@ClaudeDevs Curious to try it out, it's helpful in some cases but IMO the real quality unlock is getting another model's opinion on something, from review to debate, that's what PaF does since Feb :)
https://t.co/EWNCqO6XGz
We want to welcome a new member of The PHP Foundation Governing Board: Anne McCarthy from Automattic. Anne brings a balanced mix of thoughtful empathy and deep technical knowledge and will bring tremendous insight to the Board! https://t.co/figKjMboV1 #php#automattic#wordpress
In April, I shared phone-a-friend publicly for the first time.
Codex has come a long way since then, so I brought it in: you can now run it inside Codex and OpenCode, not just Claude Code.
Plus session persistence, so context carries across calls.
v4 is out 🚢
https://t.co/LvWrF0cJUp
Love this example from @dhh on what it means that you can tackle more ambitious work with AI agents. At 37Signals, a dev took on to make P1 (the fastest 1% of requests) ~10x faster!
@bcherny I also love /insights as a way to improve all my local skills, agents, and workflows.
I run every other week and have been getting great results.
AI is moving insanely fast.
Took 2 weeks off, and so much has happened.
OpenAI's push for cross-tool workflows with the Codex plugin for Claude Code is a big signal.
I’ve been building something similar with phone-a-friend:
https://t.co/qO7RbKQkPJ
It connects Codex, Claude, Gemini, and Ollama for reviews, handoffs, and collaborative workflows, with agent sessions and a dashboard.
Feels like the real shift is not just better models, but composable workflows between them.
@levelsio and even get them to "talk".
That's kind of how I work locally: I delegate certain Codex and Gemini tasks from a Claude session and iterate until certain criteria are met!
I have it on my personal GH, check it out, and it might give you ideas!
@levelsio As someone mentioned earlier, you could use claude -p, if you are running this inside your server. Teams of Claude Code sessions + I would add persistent sessions and...
@levelsio ... if you assign different roles/SKILLS for different "Claudes", like:
- spawn (first message): claude -p --session-id <uuid>
- resume (subsequent turns): claude -p -r <uuid>
If you can do all of that and even delegate tasks to Codex from a Claude "orchestrator".
@levelsio I do like the idea; however, for an automated workflow like this, you're trusting a single model's judgment.
I would definitely put Codex in the mix; maybe Claude does most of the lifting while Codex handles the code reviews.
Automattic turns 20 this month, and we’re celebrating with a look at the stats behind the story. 
First up: @WordPress powers 43% of all websites across the internet. That’s a number worth celebrating. And we’re just getting started.
FrankenPHP is now officially supported by The PHP Foundation
The project repository will be moved to the @official_php organisation on GitHub.
https://t.co/gKiMwRUZhe
Just dropped: The new, free AI website builder on https://t.co/eRvNKWaolr!
Tell AI what you need + get a fully built website in minutes:
🪄 No coding. No drag-and-drop. No stress.
🪄 AI handles the layout, images, and text.
🪄 You can tweak anything—just ask AI! 🧵
In 2025, @Automattic, a founding member of The PHP Foundation, is increasing its contribution to $250,000—a 2.5x increase!
We are very grateful for their ongoing support and firm commitment to the PHP community.
Thank you, @Automattic and @WordPress!