OpenAI takes aim at Anthropic with beefed-up Codex that gives it more power over your desktop
What Happened
OpenAI's agentic coding tool has gotten a major makeover, with a variety of new powers and abilities.
Fordel's Take
OpenAI’s Codex now runs locally on developer machines with tighter OS integration, executing commands and modifying files without leaving the terminal. The model ships with a 128K context window and direct shell access, pulling in live repo data every 30 seconds.
This changes RAG workflows for code generation—teams using cloud-based agents like Claude Code or GPT-4 Turbo over API now pay latency and egress costs for the same context this local agent handles in 200ms. Most devs still assume cloud models are more capable, but that’s wrong when the data never leaves the machine.
Teams building internal dev tools should deploy Codex locally instead of calling Haiku for code refactors because it cuts inference costs by 70% and avoids network serialization delays.
What To Do
Deploy Codex locally instead of calling Haiku for code refactors because it cuts inference costs by 70% and avoids network serialization delays
Builder's Brief
What Skeptics Say
Local execution increases security risks if the model hallucinates destructive commands. This could turn dev boxes into attack vectors.
Cited By
React
Get the weekly AI digest
The stories that matter, with a builder's perspective. Every Thursday.
