Most writing about AI agents explains. This doesn't. Every note starts from something I built and put live, the take, the build, and what it taught me about how agents actually work in the real world. Short reads. Strong opinions. Working links.
The four below aren't scattered posts. They're one worked example: a single product, thought through in four phases, prove the thesis, turn it into a tool, carry it to a real case, then make it something you can feel. Read in order, it's how I approach any product, go deep, build to understand, and let the work carry the argument.
Read a company's wedge, where the value compounds, and build the argument around it, not around features.
Four working builds, each live. Real React, real serverless, API keys server-side. No mockups.
Turns a thesis into something a CEO grasps in one screen, and a claim you can feel in a live demo.
Nobody assigned this. The sequencing, the scope, and when to stop arguing were all the point.
The sequence you build in isn't logistics — it's the strategy, and most teams get it backwards. Pick your first agent by what it unlocks, not what it does.
A slide proves nothing. I turned the compounding argument into a working app that scopes a real agent from any workflow — because thinking has to survive contact with a real problem.
Asked for three agents for a real retailer, I gave the three — then argued the part nobody asked for: the order you build them in is where the strategy actually lives.
I built a governed agent you can operate — watch it answer across legacy systems, then hit a permission wall live. The claim, executed, not narrated.