Applied AI & Engineering Leadership

Head of AI at ArmorCode
I build AI agents that do real work.
I turn AI capabilities into useful products and build the teams that ship them.
Get In Touch Read the Writing
Let's talk → Read the writing
Scroll
About

Builder

10
Years
3
Teams
1
Exit
3,200+
GitHub Stars

My focus is applied AI and AI-native engineering, grounded in a decade of security engineering.

At ArmorCode, I built the AI team zero-to-one and created Anya, an autonomous AppSec agent.

I build in public through open source and write about what I learn from building and using AI systems.

Open Source

cache-tax
30+ ★ Creator
See the estimated cost before a cold Claude Code send, and keep selected sessions warm during breaks.
itr-wala
870+ ★ Creator
File Indian income tax returns from the terminal. The model reads documents; tested Python computes every rupee.
claude-code-hooks
520+ ★ Creator
Installable Claude Code plugins for safety, cost tracking, and observability. Each hook also works as a standalone script.
reddit-mcp-buddy
840+ ★ Creator
Clean, LLM-optimized Reddit MCP server. Browse posts, search content, analyze users.
awesome-claude-code-mods
50+ ★ Creator
A directory of Claude Code mods with scanner results showing what each mod can access.
awesome-claude-skills
520+ ★ Creator
Curated collection of 50+ verified skills for Claude Code, Claude.ai, and API.
SGLang
36k+ ★ Contributor
High-performance serving framework for large language and multimodal models.
vLLM
93k+ ★ Contributor
High-throughput LLM inference engine powering production AI at scale.

Experience

2023 - Present
Head of AI
ArmorCode · Series B
Built the AI team zero-to-one. Creator of Anya, an autonomous AppSec agent. Full career details in my resume.
2021 - 2023
Engineering Manager
Urban Company · $2.8B Unicorn
Built 12-engineer security organization from scratch. Drove platform initiatives achieving 99.999% uptime. Implemented GDPR-compliant privacy architecture for IPO readiness.
2018 - 2021
Founding Engineer → EM
AvidSecure → Sophos
Employee #1. Built MVP leading to acquisition in 6 months. Scaled platform from 10 to 10K customers and team from 0 to 25 engineers.
2016 - 2018
Sr. Software Engineer
NTRO
Built secure software systems for critical infrastructure protection.

Writing

Claude Code Hooks Just Moved Inside the Process
September 20, 2026
A Mod is a Claude Code hook that moved inside the process. Three diagrams, one twelve-line mod, and by the end you can read what any mod can touch before you install it.
Claude Code Rewrote 314k Tokens for a Two-Line Recap
September 19, 2026
A cold Claude Code cache turned a two-line recap into a $6.28 write at API list rates. One tool shows the price at send time or keeps selected sessions warm.
Claude Code Plugins Can Now Run Inside Claude Code. 14 of 31 Can Run Commands on Your Machine.
September 15, 2026
Claude Code has a new kind of plugin that runs TypeScript inside Claude Code. I scanned 31 with Claude Code's own validator. 14 can run commands on your machine, and 4 can reach the network.
Switching Effort Nuked My Prompt Cache. The Next Day It Didn't.
September 4, 2026
An effort switch used to re-read your whole conversation. I caught the warning on the 3rd on 2.1.259, and 2.1.260 killed it on the 4th. Paper trail inside.
The Model Is Banned From Math
August 13, 2026
An ITR-filing skill went viral off one stranger's tweet. The design rule that earned it: the LLM reads your documents, tested Python computes every rupee. Including the bug the fuzzer could never find.
How to Write Loops with Claude Code
July 20, 2026
I read 182 replies to “can you share loops you regularly use?” One linked a loop you could actually run. Seven you can copy, with the checks and guardrails attached.
Anatomy of a $5 “Good Morning”
July 8, 2026
My most expensive API requests ever are an acknowledgment, a menu choice, and a parade of good mornings. Where agent tokens actually go, receipt by receipt.
Prompt Caching Doesn't Cache Prompts
July 7, 2026
Prompt caching is the single biggest cost lever for LLM agents, but it doesn't cache your prompt. What it actually caches, and why it matters.
You're Slower Than You Think: The 4 Levers of AI-Era Speed
May 19, 2026
AI made optimizing your own workflow nearly free. Four levers (Connect, Encode, Teach, Parallelize), each with a tell you can catch yourself in.
The 2-Minute Claude Code Upgrade You're Probably Missing: LSP
February 28, 2026
Every Claude Code user is running without LSP. That means 30-60s grep searches instead of 50ms precise answers.
Claude Code's Most Underrated Feature: Hooks
January 25, 2026
How Claude Code hooks log events, block dangerous commands, and protect secrets, with ready-to-use code.
All posts →

Let's Build

VCs

I translate AI hype into ROI reality

Due Diligence Calls →

Founders

I invest time in breakthrough ideas

Technical Advisory →

Engineers

I build teams that change trajectories

Open to Coffee Chats →

Speaking

Black Hat · DEF CON · OWASP · c0c0n

Invite to Speak →
karan@karanbansal.in LinkedIn → X → GitHub (3,200+ ★) →