Dean

Vibe Coding Is the New Product Management “There’s been a shift—a marked pronouncement in the last year and especially in the last few months—most pronounced by Claude Code, which is a specific model that has a coding engine in it, which is so good that I think now you have vibe

Day 7 - End of week 1 ✅ What I built this week: ✅ Core AI chat engine ✅ Knowledge base ingestion (PDF, DOCX, websites) ✅ Chat widget (embeddable) ✅ Human escalation detection ✅ Basic dashboard What's next: → Live support interface for agents → Analytics & insights →

Day 6 - Building the human escalation system This is the core of People Loop. The AI needs to detect: → User explicitly asks for a human → AI can't find an answer in the knowledge base → User is frustrated (sentiment analysis) When any of these trigger → instant handoff to

Day 9 - Added real-time notifications When a chat gets escalated, your team gets notified instantly: → Browser notifications → Email alerts → Shows customer sentiment → Displays conversation priority No more customers waiting because nobody saw the escalation.

UI design challenge today: Should the chat widget be: A) Always visible (like Intercom) B) Hidden until user clicks C) Pops up after 30 seconds Went with A. If you added a chatbot, you want people to use it. But I'll make it customizable. Every brand is different.

what I would be working on if I started another company today

Day 5 - Built the embeddable chat widget You'll be able to add People Loop to any website with just 3 lines of code: <script src="..."></script> And boom - AI chat appears in the bottom right. Spent way too long making the bubble animation smooth. Worth it.

Just tested People Loop with a 200-page employee handbook. Questions it handled perfectly: ✅ "What's the vacation policy?" ✅ "How do I submit expenses?" ✅ "Who do I contact about benefits?" This is going to save HR teams SO much time.

Day 4 - Knowledge base ingestion is working Now People Loop can: → Parse PDFs → Extract text from DOCX files → Crawl websites → Chunk everything into searchable pieces Uploaded 50 pages of docs, and the AI can now search through them in milliseconds.

The gap between "it works on my machine" and "it's ready for customers" is massive. Today's prototype: ✅ Answers questions ✅ Searches knowledge base ❌ No error handling ❌ No UI polish ❌ No human handoff yet ❌ Probably breaks with edge cases 30 days to close that gap.

Small milestone today: My AI chatbot just answered its first question correctly using a PDF I uploaded. Question: "What's your refund policy?" AI: *pulls exact text from knowledge base* It's basic, but man... seeing it work for the first time hits different.

Here's what most people don't know about AI chatbots: The AI doesn't "read" your docs every time someone asks a question. Instead: 1. Docs get split into chunks 2. Each chunk gets a "vector" (like a fingerprint) 3. When asked, AI finds matching fingerprints 4. Only sends

Day 3 - Built my first working prototype 🎉 It's ugly, but it works: → User sends a message → AI searches the knowledge base → AI responds with an answer Next up: Adding the "I need a human" detection logic. This is the fun part - seeing it actually work.

Building People Loop and making a big decision: Which LLM should power the chatbot? Options: • GPT-4o - smartest, but slower • Claude - great at following rules Thinking I'll let users choose per agent. Different use cases need different speeds.

Day 2 - Setting up the foundation Today I'm working on the core architecture: → Database schema for conversations → Knowledge base structure → Agent handoff logic The hardest part? Deciding WHEN the AI should escalate. Too early = wasted AI. Too late = angry customers.

Most AI chatbots are built on a lie: "AI will handle 100% of your support tickets" Reality check: ❌ Complex refunds ❌ Angry customers ❌ Edge cases ❌ Unique situations AI can't do it all. But AI + humans? That's the sweet spot.

It’s happening There is no way solo founders with AI agents as cofounders and employees won’t take over the world. Solo founders are like Olympic champions in agency&taste The thing I hate the most is to work with other people, so now I never have to

I travel to around 20 countries a year. In 5 years, I have spent thousands of dollars on eSIMs. The big eSIM providers out there pay pennies for eSIMs and resell them to users at a 500% markup. They buy ads and run huge, inefficient teams that the user pays for. So I built

Why am I building this? I tried adding AI support to my last project. The AI was fast, but when it didn't know something, it just... made stuff up. Customers would get angry. I'd lose trust. What if the AI was smart enough to say "let me get a human for this"? That's People

Day 1 of building in public 🚀 I'm building People Loop - an AI chatbot that actually knows when it can't help. The problem: Every AI support tool I've tried fails silently. Customer asks something complex → AI hallucinates an answer → customer frustrated. My solution: AI