The Bottleneck was Never the Model
The companies pulling ahead are not waiting on a smarter model. They noticed the bottleneck was never the model — and that since they fixed it, it has moved twice more. Read more: The Bottleneck was Never the Model
Product work, systems, and writing
I make AI products survive contact with real customers.
I've spent 11+ years in product, including 3+ shipping AI. Led $1B+ device-financing platform in 18 live markets, enhanced LLM search on Galaxy phones, drove experimentation culture and built the evaluation loops that keep AI features honest after launch. Over the years, I have worked across consumer, fintech, B2B SaaS and Enterprise SaaS domains.
Working Memory
- Context windows are a budget, not a bucket03 Oct'26
- Memory is a product decision01 Oct'26
- Write the eval before the feature30 Sep'26
- Latency is a UX problem wearing an infra costume28 Sep'26
- The harness is the moat25 Sep'26
The Log
What changes when the decision meets reality?
Agentic Memory Taxonomy
How we structure context across short-term reasoning and long-term storage in agentic loops. Defining the boundaries between RAG, vector stores, and persistent world Read more: Agentic Memory Taxonomy
RLHF for Product Managers
Reinforcement Learning from Human Feedback isn't just a technical training phase; it's the new "User Research". How to structure human feedback loops to align agentic Read more: RLHF for Product Managers
In progress
NextBite
A meal recommendation generator that explores how small models handle hyperlocal preferences.
Try NextBite