#262 · how to package a $3,500/m service
Speed and quality aren't a trade-off; you generally get both. Only in extreme low-latency cases (like real-time voice agents) might you sacrifice quality for speed, using cheap, fast models. For most use cases, the real trade-off is speed and quality versus cost.
Yes, I have a framework. First, set your accuracy constraints, determine how accurate you need to be, for example 95%. Second, build your flow. Third, choose the best model, starting with the most expensive option. Run an evaluation to see if the model's accuracy meets or exceeds your goal. If it does, reduce model quality (choose a cheaper model) and rerun the evaluation. For instance, GPT-4.5 might give 97% accuracy, but if your cutoff is 95%, you can try GPT-4.1, which gives about 95% accuracy, saving cost. Repeat this process, testing cheaper models until you find the right balance between cost and ability to perform the desired task.
I’m still not a fan of voice agents overall, even with low latency, because there’s always a chance a human will notice something off and realize they’re talking to a robot, which usually leads to disgust. I don’t want to risk that trust with people calling to book appointments or rebook after paying. While I’ve tried models like GPT Realtime 2.1 and they’re faster and handle interruptions better, they still have gaps in contextual awareness and can bug out occasionally—something humans don’t do. My take is that, for now, the reliability and grace of a real person outweigh the speed benefits of these AI voice tools.
Thanks for the request—I'd love to make a video on building a humanlike chatbot. Right now, the value of an AI assistant feels small because I can do tasks like scheduling appointments myself in just a few seconds, whereas earlier it would have taken much more day-to-day effort. The question is when the speed gap becomes so tiny that it's worth using an AI, or when the value becomes so high that I'd need an AI secretary. For me, it doesn't make much sense yet because I type extremely fast—about 186 words per minute—thanks to years of RuneScape grinding where I had to type the same thing repeatedly without copy‑paste. I'm also very fluid with computers and know hotkeys, so doing things myself is often faster than telling an AI to do them, especially given the current lag and the few words needed to prompt the model. That said, the landscape is changing: soon we'll be able to chain actions, so a single prompt could trigger dozens of steps like looking up flights, comparing costs, and booking them via APIs. We're approaching that point, but the models aren't there yet, which is why I've held back on chatbot videos. I will definitely build a humanlike chatbot for tasks like this in the future. A community member, Mehmud from Maker School, already sells Vapy chatbots and showed how powerful simple prompts can be, so a Vapy‑focused video might be worthwhile as well.
In most industries, AI voice agents for speed‑to‑lead don’t work well because even a small chance of the prospect realizing it’s an AI destroys trust and willingness to pay. That tiny risk can cost you a significant portion of revenue—often closer to 20‑40%—rather than just saving 1%. The only viable use is a highly nuanced approach: have the prospect fill a form, then immediately show a countdown (We're calling you in 3-2-1) and connect them to a live person, using the AI voice only to announce the imminent call and buy your team a few seconds to prepare. In that scenario the AI acts as a bridge, not the main interlocutor, and can be effective. Outside of that specific setup, AI voice agents are not ready for sales closing or broad lead conversion, and relying on them will likely hurt performance.