well as possible. Then if you’re like classifying whether Bardi should speak or not, then it’s like the latency on that and everything. Um and so we There’s one engineer on the team in particular who has spent like a ton of time working on this and getting it into like a really good state. And I think it’s really cool because there’s like no one else really has having an AI that can probably participate

15:28 in video meetings and where you can. And essentially part of the way we do that is we do have like a classifier classifier deciding, okay, should we already speak or not based on the conversation so far. And then we preemptively generate like the entire voice pipeline. So like constantly Wording is generating, you know, responses and then we wait when it is time to respond, he speaks very quickly. Which again, like the turn taking when you are in a video meeting you don’t realize this, but

16:02 as people are ending their sentence, you start jumping in. Like like right as they finish and if you wait until that’s done and then go through your little pipeline, respond 3 seconds later, it’s way too late. So you need like sub-second latency in order for it to actually feel like a natural conversation. So that’s been like a big challenge, but I think something that we’re pretty proud of to have built. What would you say about that, Wording?

16:49 All right, thanks, Wording. I’d like to hear from you uh specifically on this like handling messy stuff. I’m I’m curious like what are some of the I know words, entities, whatever that are hard to get right and uh especially maybe like you probably have a lot of connections and integrations downstream to do these calls. Really, how how are those set up and maybe some insight there would be interesting to me. Sure, yeah. So Wording voice engine is using to systems which are Vocals not pretty great if any of you guys have worked with them. Um you guys have an idea like similar to

17:21 what is a spin-off from like browsers to IVR and routing agents from there just to market up to you guys to use it or like that’s really what we use and like other routing agents. Um Like earlier, yeah. But anyways, I guess what I’m Those integrations are for you to see like some of the worst software you can do in like a like change the name of that in the creative side of conversation. They use a thing uh like it’s slow. Uh sometimes it increases expectations. Uh like oh, we’ll look that up for you.

17:52 Like maybe some type of sound. Um Um and then also you do need like you’re saying you usually being the collection of the email address. Like you don’t want to mess that up email address or phone number collection um or like their insurance number collection. So really mainly the um the seeing people and understanding what they do to you or something. Getting to review. Um And then every time you do eventually have a long time that you do it with um So with our pre-op instructions, there’s like a set of instructions that need to be delivered to the patient. I mean you have to be sure that they heard it. And um it’s hard with the cascading voice

18:28 uh like the agent stack we use. Um you don’t know for sure that it wasn’t being opted in with uh saying these things. So it’s like we know the outline of generic phrases spoken text is going to say this and send it off. It’s we’re playing the audio and then the patient interrupts midway. Um The agent that we have right now doesn’t listen to itself speak. So it knows what it wants to say. It doesn’t know what it actually thinks it’s going to say. Um The first and like at this point you just can restart the whole like instruction block of it and try it on. You know, if they put on a long

18:59 instruction to break it down and not have a too big of an instruction set. But like you don’t want the AI to keep talking to somebody cuz if someone interrupts it, it’s going to have no idea where it was. And you have to restart or skip forward and you lose everything. Yeah. >> the context of the patients and the doctor they’re going to, etc. Is part of that voice engine experience or is it more like unscripted or sorry, scripted in the sense where there’s no context for each for each individual user? Like you calling me by name and mentioning what doctor you’re going to. >> Yeah, we do a little bit of that. If you

19:34 have ones like we know what procedure is and talk about that. You can ask us any questions about what procedure is going to happen. For like I think I see who wants to obtain the patients. So we have no information on what first to call them. You don’t You may just be able to contact but I think you don’t save like clean phone calls feature of value in this patient like holiday this is who knows. They say they are or they’re about it. They’re definitely like over 10,000 calls on It’s about three years risk to leave anything that seems like a complication

20:04 of paying to have to do with anything like really it. >> There’s maybe a few talking about it from the Bodhi side. You probably have a lot that goes into like context and personalization. How how have you set that up? Like what does that look like under the hood for Bodhi? >> Yeah, within a single voice conversation, I mean LLMs have huge context windows now so it can just handle all of that naturally. Uh the difficulty is then when you go to multiple channels trying to Bodhi over email, text expecting Bodhi to remember all of that. Um and then some people

20:35 have been talking to Bodhi and having calls for like two years, right? Um so like how do you handle all of that? Uh it’s a bit of a nightmare. Um but yeah, right now we recently completely like revamped Bodhi’s harness, you can say, over the text messaging channels and so now that works very well in terms of like you know, similar a similar way of handling everything to say like GrokBot or coding agents where there’s compassion, there’s a memory and there’s all that different things. So,

21:08 that works quite well. Also, our whole over voice voice is still like, you know, has room to improve, but it’s actually right now in development cuz they do to overcome the ads with our voice. Um I it seems like not too many people are doing this right now, but the a similar system I mentioned where it is possible and we can get into that a bit more, um but having those same things where there’s the chat, there’s like the ability to search previous conversations, all this stuff. The challenge with voice is then if you do

21:41 that and they say a tool call, are you just leaving, you know, the person hanging for like seconds while doing that? Um and so, for voice, there’s there’s a bit more complexity, and you have to be a bit smarter about it. Um again, we can get into that more, but say with like ChatGPT, I have like the new the new thing they released, which has like full duplex, and there’s like a front end and a back end model, so that allows it to um continue talking while in the background a bunch of things are going on. Um and so, I think that’s like pretty

