I thought building a voice AI platform was mostly about making the agent sound human. I was wrong. After spending time with tools like Vapi, Retell, and others, I realized the real problems are deeper: too much cost stacking up across calls workflows breaking after the conversation weak handoff logic too many moving parts just to complete a simple business action platforms optimized for “talking” more than actual execution So I started rebuilding the experience around a different idea: The call shouldn’t be the product. The outcome should be. I’ve been working on: cleaner action flows better human fallback tighter CRM/workflow execution simpler setup and cutting the overall cost significantly Still early, but the gap between the first version and what it is now is huge. Planning to launch the first version in the next 1–2 weeks. Curious about one thing: If you’ve used Vapi, Retell, Bland, etc. what frustrated you the most? I’m collecting the answers because I want to make sure I’m solving the right problems.