Notes on reverse-engineering a mobile API with Claude Code in 2 hours

Ground Truth is a Human Job

"What's the Use Case?" Is a great question. Until it Isn't.

Notes on figuring out how to work with LLMs, from someone still figuring it out

Introduction

Every experience teaches you something. The only question is what.

Notes from my first job: research, teams, and the things that actually mattered

Things I Learned in 3.5 Years at my first job

A little bit of slope makes up for a lot of y-intercept

Control Wipro smart plugs programmatically using TinyTuya and the Tuya IoT platform, with an example battery-based charging automation script.

Automating Wipro Smart Plugs with TinyTuya

Set up instant, end-to-end encrypted Obsidian note syncing across all your devices using a self-hosted CouchDB server exposed via a Cloudflare Tunnel.

Self-Hosting Obsidian Live Sync with Cloudflare Tunnels

Guide for downloading HTML presentations as PDF files using DeckTape.

Converting HTML Presentations to PDF with DeckTape

How to configure a Jupyter notebook server to start automatically at system boot using systemd.

Creating a Systemd Service for Automatic Startup

I've written before about starting with the problem, not the solution in my reflections on my first job^[1] This post complicates that a little. It's about a blind spot I recently noticed.

From what I've seen in product-oriented organisations, asking "what's the use case?" is a good question. It keeps everyone grounded.

But it has a blind spot. 

Nobody asked for a model trained on the entire internet. No developer woke up thinking "I need something I can talk to in plain English that writes code, navigates complex codebases, and knows more about more things than any single human." That mental model didn't exist before the capability did. These use cases were not sitting there waiting to be found. The capability unlocked them and made them thinkable. 

Asking for the use case of general language understanding in 2018 was a little like asking for the use case of electricity before the modern world had been built around it. The question sounds responsible. Sometimes it is just too early.

Some breakthroughs create their own use cases. You don't discover those by talking to customers.

Problem-first thinking is still the right instinct when the possibility space is known. But some work expands the possibility space itself. In that phase, product thinking can become a constraint on imagination.

The mistake is not applying product thinking to research. The mistake is applying it uniformly, without asking which phase the work is in.

Product sense is about reading a room. Some rooms haven't been built yet.

[^1]: https://log.chaitanya.one/fujitsu-reflections
