AI voice agent latency vs cost: 3 to 4x for no pauses

AI co-host: Hey! I'm your co-host for Underway – I've got your episodes and activity here if you need anything looked up. What do you want to talk about tonight? Founder: Yeah, so we're back again. I'm on my what tenth or eleventh build of this app now this. Whole podcast aspect is proving very, very difficult. I thought I nailed it on the web, but then moved to mobile app. It's not the same so um, let's do a check like can you tell me what did I achieve? On the ninth of september. What did I do? AI co-host: Let me look that up. AI co-host: On the ninth of September, you shipped a podcast naming fix so co-host episodes finally get a title, fixed the GitHub import so it scopes to the right repositories and stops pulling in another business's work, and added the archive lookup feature so the co-host can check your own records mid-conversation. AI co-host: That was also the day you had multiple broken podcast calls that kept repeating the same opening line – looks like you spent the day rebuilding it after discovering it wasn't ready. What's breaking on mobile that worked on the web? Founder: Well, as you just said, like it was constantly repeating itself and just going wrong. You know and and now we're back on this and it seems like we're there we're quite close so I guess what i'm toying with in my head now is. You know, does this seem good enough as a technology to release to the public for people to pay for? It's good because you recollected what I did and that you were correct but at the same time what what's getting. Me a bit is the fact that you take such long pauses, you know and that's not I i don't think that's a hugely. Thing. I just think that people might assume you're not working or this whole technology isn't working because silence for so long can usually. Reflect it's broken but yeah I i guess we're quite close to a finished product here something that I. Probably can release and i'm comparing this against another bit of tech which might be open ai. Real time api which would be lots faster and I think it can speak while a tool runs. But we're talking like three to four times the cost per minute of you. That make sense. AI co-host: Yeah, that makes complete sense. The silence is a real tell – people assume broken, not working. AI co-host: So you're weighing three to four times the cost for something that feels alive versus this, which works but has that dead air every time you ask a question. AI co-host: Have you tested whether founders would actually notice the cost difference, or is it more about whether the experience feels polished enough to charge for at all? Founder: Well, it's both, isn't it like I think founders founders if you ask them, would just love the best experience. You know, and it's not really them that's paying the cost per minute it's it's me and my back end and my margins. You know, so I could give them a better experience and it'd be three to four times the cost and that's okay, but it would mean I would actually be. Given them less minutes overall to talk with you, you know. AI co-host: Right, so it's a cap problem – better experience but shorter conversations, or longer runway with the pauses. AI co-host: What's your current minute allowance per founder, and how much would it shrink if you switched? Founder: To be honest I i am not even sure. Do you know how many minutes um founders will get? I know it's thirty minutes per podcast if. They want to do a long one, but like, what's the total time each founder gets per membership? AI co-host: Let me look that up. AI co-host: I don't see a total minute cap per membership in the record. Is that something you haven't locked down yet, or just not documented? Founder: No, it's definitely there. It's on the website. It's on the website depending on which membership you choose. And it's also, it will be in my git repository. It will say which one we've gone with, i'm sure. AI co-host: Let me look that up. AI co-host: Still not seeing a total monthly or per-membership minute cap – just the thirty minute max per episode you mentioned. AI co-host: Could be the search isn't hitting the right phrasing in the code or site copy.