Reddit debate asks why coding agents still run planning and review on the same expensive model
Neat_Initiative_7780 · reddit · 2026-07-28
A Reddit post argues that agentic coding tools may be overpaying by routing every phase of a session through the same expensive model.
The author suggests splitting the workflow into stages—planning, scaffolding, implementation, review, and push—and routing each stage to the model that actually fits it, using cheaper models for boilerplate and stronger ones for architecture and review. The post asks whether anyone is already doing this well, and whether the lack of a polished UI is the main reason more tools do not ship it.
More from coding & agent
- CTI Expert turns Claude into a cyber threat intelligence analyst with 74+ commands — tom_doerr · 2026-07-28
- Agent Mini offers a 3,000-line local-first AI agent with shell, memory, and vision — Lordrovks · 2026-07-28
- In-Context launches an AI-native portfolio built for assistant browsing — modelcontextprotocol · 2026-07-28
- A Midjourney MCP server brings image generation and editing into agent apps — modelcontextprotocol · 2026-07-28
- Agensis pitches a shared workspace where AI agents reply by default and keep memory — jasonkneen · 2026-07-28
- Demo shows an Android app workflow using Antigravity, BigQuery, and Cloud Run — rseroter · 2026-07-28