The Primitive Manifesto

Why I am building a from-scratch systems stack in the age of AI and programmable money.

Written byazizadx
Published6/13/2026

The Primitive Manifesto

Modern software is a stack of black boxes. We’ve traded agency for convenience, hiding behind layers of abstractions we don't own and cloud providers we can’t audit.

The Abstraction Tax

Every layer between the developer and the metal is a tax on performance, privacy, and understanding. I am trading convenience back for primitives.

Core Convictions

In the age of AI and programmable money, the "Standard Stack" is a liability. I am building the Aziza Stack based on three technical pillars:

  • Latency is Intelligence: If an AI agent takes 5 seconds to "think" because of bloated middleware, the system is broken.
  • Local-First Privacy: Sending proprietary code to a third-party API is security debt. Inference belongs on the metal.
  • Ownership vs. Tenancy: If you only build on top of APIs, you are a tenant. To be an architect, you must own the primitives.

The 6-Week Roadmap

I am replacing my entire developer environment with tools I’ve built myself.

Week 1: Foundations

Building zlog — a zero-allocation, machine-native logger in Zig. Logs optimized for LLM parsing, not just human reading.

Week 2: The Core

Building koda — a high-performance text buffer engine in Rust using Piece Tables. Eliminating the "mushy" feel of Electron editors.

Week 3: AI Layer

Local inference daemons. Treating an LLM like a filesystem—a resource managed by the OS, not a cloud API.

Week 4: Programmable Money

Integrating stablecoin payment rails. Giving AI agents wallets, not just API keys.

Week 5: The Trust Layer

Implementing ZK-proofs for verifiable machine logic. Proving the agent did the work without leaking the prompt.

Week 6: The Personal OS

The final integration. A unified system where I own the editor, the brain, and the ledger.

What's Next: zlog

The first primitive off the line is zlog — a zero-allocation, structured logger written in Zig. It emits machine-native binary logs designed to be parsed by LLMs, not just read by humans. No heap, no format strings at runtime, no hidden allocator. Just raw, deterministic output at the speed of write(2).

zlog is the foundation everything else builds on — if you can't observe your system without overhead, you can't reason about it.

I stopped using tools. I started building them.