---
title: "Sahil Kathpal (@sahil-kat) | Hashnode"
description: "Software Engineer building Grass — mobile access for Claude Code and AI coding agents. Read the latest articles by Sahil Kathpal on Hashnode."
url: "https://hashnode.com/@sahil-kat"
---

@sahil-kat

# Sahil Kathpal

@sahil-kat · Mumbai, India · Joined April 2026

Software Engineer building Grass — mobile access for Claude Code and AI coding agents

### [Inside-the-Loop vs. Outside-the-Loop: Evaluating Agent Architectures](https://codeongrass.hashnode.dev/inside-the-loop-vs-outside-the-loop-evaluating-agent-architectures?utm_source=hashnode&utm_medium=feed)

Inside-the-loop and outside-the-loop are the two architectural modes that determine whether your AI coding agent feels controllable or like a coin flip. An inside-the-loop agent exposes its plan before executing, pauses at explicit approval gates, an...

13 min read · [codeongrass.hashnode.dev](https://codeongrass.hashnode.dev/inside-the-loop-vs-outside-the-loop-evaluating-agent-architectures?utm_source=hashnode&utm_medium=feed)

[May 6](https://hashnode.com/posts/inside-the-loop-vs-outside-the-loop-evaluating-agent-architectures/69fb7aabae4345219b34eec3)

### [Cut Claude Code Token Usage 98% with Purpose-Built MCPs](https://codeongrass.hashnode.dev/cut-claude-code-token-usage-98-with-purpose-built-mcps?utm_source=hashnode&utm_medium=feed)

Running Claude Code against a large codebase or a corpus of financial documents will drain your token budget fast — not because the tasks are conceptually hard, but because Claude's default behavior is to read entire files into context. Two recently ...

12 min read · [codeongrass.hashnode.dev](https://codeongrass.hashnode.dev/cut-claude-code-token-usage-98-with-purpose-built-mcps?utm_source=hashnode&utm_medium=feed)

[May 6](https://hashnode.com/posts/cut-claude-code-token-usage-98-with-purpose-built-mcps/69fb7aa9e541678fa29a1719)

### [Catch Agent Mistakes Before They Execute: Agent Verifier + Conduct](https://codeongrass.hashnode.dev/catch-agent-mistakes-before-they-execute-agent-verifier-conduct?utm_source=hashnode&utm_medium=feed)

By the time a manual code review catches a hardcoded API key or a retry loop with no exit condition, an AI coding agent has already written it to disk — and possibly already run it. Two freshly shipped open-source tools — Agent Verifier and Conduct —...

14 min read · [codeongrass.hashnode.dev](https://codeongrass.hashnode.dev/catch-agent-mistakes-before-they-execute-agent-verifier-conduct?utm_source=hashnode&utm_medium=feed)

[May 5](https://hashnode.com/posts/catch-agent-mistakes-before-they-execute-agent-verifier-conduct/69fa292b2cc34a7099ce089c)

### [When Should Your Agent Ask Before Acting? A 3-Tier Risk Framework](https://codeongrass.hashnode.dev/when-should-your-agent-ask-before-acting-a-3-tier-risk-framework?utm_source=hashnode&utm_medium=feed)

Every developer running AI coding agents eventually hits the same wall: the agent does something destructive without asking, or it interrupts flow by asking for approval on every file read. The debate plays out publicly as a Codex vs. Claude Code arg...

12 min read · [codeongrass.hashnode.dev](https://codeongrass.hashnode.dev/when-should-your-agent-ask-before-acting-a-3-tier-risk-framework?utm_source=hashnode&utm_medium=feed)

[May 5](https://hashnode.com/posts/when-should-your-agent-ask-before-acting-a-3-tier-risk-framework/69fa292923d8e9a122973f0c)

### [AI Agent Disaster Postmortems: The 3 Structural Guardrails](https://codeongrass.hashnode.dev/ai-agent-disaster-postmortems-the-3-structural-guardrails?utm_source=hashnode&utm_medium=feed)

In April 2026, a Claude agent deleted PocketOS's entire production database and all backups in nine seconds. No confirmation prompt. No approval checkpoint. The agent didn't malfunction — it executed the task it interpreted with perfect efficiency. A...

13 min read · [codeongrass.hashnode.dev](https://codeongrass.hashnode.dev/ai-agent-disaster-postmortems-the-3-structural-guardrails?utm_source=hashnode&utm_medium=feed)

[May 4](https://hashnode.com/posts/ai-agent-disaster-postmortems-the-3-structural-guardrails/69f8d7accd438d093bc3f764)
