Stop Letting LLMs Orchestrate Your AI Agents
Comparing interactive and headless AI agent orchestration on reliability, prompt control, observability, and cost, using bug reports from Claude Code's issue tracker and four multi-agent systems I built.
Hi, I'm Abdelaziz.
I'm an AI engineer at Vector Informatik in Stuttgart. I build AI coding agents and the evaluation, observability and deployment tooling around them.
Recent posts are about AI coding agents: how to orchestrate them and where they fail. The older ones are from my embedded years and cover the C build process, PWM on the ATmega32A, and test-driven development with Unity.
Latest writing
All posts
Test-Driven Development(TDD) using Unity
Learn how to perform Test-Driven Development (TDD) for embedded systems using the Unity testing framework.
12 min
Generating PWM Signal Using CTC Mode
A step-by-step guide to generating a PWM signal using CTC mode on an ATmega32A microcontroller.
3 min
The C Build Process in Detail
A deep dive into preprocessing, compilation, assembly, and linking.
5 min
Open source
orchcore Python Python library that runs AI coding agent CLIs (Claude, Codex, Gemini, Copilot, OpenCode) as subprocesses in phase pipelines, with one event model for their five stream formats. planora Python CLI and TUI where one agent writes an implementation plan, other agents audit it in parallel, and the planner revises it.