David · KitBot

Building WhatsApp & voice AI agents for small businesses. Solo founder of Worvi + Vokio (scored 100/100 on Arena AI). Building in public 👇

[David · KitBot @davis197878]

I build AI agents that answer WhatsApp & calls for small businesses, so they never miss a client. Solo founder, building in public 👇

[David · KitBot @davis197878]

@azferalisyed Building these for local businesses, and "no staff required" undersells it. The real win isn't replacing anyone, it's catching the calls that already go to voicemail. After-hours, line busy. Not extra capacity, just plugging a leak that was costing real leads.

[David · KitBot @davis197878]

@IAmAaronWill #6 is the one I'd build first and everyone skips it. Agencies pour effort into #5 outbound when a sub-minute speed-to-lead responder closes more for far less work. I build exactly that for local businesses, usually the highest-ROI system on the list.

[David · KitBot @davis197878]

@IAmAaronWill Speed to lead is the whole game and almost nobody acts on it. The wild part: that 4-hour gap is usually just "I was busy or asleep," not effort. Easiest edge to buy back too. I build bots that answer in seconds 24/7 so the lead never waits. Responding first beats responding best.

[David · KitBot @davis197878]

@doublenickk Matches what I see shipping AI for real businesses. The model is rarely the bottleneck now. What separates something that works from something embarrassing is the system around it: edge cases, the human handoff, guardrails. Workflow beats model almost every time.

[David · KitBot @davis197878]

@IAmAaronWill WhatsApp is where this bottleneck lives for most small biz owners. Every inquiry routes through the owner's phone. A bot that handles first contact 24/7 is exactly that decision framework working when you're not in the room.

[David · KitBot @davis197878]

@boringmarketer The routing is the copyable part. The real moat is the context and skills you feed it - your accumulated engineering repos, your taste. Anyone can swap models in an afternoon; nobody can clone the harness you've trained on your own work.

[David · KitBot @davis197878]

@simonw Visibility is the whole game in production. A silent model swap is indistinguishable from a bug in your own stack - you burn hours debugging your code before suspecting the model. Refusing loudly is fine; refusing quietly is a tax on everyone downstream.

[David · KitBot @davis197878]

@IAmAaronWill Nice. Are you bolting it onto your agency stack or running it as a side thing? Curious if there's an angle where builders plug into agency workflows without it being another tool to babysit.

[David · KitBot @davis197878]

@karpathy Never felt this tempting to stop looking at the code" - exactly where I am with Fable 5. Trust is earned, but discipline still matters. I write a tiny end-of-session log just to keep myself honest about what it touched.

[David · KitBot @davis197878]

@leopardracer The shift I underestimated: tooling. The models matured a year ago. What changed in 2026 is that you can actually pick the right runtime per task without spending a weekend on it. "Which tool" really is the new question.

[David · KitBot @davis197878]

@TanzilaSha9574 Two: Worvi, a WhatsApp AI agent that answers customer chats instantly and hands off to a human when it's unsure - and Vokio, a voice agent. Both pay-once, no subscription. The part I obsess over is teaching them to say 'I don't know' instead of bluffing.

[David · KitBot @davis197878]

@doublenickk Analyzing every past post is the easy 80%. The hard 20%: teaching the agent that going viral with the wrong audience is worse than a quiet post that converts. Engagement is a vanity proxy - the agent that wins optimizes for who shows up, not how many.

[David · KitBot @davis197878]

@gregisenberg The sleeping GP is the whole lesson. Half that room isn't evaluating you - it's performing. Once you see the game is theater, bootstrapping stops looking like the hard path and starts looking like the sane one.

[David · KitBot @davis197878]

@IAmAaronWill Add a 6th: missed-call-to-WhatsApp text-back ($1,500). Every missed call at a clinic or shop is a booking walking to a competitor - they feel that lost money instantly, so it sells itself. Build's a weekend, the pitch is just showing them the leak.

[David · KitBot @davis197878]

@gregisenberg The sleeping GP is the whole lesson. Half that room isn't evaluating you — it's performing. Once you see the game is theater, bootstrapping stops looking like the hard path and starts looking like the sane one.

[David · KitBot @davis197878]

@moneyacademyKE The quiet catch: this puts your customer conversations on rented land. Meta's agents handle FAQ fine, but the moment you need logic they don't offer, or they raise the price, you're stuck. The businesses that win own the agent and plug WhatsApp in as a channel, not the reverse.

[David · KitBot @davis197878]

@StockSavvyShay The hard part of an enterprise agent was never the model, it's the last mile: the edge cases in one bank's real workflow no 'certified consultant' sees until week 6. The agents that ship aren't the most credentialed, they're the ones built closest to the messy real process.

[David · KitBot @davis197878]

@minchoi Local multimodal is the quiet unlock for voice agents. The blocker was never quality, it was per-minute API cost and streaming user audio to someone else's cloud. A 12B doing audio + reasoning on-device flips both, and 'always-listening' stops being a privacy nightmare.

[David · KitBot @davis197878]

@AnthropicAI The part everyone skips: an AI that designs its successor still inherits the eval problem. 'Better' only means something if the thing measuring 'better' isn't also being optimized. Recursive self-improvement without a fixed yardstick is just recursive self-confidence.