Kojo's blog

One way to slice the inference market is to cluster workloads by intelligence saturation.

I've been trying to understand how intelligence saturation affects model selection.

There's no such thing as a good AI model, only a model that is good for your workflow.

Recently, I spent several hours getting Claude Code to reverse-engineer the API for my Wyze smart bulbs.

The core premise of git is that you version-control individual characters.

A style reference in Midjourney captures the feel of an image.

When I first tried BoldVoice, it immediately captured what makes an AI-native consumer experience feel truly magical.

Conductor is a coding IDE that lets you run a bunch of coding agents in parallel.

Just a few years ago, if you wanted to ship production-grade software, you had to raise a small army.

Sonnet 4 has a relentless desire to make artifacts unprompted.

I've been using the term context gravity to describe how startups can build defensibility in the AI era.

MCPs (Model Context Protocols) are creating new user behaviors, starting with developers.

Codex and Jules are the first AI products I've felt I could trust to go off and accomplish a task async.

One of the most limiting recent UX patterns is the AI button.

All AI-assisted coding exists on a spectrum from vibe coding to precision coding.