Arun Prakash · Reel 2026

SC 09 · Notes

Notes, written after the build

14 posts · agents, benchmarks and the builds behind them

  1. New · 15 min read · WordPress · Performance · PHP · Redis · Cloudflare · Case study

    The Plugins Weren't the Problem: Speeding Up a 280,000-Story WordPress Site

    How I made a Malayalam news site on WordPress fast and cheap to run: hosting from about ₹7,500 to ₹1,500 a month, a database cleanup, a Next.js detour I shouldn't have started with, and the custom theme that took mobile Lighthouse from 48 to the 90s.
  2. New · 5 min read · AI Agents · Security · Prompt Injection · Browser Automation · Agentic Routines

    Prompt Injection and What a Browser Cache Must Refuse

    A tool that replays an agent's browser work reads websites and runs routines other people wrote, and both can contain text meant for the agent. What Agentic Routines does about it, the limits the extension holds whatever a page says, and what it cannot cover.
  3. New · 5 min read · AI Agents · Browser Automation · Reliability · Agentic Routines

    How an Agentic Routines Replay Checks the Page Before Every Step

    A replay with no model has to know when the page is no longer the one it recorded. Three checks run in order before each step, a changed step goes to the agent once, and a few edits are refused whatever the agent asks for.
  4. New · 4 min read · AI Agents · Browser Automation · MCP · Claude Code · Product

    Agentic Routines: A Cache for Your Agent's Browser Work

    Coding agents drive a browser one step at a time and pay for the whole run again tomorrow. Agentic Routines records the run, lets the agent tidy it into a routine, and replays it as one tool call with no model. Why I built it and the numbers from a small benchmark.
  5. · 8 min read · Generative UI · AI Agents · WebGL · GLSL · Claude · Prototype

    Generative UI Without HTML: Letting the Model Design in a Shader

    I wanted Claude to design anything, not fill in a component catalog. I tried a small language, then free-form HTML in a sandbox, then a single GLSL fragment shader with compile errors and a rendered frame fed back to the model. What each attempt got right, why I dropped it, and what I haven't measured.
  6. · 10 min read · Reverse Engineering · Android · Wear OS · Motorcycles · Personal project

    Twistgrip: Reverse-Engineering the RTR 310's Bluetooth Stream

    The RTR 310's cluster broadcasts far more than a speedometer shows. How I studied the official app, taught a capture script to speak its controls out loud, and mapped the bike's Bluetooth stream one field at a time.
  7. · 7 min read · Game Dev · WebGL · Rust · Product

    Impact Frame: A Browser Brawler Where the Hit Is the Whole Game

    Impact Frame is a free stickman brawler that runs in your browser. Every hit is told as loudly as it landed, on a fixed 120 Hz simulation the server can replay. How the hit, the print look, the sound, the enemies and the leaderboard were built.
  8. · 4 min read · AI Agents · Claude Code · Open Source

    Wave Planning, Packaged: The Claude Code Plugin

    Wave planning is now a Claude Code plugin. It plans a build into waves, routes each phase to the cheapest adequate model, and runs every wave in parallel git worktrees with a gate before the next one starts.
  9. · 4 min read · Go · Networking · Product

    xposed: Sharing a File Shouldn’t Need an Upload

    Right-click any file and get a public HTTPS link and a QR code. Nothing uploads: it streams from your disk, and the link expires after 24 hours. How it works, and what shipping 13 releases in 7 days taught me.
  10. · 10 min read · AI Agents · Software Engineering · Systems Design · Research

    Wave Planning: What a Wave Actually Is

    A follow-up to wave planning: what a wave actually is, why dependency alone isn't enough to make one safe, and why the real bottleneck is decomposition, not running more agents at once.
  11. · 9 min read · Performance · Go · Java · C# · Rust · Node.js · Deno · Benchmarking

    C# Was Losing to Node Because of a Logging Call

    Adding Java and C#, giving all 10 services 4 real CPU cores instead of 1, and finding a default-logging handicap that was costing C# half its throughput. Three fixes flip the whole comparison.
  12. · 6 min read · Performance · Node.js · Bun · Deno · Benchmarking

    Bun and Deno Crush Node, Until the Work Gets Real

    Benchmarking Express, Fastify, Bun, and Deno under a trivial endpoint and then a CPU-bound one, in identically resource-capped Docker containers, tells two completely different stories.
  13. · 6 min read · Market Research · Hardware · AI Agents

    We Researched the Entire PC Building Market. Here's the Verdict.

    30 buyer types, 1,000+ sources. What actually matters when building a PC in 2026: the prices, the value picks, and the mistakes almost everyone makes.
  14. · 4 min read · AI Agents · Software Engineering · Systems Design

    Wave Planning: Parallelizing Software Projects with AI Agents

    A structure for breaking software projects into steps, phases, and waves so independent work can run in parallel across AI agents instead of one task at a time.