I’m Paweł, and I’m spending the next few months going deep on AI agent orchestration — orchestrators, sandboxes, protocols like MCP and ACP, observability, the whole stack. This blog is the lab notebook from that journey.
Why “Goodbye World”
“Hello world” is what you write when you start. “Goodbye world” is what an agent does after it’s spun up its own sandbox, executed a plan, and shut itself down — leaving only an artifact behind. That feedback loop, where you orchestrate something and then step out of the way, is the part of agentic systems I want to understand cold.
What’s coming
A six-post series, written in public as I learn:
- What is AI agent orchestration and why I’m diving in — mental model, MCP vs ACP
- The 2026 landscape of agent orchestration tools — 16 tools, condensed and categorised
- CrewAI vs the Anthropic SDK — two approaches to multi-agent pipelines, side-by-side code
- Designing a pipeline from brief to deploy — DAG scheduling, human gates, the Dev→Review→Test loop
- Archon — my first POC — GitHub issue → plan → implement → validate → PR
- Agent observability — what’s worth watching and what’s noise
Each post follows the same shape: what I wanted to understand, what I tried, what I learned, working code, and the next question that opened up.
Stay in the loop
The blog has an RSS feed — that’s the only follow mechanism for now. Newsletter comes later, once there’s enough material to be worth subscribing to.
Code from each post will live in a companion repo. First post drops soon.