What I learned running local LLMs

• #AI
• #LLMs
• #Development
• #Infrastructure

Jul 03, 2026

Scaling my development environment by using specialized local LLMs for different tasks.

Making AI Coding Agents Smarter About Context

Jul 01, 2026

How two lightweight filters — coordination message detection and session-level deduplication — cut wasted context injection in AI coding agents.

Dev Setup

• #macos
• #terminal
• #tools
• #productivity

Jun 15, 2026

A rundown of the tools I use day to day, from the terminal to local LLMs.

Jun 14, 2026

June 14th. It is raining. For me, rain usually means one thing pakoras, samosas, and a hot cup of chai. So there I was, sitting by the window, watching the rain and enjoying some chai, when my mind started wandering to a completely unrelated topic developer productivity. Are we actually getting more productive with AI? And how would we even know? Most organizations try to measure it with DORA metrics: - Lead time for changes - Deployment frequency - Change failure rate - Mean time to recovery (MTTR) These are useful. They tell us how efficiently we are shipping software. But - Are we solving the right problems? - Are we reducing complexity or just moving it around? - Are we making good technical decisions? - Are we making it easier for other teams? - Are we leaving things better than we found them? AI can help us write more code, close more tickets, and merge more PRs MRs. But output is not the same as value. It does not push back on bad requirements. It does not ask why. Take code reviews as an example. AI can catch formatting issues, point out potential bugs, and suggest improvements. But it won't tell you that the feature itself should not exist. That still takes someone who understands the business, the customer, and the tradeoffs. The more I think about it, the more I feel that quantitative metrics tell us how fast we are moving. The harder question is whether are moving in the right direction.