Loop Engineering Is a Pattern, Not a Feature
13 hours ago
- AI infrastructure in 2026 is driven by cycles where useful techniques become marketed as features, often as walled-garden products.
- Loop engineering is a basic programming construct (a 'while' statement), not a product category or feature to be purchased.
- The core of loop engineering is the turn orchestrator, which manages loop turns and exposes lifecycle hooks.
- Fully agentic orchestration using LLMs for every turn is costly, slow, inconsistent, and often unnecessary.
- Functions are superior to LLMs for most turn orchestration tasks because they are consistent, fast, cheap, and accurate.
- The 'iii' harness uses Workers, Functions, and Triggers to orchestrate loops deterministically, avoiding unnecessary LLM calls.
- An example workflow scans a GitHub repo for vulnerabilities using iii triggers and functions for reliable, parallel, and efficient completion detection.
- Workflows should start with exploratory LLM sessions, then move logic into deterministic code where possible to improve cost, latency, and reliability.
- Mature systems use LLMs only where genuine judgment is needed, replacing them with code or smaller ML models elsewhere.
- iii supports this progression because its primitives make LLM-based and code-based components interchangeable and easy to swap.
- iii is open source and designed for observable, testable, and independently swappable components in loop engineering.