TLDRocket
Sign in

Building an AI Coach to Help Tame My Monkey Mind

Eugene Yan

A guy built himself a voice-based AI therapist using Claude 3 and a phone number. You can even call it right now to try it out.

Based on reporting by Eugene Yan — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

Eugene Yan has a familiar problem: a restless mind, imposter syndrome, and the endless chase for the next dopamine hit. His usual fixes—meditation, morning pages, gratitude journaling—all shared the same blind spot. They're introspective. None of them talk back.

So he turned to ChatGPT as an ad hoc therapist, and found it genuinely useful for a year. Then a tweet from Swyx nudged him toward voice, a modality he'd previously written off thanks to years of shouting at Siri and getting nowhere. This time it stuck. Yan built a voice-based AI coach he now talks to on his morning walks, using it to rehearse hard conversations, sort through half-formed ideas, and reflect his own emotions back at him before he even reaches the office.

The setup is Claude 3 Sonnet paired with Vapi, a voice API platform that handles the messy parts of conversation—knowing when to pause for an interruption, knowing when to shut up while someone thinks. Yan says that combination is what finally made voice feel natural instead of stilted. He named the assistant Tara, gave her a system prompt asking for casual speech, filler words, and radical candor capped at 200 words a response, and hooked her up to a real phone number anyone can dial.

He's also published the entire recipe: create an assistant on Vapi, write a first message and system prompt, rent a virtual number for two dollars a month, assign it, publish, done. Ten minutes, he claims, and no coding beyond copy-pasting a prompt. It's a small, almost domestic use of AI—no benchmarks, no funding round—just someone testing whether a language model can substitute for a sounding board that's awake at 6 a.m.

What's notable isn't the tech stack, which is fairly plain, but the framing. Yan isn't pitching a startup. He's describing a coping mechanism, one he's openly inviting strangers to dial into, which says something about how comfortable talking to a machine about your anxieties has quietly become.

My take — AI-written commentary, not fact-checked reporting

Personally I think this is the most honest use of an LLM I've seen in months—no roadmap, no monetization, just a guy admitting a chatbot helped him process feelings faster than journaling did. The bigger pattern here is that voice, not chat windows, is what will make AI companionship stick, and most labs are still underrating that shift.

Read more about this at: Eugene Yan

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.