Codex Boom and Agent Infrastructure Upgrades
Latent Space · rss · 2026-07-14
AINews Focus: Coding Agents and Infrastructure
This Latent Space AINews issue centers on Codex, Claude Code, agent harnesses, evaluation, and inference infrastructure.
Codex Usage Surges
The article notes:
- Codex usage grew over 10x in 6 months, hitting 7 million users
- Added 1 million users in just the last day or so
- The author speculates Codex's growth is so explosive it might be approaching or exceeding Claude Code's trajectory
Agent Infrastructure Upgrades
Highlights Prime Intellect's verifiers v1:
- Redesigned environment stack for agentic RL and evals
- Splits environments into taskset / harness / runtime
- Supports "bring your own harness" for coding and computer-use agents
- Changed rollout traces to message DAGs, preventing history duplication and reducing growth from O(n²) to O(n)
- Mentions a training config case: 100B inference model, 40-turn SWE agent task, 1000 RL steps, 6 H200 nodes, completed in two days
- vLLM support ensures rollout paths retain exact token IDs / logprobs, reducing tokenization drift between serving and training
Product and Eval Trends for Coding Agents
Key trends summarized:
- Harnesses are becoming the product itself, not just wrappers
- Coding agent competition is shifting from token pricing to cost per task
- Multiple evals and leaderboards now emphasize task success rates, controllability, and unit cost in real agent sessions
OpenAI Codex / GPT-5.6 Sol Usage and Fixes
Community feedback on OpenAI:
- Explained fixes for GPT-5.6 Sol in ChatGPT Work / Codex
- Inference optimizations boosted usage by 10%
- Reverted context limit from 372k to 272k due to billing / usage impacts
- Rolled back some experimental reasoning-effort changes
- Fixed excessive multi-agent behavior in high-intensity modes
Product Layer Expansions
Also mentioned ChatGPT restoring WhatsApp support in the EEA region and expanding to Kakao / Viber in select markets.
More from coding & agent
- Coding agents are heading toward an AI-writes, AI-reviews, human-approves workflow — aftahi_ai · 2026-07-22
- oMLX 0.5.2 adds Mac menu-bar stats, low-bit decode kernels, and faster downloads — awnihannun · 2026-07-22
- GitHub review bot hits its PR limit and forces a 39-minute cooldown — DanielLockyer · 2026-07-22
- Max reasoning effort appears to be mobile-only in Codex Remote, not desktop — GabGarrett · 2026-07-22
- A Reddit demo argues online stores should expose carts and pricing through MCP — gelembjuk · 2026-07-22
- Open-source AI SDK provider routes Vercel apps through a local Codex subscription — lgrammel · 2026-07-22