LOCAS: An AI-Assisted Coding Workflow to Unite Frontier AI Efficiency, Local Compute, and Human Craftsmanship

Current AI-assisted coding often relies on large cloud-based models for end-to-end planning and implementation. This usually requires constant internet connectivity and can offer limited granular oversight, which sometimes introduces architectural blind spots. This paper proposes LOCAS (Local-Centric Asynchronous AI Workflow with Frontier Planner and Small Executor) as a workflow to unite frontier AI efficiency, local compute, and human craftsmanship. Drawing on prior work in tiered AI orchestration, LOCAS is designed to complement existing agentic and traditional coding workflows by introducing a cloud-to-local asynchronous handoff. In this process, a centralized "Big Planner" generates a high-fidelity blueprint that local Small Language Models (SLMs) implement independently. This handoff allows developers to leverage high-capacity AI reasoning while performing most implementation tasks using models running on local hardware-an approach that remains productive even in low-connectivity environments or under strict AI budgets. By encouraging a structured, stepwise, local-centric process, LOCAS keeps developers intimately involved in the development process. This helps mitigate the risks of difficult-to-trace bugs in large AI-generated code, preserves architectural clarity, and ensures better accountability. Ultimately, it enables a more sustainable, cost-effective, and resilient execution using a developer's own machine compared to traditional cloud-tethered workflows.