available · open to interesting work | loc bengaluru, ist | local --:-- srijanshukla18@gmail.com
[category]/writing/ai

ai.

Filtered public notes from the archive.

[01]

entries

08
2026-05-17 A quantified-self setup with one AI agent and two readouts I have a single WhatsApp thread with Hermes Agent capturing what I do every 15 minutes and what I eat. Two skills own the capture loops, write the JSON, and render two public readouts on this site. Notes on why this works now and why I'm tracking exactly two things. 2026-04-14 Model overthinking is a control problem High thinking budgets can reduce control, especially when stale context and autonomous tool use mix together. 2026-04-14 Skill Sync Cron The technical problem was simple: one skills repo, two laptops, five agent runtimes. The real problem was that every unstated decision gave the agent room to optimize for control instead of maintainability. 2026-04-12 Sequented Gated Prompting with a Pi extension How I wanted one cleanup workflow after each feature, tried building a Pi extension for it, and learned that sequential gated prompting is more reliable than bundling everything into one turn. 2026-04-12 Frontiers in Agentic Design A systems view of modern AI agents: models are only one layer; the real frontier is the harness, context, control loop, memory system, safety boundary, and orchestration around the model. 2026-04-12 What Happened When I Tried to Coordinate Two AI Agents Over NFS A shared mount looked like the cleanest way to make two agent instances talk across machines. A 30 minute directory cache made the whole thing unreliable, and Git ended up being the simpler bus. 2026-04-12 Vibe Coded a `watch` Command for Autonomous LLMs, and It Became My Debugging Loop Once I had LLMs editing code autonomously, I wanted a better way to watch a directory breathe in real time. Existing tools covered file events or diffs or a TUI, but not all three together. 2026-02-22 TAO - The Prompting Pattern That Makes AI Agents Super Effective Most multi-agent failures are not intelligence failures. They are ownership failures. Territorial Agent Orchestration fixes that by making the coordinator own shared work and parallelizing only isolated slices.