What started as a simple experiment has turned into 7 different AI call assistants, with another one currently in development. Some of the capabilities I've implemented so far: • Inbound call handling• Outbound calling workflows • Lead qualification• Appointment scheduling • Rescheduling and cancellations • SMS confirmations and follow-ups • Reminder calls before meetings • CRM updates and record management • Call transfers when needed • Multi-step business workflows The most interesting challenge wasn't getting the AI to talk. It was getting it to remember. One of the assistants can remember previous conversations across multiple calls, retrieve historical context, and continue the conversation naturally instead of treating every call like it's the first interaction. Building these systems taught me that the hard part isn't the voice model itself—it's the orchestration behind the scenes. Memory, workflows, CRM integration, scheduling logic, error handling, and all the small edge cases end up being where most of the work happens. Still a lot to improve, but it's been fascinating seeing how far AI voice technology has come in such a short time. Curious what everyone else is building in the voice AI space right now.

Recently worked on an AI system for a music industry website redesign. There were two parts to it: 1. A public-facing AI chat widget for general queries + lead capture 2. A private, full-screen AI assistant inside the client portal. The second one was more interesting. The client didn’t just want “AI support” — they were concerned about users overusing it and burning unnecessary tokens. So instead of a typical chatbot, I designed a session-based interaction system: - Each user gets a 5-minute active window - At minute 4 → a warning is triggered (“1 minute left”) - At minute 5 → session ends + redirects to an external GPT - Then a 5-minute cooldown kicks in - After cooldown → user can start again It basically runs in a loop. What’s interesting is how this changes behavior: Users become more intentional with their questions instead of treating it like an endless chat. Also ended up optimizing one of their internal workflows in the process: A task that used to take ~6 hours manually is now done in under 10 minutes. No fancy theory here—just structuring AI usage in a way that actually makes sense operationally. Still experimenting with how far this “controlled AI interaction” approach can go.

Anyone working with the new AI Agents yet? I build an agent that know all of my IP and cross references it when producing content for social. I’ve been meaning to do this with Pinecone but didn’t want to go through the ingestion process. Now I can do everything in Make. It’s dope