The Illusion of Control: Building a 100+ Agent Swarm in Web3 (Part 3)
Part 3 of my HackerNoon series on building a 100+ agent swarm just went live. Part 1 covered context management. Part 2 covered skill engineering. This one is about why I stopped refining prompts and started building enforcement instead.
If you only improve the prompt, you get higher quality mistakes.
Prompts control objectives. They don’t control boundaries. I wrote “use theme tokens, don’t hardcode colors” in my agent instructions. Clear, direct, explicit. The agents did it anyway. So I built ESLint plugins that make the violations impossible to commit.
The article walks through the full progression:
- Explicit prohibitions — engineering principles your agents lack, written down because the agent needs them to be
- Rule graduation — encoding those principles in tooling instead of instructions
- Deterministic enforcement — ESLint plugins that reject bad code before it commits
The jump from vibe coding to agentic engineering is not a tools upgrade. It is a behavior change. And that’s why most people stall at vibe coding. Tools are easy to adopt. Behaviors are hard to change.