Claude Code v2.1.286 (released September 30, 2026) changes behavior so that when the default model is rejected by the API, it retries once with the previous model of the same tier. It closes the "every-turn failure" in unattended runs, and also refines the approval counter display and how remote-con
Claude Code v2.1.285 adds allowedProviders to limit which API endpoints may be used, CLAUDE_CODE_DISABLE_WEB_FETCH to turn off WebFetch, and a sandbox that lower-level settings can't loosen. In the age of unattended operation, the controls for administrators to tighten the agent's "outbound mouth" a
At DevDay 2026, Codex is reworked into an always-on development platform that spans devices. It adds voice control and parallel work via /agents, automated code review, and Codex Security Cloud, while GPT-6.1 Sol cuts unit price to about a fifth and Ultrafast reaches around 300 tokens per second.
OpenAI's Codex made git worktree sessions the default in 0.156, released September 23. Disposable conversations become durable workshops you can isolate, resume, and review, leaning the tool toward a parallelism-first workflow. A /usage dashboard and a full-screen TUI were tidied up at the same time
Claude Code's default model was updated twice in one week. v2.1.284 makes Sonnet 5.5 the standard, while mechanisms like deniedModels and a dollar-denominated /usage put 'which models, and up to how much' in administrators' hands.
Codex v0.152.0 on August 31 and its next-day fix release added explicit ceilings on MCP tool output volume and execution time, and switched the planning tool off by default. Here's a rundown of the changes that matter for long-running unattended and semi-autonomous agent operation.
Claude Code v2.1.257, released September 1, 2026, swaps its default model to Fable 5.1 with its one-million-token context. It also adds guardrails to auto mode that stop credential retrieval and out-of-scope reads from slipping through. Here's a rundown of the changes that matter to developers.
Gambit Security and CloudSEK report that the ransomware group Aurora abused Cursor's AI agent for real intrusion work. Posing the tasks as an "authorized exercise" to slip past the safeguards, they had it handle reconnaissance and privilege takeover on the back of stolen credentials—a warning that a
McKinsey's annual survey found that about 30% of respondents passed on buying software because they could build it in-house with coding agents. We unpack the procurement shift from buying to building — and the current reality that productivity is up while profits stay flat.
On August 31, OpenAI is ending availability of GPT-5.4 and 5.4 mini in Codex. But it only affects ChatGPT-login users—with API key authentication or via the API, they remain. The replacements are gpt-5.6-terra and gpt-5.6-luna. Here's a rundown of the settings to review and the three paths you can t
Two weeks after SpaceX acquired Cursor, OpenAI served notice that it would cut off access to its GPT models, with a planned shutoff on November 12. We break down — from a developer's perspective — the features a personal API key can't cover and Anthropic's swift move to expand its own capacity.
Cursor's cloud agent can now start building without a GitHub connection. Give instructions from scratch → preview instantly in the browser → save to Origin once you're satisfied, and even get a public URL via Vercel. The entry point to development flips from "container first" to "build first."
