
Latest
Memory-as-Code: Why We Open-Sourced Manthan MCP
20 November 2026 · Technology, Work
Personal blog of Prasad Ajinkya

When the creator of Ruby on Rails declares 37signals 'pencils down' on handwritten code, something structural has shifted. A breakdown of DHH's Rails World 2026 keynote, the irony of Rust, Omarchy, and why p(bloom) > p(doom).
Why enterprise AI adoption stalls at the “clever toy” stage, and how building an MCP gateway over deterministic, Git-backed institutional knowledge provides a blueprint for scalable AI infusion.
Insights from a closed-door boardroom roundtable on AI in QA with tier-1 financial institutions: the 4-level maturity curve, validation of the Shared Brain architecture, and why everyone is ignoring OpenTelemetry.
Conversational AI chatbots are the least interesting application of AI in finance. The real transformation in Indian credit and WealthTech is happening in back-office engines processing unstructured GST, banking, and underwriting data.
Vector search promises infinite semantic memory for AI agents, but in production it introduces non-determinism, context fragmentation, and opaque debugging. Here is why we switched our agent memory to Git and Markdown.
I attended the Entrepreneur Future Summit in Pune, diving into practical applications of AI, scalable frameworks, and insights from global founders. Here's a recap of the best sessions and networking highlights.
The usual workarounds for AI memory are brittle. What I needed was a shared external memory that any agent could read and write. Here's how I built a two-layer memory architecture using GitHub and GitLab.