The AI Flywheel: Why Some AI Products Become Viral
Why do some AI products get smarter the more they are used, while others stall? It's because of the data flywheel: the loop that separates a feature from a moat.
What Is Context Rot And Why Your AI Gets Dumber The Longer You Talk to It
Bigger context windows didn't fix AI quality but made it easier to overload the model without noticing. Here's what context rot is, why it happens, and the TRIM framework to fix it.
What Is Synthetic Data And Why Is It Replacing Human Labelling?
Most AI training data is no longer written by humans. Synthetic data, generated by AI models, is faster, cheaper, and often better. Here's how it works.
Airbnb Replaced Press 1 With Four ML Models That Actually Listen
Airbnb replaced rigid phone menus with four ML models running in real time: ASR, intent detection, semantic retrieval, and paraphrasing. WER dropped from 33% to 10%.
How Netflix Turns Generic LLMs into Recommendation Engines
Netflix's recommendation engine drives 80% of viewing. Here's how they built a post-training framework to adapt LLMs for their catalogue from SFT to RL.
Claude Channels lets external systems push events into a live Claude Code session. Here's what it is, how it works, and why it matters for AI product teams.
What Is Agentic RAG And Why Regular RAG Isn't Enough Anymore
Classic RAG retrieves once and generates. Agentic RAG retrieves in loops, checks what it found, and goes back for more. Here's what that means for the AI products.
WTF Is The Difference Between Claude, Claude Code, Claude Cowork
Anthropic now has three products called Claude. One is a chatbot, one writes your code, and one manages your files. Here's how they're different and when to use each.
Everything You Must Know About Mixture of Experts (MoE)
Mixture of Experts (MoE) is the architecture behind GPT-4, Gemini 1.5, and Mixtral. Here's a PM-level explanation of how MoE works and why it matters for your API budget.
Notion reached a $10 billion valuation on just $344 million raised. Here's how a free personal tier and an accidental template ecosystem became its growth engine.
