How the magpie developer sees it
Why I crack things open
- 01
The order of making has flipped
To build something new, you used to start with the parts: learn the language, understand the framework, and only then write the first line. Now the reverse is possible. Build what you want first, then take it apart to understand it. I think this reversal has changed how fast a person can grow.
- 02
The cost of taking things apart has collapsed
If you wondered how a car runs, you could always take one apart. The problem was the time spent hunting down material for every single part. Now, wherever you get stuck, one request — “make this simpler” — is enough. Understanding has never been this cheap.
- 03
Handling the tools is becoming the model’s job
When coding agents arrived, I used them hard. At first I focused on orchestration and the craft of skills, but as the models improved they took over most of that on their own. The edge that remains, I decided, is not the hand on the controls but the speed at which I grow my own knowledge and perspective with AI.
- 04
Building with understanding, not just building fast
If I stack up what an agent builds in a day without understanding it, it is not mine. So even when implementing, I open each change step by step and turn it into knowledge. The model raises the speed of building; the depth of understanding is mine to build.
- 05
Choosing what to crack open is the skill
Anything can be taken apart now. But everyone’s hours pass at the same rate. Not cracking things open at random, but choosing what to crack open — that choice, I think, is the skill of this era.
- 06
The more you have cracked open, the better you delegate
A model picks its answers probabilistically from an ocean of general knowledge, so a single word in a prompt can decide the quality of the result. The vocabulary you build by cracking a domain open makes better questions, and better questions bring better results. Cracking open and delegating are one and the same.
So, here
I crack things open while building them. Right now I am writing a series on building an enterprise AI bot from scratch, recording the decisions I run into along the way.
See what I am building