Nine coding harnesses vs. your laptop

6.9 relevance

AI coding harnesses compared, highly actionable for developers

General nasutton.notion.site

Answer the board on AI readiness

Arrive with evidence, not opinion. See how your foundation benchmarks against your peers, get a clear read, and leave with a ranked place to start.

Why It Matters

As AI agents move to local-first architectures for privacy and latency, this exposes a critical gap: current agent orchestration tools are optimized for server GPUs, not Apple Silicon or edge devices, directly impacting developer experience for platform engineers building offline-capable workflows.

Key Takeaways

Audit your coding harness for system prompt bloat and side-request patterns before committing to local inference—prefill time can exceed 50% of your context window on laptop hardware.

Summary

A developer benchmarked nine coding AI harnesses against a local Qwen 3.8 27B model on an M4 MacBook Pro, finding three systemic problems: oversized system prompts and tool schemas (2,008 tokens for pi vs 18,046 for Opencode), reduced effective context windows (Opencode leaves only 44% of 32K tokens), and excessive side requests that overload the local GPU. The test used auto-approve Exercism tasks with a shared llama.cpp server, revealing local-local design flaws that make many harnesses unresponsive compared to data-center usage.