22:14 interesting, and it would be in the full overhaul, as I mentioned. Um so, I’m excited for that soon. >> That’s awesome. Um I think we have one or two more questions. Sorry for interrupting today. We’ll do this. Um I’m curious maybe what surprised you most as you scaled your voice agent. It seems like it’s really easy to build some demos and kick quick things out that look great on screen and you tell me to meet up from the bathroom. What surprised you most about taking that like demo or your new POC all the way into production? >> I mean, one thing I’m surprised with is that

22:45 like we haven’t run into like more issues. I think a lot of the providers we’ve been using, considering how like complex voice AI is and everything happened in the background, like yeah, it’s it’s it’s been good. Um but one challenge compared to say more like text base agents is actually verifying and like essentially running e-values or or simulations on it because

23:15 with the text base agent what we do at Audi we have something that we call messaging agent simulation scenarios where we can set up an exact scenario. Okay, like here is this person and here’s the situation they’re in and this is exactly what Audi should do in this situation and we have like 200 of these so we know what Audi should do in like every situation and all their like behaviors and everything like that. With voice you can do a similar thing but then when there’s actually a person talking that ends up being different

23:48 than like you know, simple text input or even you know, in the case there’s these services which are really cool which have other agents which will then interact with your agent to like test different scenarios or edge cases and things like that. But even an agent interacting with your agent is like very different than humans interacting with their agent so there’s always things that go wrong which are really hard to to catch ahead of time with e-values and simulations so that’s something that we’re

24:19 thinking about especially as we develop agents to something more sophisticated as I mentioned because right now over over voice that’s like not crazy complicated for voice. Um for one to have longer two or pulse in the project kind of thing um going to be some some fun stuff to figure out there. Yeah, so what is going to be just nothing like to do with how about new calls is a lot harder than you would think it would be. Um so like most time we call people and they’re not

24:51 going to answer if they go to voicemail. So if an Audi goes to voicemail then kind of voicemail like and um there’s all three numbers that of detect voicemails perfectly when I’m like listening to pure silence and you say something. That’s actually really hard. I mean, people tell me that this like sound perfection because they would have like the um >> [clears throat] >> tokens and all that. It would be like the voicemails and all that. A lot of times it’s just like hearing text to voicemail. Does it too early to start trying to hear a voicemail and then making sure you know like restarting stuff at the right time and do it good here and there and all that. Really it’s like all of these voicemails

25:24 and all that and then voicemails and all that and then location say it calls back and then it will call us back and then the whole thing flowed. Um look at beyond these voicemails. That was the kind of new um When we launch uh if you have the call like everyone in your company and we’re okay. So we’re like patients and then immediately on the first day we bring new iPhone screeners. Um so these uh new app voicemails they ask like explain your armor, why you’re calling in and then all of them talk honestly. I’ll get back to you and this whole thing that you have to explain.

25:55 It’s a first call in and then we from what person is there and then we jump to my friend there and we all hear my saying “I’m trying to ask you a lawyer.” I do the flow. Um I mean, it’s like this whole and ashes deals very not uh It’s not the box of laughs that I guess. And I would I know a person who had a vehicle that had that in the uh I guess dealing with that sort of timing and also the phones like um So first cycle use like Twilio and then Twilio actually initiates the call also and then all these things are in my field. You have a sub pump and then

26:28 once they fix all that and then place a call to the location. Knowing making sure that life to knows when person has to picked up. So you can start doing this time of like sound perfection of the week uh the question or something. Uh let’s go. I guess it hasn’t been a reason back of what was going on here and then all my uh I got to go to Twilio and then hopefully find a promise like there with an athlete. I think the mail data on the ring and then I lost the video. know uses all the data by And for this reason, I keep holding my phone because not

27:00 I don’t know. I’m surprised that you’re [clears throat] seeing me. >> Yeah. And you are more of the audience, I guess. Maybe you’re looking at And usually I’d say what are your predictions, but we’ll do it the other way. Waving your magic wand, what do you want to see solved in the next say 12 months? Next 12 months that will help you as builders. >> Maybe going off me, but I would be as well to see five >> phones like lighter kids. I don’t know man. >> Hopefully you guys have looked into it. I think the five phones of new associates called my phone to be >> Right now we are forced to inspection after the ball is in our court. >> Through some of the eyes like kind of

27:32 dumb or dumb that beyond the valley. >> The same way I am. >> But it’s good to say the name of the person you >> It’s not. >> I don’t know if it feels so long. >> You like to get >> You just want to say anything like >> You can’t bring your son in here. >> He transferred me to >> I really say >> I really say my rebound >> I just say >> I am. >> You like >> I am currently >> You see I am. >> I am currently

28:03 >> I am currently >> I am currently >> For me, I think I am just really excited about giving so much that I have just mentioned in like people now use the high quota products like the capabilities are are quite good and I think we’re going to start expecting more of that in terms of like voice agents and really the capabilities and and what all they can do. Um and right now we’re currently hearing the bottom neck right having to do with the text speech pipelines where if

28:38 you have to transcribe the text first there’s always going to be things that that’s wrong about like you know isn’t quite right. Um so I’m looking forward to like the future as like uh speech and speech models developing even further, getting more intelligent and everything. And then also that uh you know, front and back end on the lecture where you know, you can have the engine continue to speak while doing all kinds of stuff in the background. And so I’m looking forward to you know, seeing that develop more. Thank you all for coming out tonight. See you in the

29:10 next hour or so. Thank you.