Codex GitHub repo has only a few open issues labeled macOS, but hundreds for Windows. Windows has the extra challenge of supporting both native Windows and WSL workflows, and that complexity shows up in the issue tracker. I think it’s time for me to pause on WSL mode in Codex Desktop and go back to running Codex CLI directly inside WSL until things stabilize.

Compared two implementation plans for the same feature today: • Plan 1: Composer 2.5 in Cursor • Plan 2: Gemini 3.5 Flash in Google Antigravity 2.0 • Review: GPT-5.5 Both got the product direction right. But Composer’s plan was stronger on architecture: clearer file boundaries, maintainability, responsive behavior, future extensibility, and verification. Gemini’s plan felt more like a UI sketch, while Composer produced a more complete implementation plan. Composer 2.5 is not considered a frontier/SOTA model, yet it still came out ahead in this comparison. Makes me think the harness, context management, and workflow around the model matter as much as the model itself.

🚀 Introducing Gemini-SQL2, our breakthrough text-to-SQL capability powered by Gemini 3.1 Pro! We've achieved state-of-the-art results on the highly competitive BIRD benchmark, translating natural language into execution-ready SQL queries. 🧵👇

Communication is now the most important skill for developers

GitHub Copilot app is getting surprisingly good. The My Work section works like a personal task manager for developers. It keeps assigned issues, PR review requests, mentions, and other pending work in one place so nothing slips through the cracks. Simple, but very useful.

Great software always took shape in conversation, not the commit. With agents, the conversation that generates the code is becoming the true source of our software. And Git can't keep up. So we built something that can. Meet DeltaDB: zed.dev/blog/introducing-del…

Antigravity 2.0, both the desktop app and CLI, still feel quite far behind the competition. What surprises me more is not the current state, but the pace of updates. In a market moving this fast, the gap isn’t staying constant, it keeps widening. Makes me wonder: is there a bigger plan the team is working toward that isn’t visible yet?

Not feeling much excitement for Fable 5. I have Codex Pro, and GPT-5.5 is excellent for complex tasks, but Composer 2.5 inside Cursor handles 95% of my day-to-day work more efficiently. At this point, reliability, speed, and workflow matter more to me than benchmark gains.

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Its capabilities exceed those of any model we’ve ever made generally available.

The official Todoist ChatGPT app is out! Use it to plan your day, add and manage tasks, organize projects, etc. We’d love to hear what you think and how we can make it better.

WSL Codex tip: If your normal WSL terminal finds node, pnpm, uv, etc. correctly, but Codex/non-interactive shells fall back to Windows PATH shims, check your shell startup files. For zsh, .zshrc is interactive-only. Add minimal PATH setup to ~/.zshenv: export PATH="$HOME/.local/bin:$PATH" export NVM_DIR="$HOME/.nvm" export PATH="$NVM_DIR/versions/node/v22.20.0/bin:$PATH" Keep it lightweight, but this ensures non-interactive WSL shells use native Linux tools before /mnt/c/....

If you're using Codex Desktop with WSL, project automations can still be hit-or-miss for some users. The automation run gets created, but the agent turn doesn't always start. One workaround that's been working well: thread heartbeat automations. They wake up the same chat on a schedule, keep the conversation context intact, and are great for recurring "check / review / monitor / notify" workflows that don't require access to a local project worktree. Not a replacement for project automations, but a useful pattern until WSL support becomes more reliable.

Now that the Codex 5x $100 plan quota is lower and GPT-5.3-Codex-Spark has its own separate quota, what are the best use cases you've found for GPT-5.3-Codex-Spark? How are you deciding when to use it instead of other GPT-5 models?

I’ve been using Claude Code and Codex for the last year. Recently, I tried Antigravity 2.0, but it didn’t leave me with a great feeling. After trying Cursor’s new Agents Window, Antigravity 2.0 feels more like an alpha release than even a beta. It has potential, but right now it feels immature and somewhat rushed. I wonder if the timeline was accelerated because of Google I/O.

Moved from Fedora back to Windows because of Codex Desktop. Thankfully WSL exists, so I can continue pretending I’m still a Linux user.

Downloaded Cursor today and have only used Auto mode so far. Didn’t expect it to be this good. Fast, responsive, and the output quality is impressive. Codex is still my main driver, but Cursor is off to a very strong start.

One of the most underrated things about agent skills is portability. Today I tried Cursor again after spending a long time using Codex and Claude. The same skills worked with virtually identical output. No rewrites, no agent-specific tweaks. The skill becomes the reusable unit, not the agent. That's a pretty powerful abstraction.

Spent some time with the Codex Python SDK today creating a GitHub PR review TUI. Fun project, and interesting to note that the output remains completely consistent between Codex Desktop and the SDK.