ThursdAI - The top AI news from the past week · From Weights & Biases, Join AI Evangelist Alex Volkov and a panel of experts to cover everything important that happened in the world of AI from the past week

📆 ThursdAI - Jan 15 - Agent Skills Deep Dive, GPT 5.2 Codex Builds a Browser, Claude Cowork for the Masses, and the Era of Personalized AI!

·1 hr 41 min·5 clips

Cloud co-work learned the folder, the extension, and the model choice without one line of code shown.

1. ThursdAI on January 15 focuses on agent skills, Cloud co-work, GPT 5.2 Codex, and personalized AI systems. 2. Alex Volkov hosts with Nisten Tahirai, Wolfram, Jan Peleg, Raven Wolf, and guest Eleanor Berger, who joins later for the agent-skills deep dive. 3. The episode asks what happens when coding agents become generalized agents that can help developers and non-developers do real work. 4. Baichuan M3 is presented as a 235 billion parameter medical LLM fine-tuned from Qwen 3 and claimed to beat GPT 5.2 on healthcare benchmarks. 5. The panel also covers LongCat FlashThinking 2601, a 560 billion parameter MoE with 27 billion active parameters and MIT licensing. 6. Anthropic's Cloud co-work is described as a Mac-only research preview for Max subscribers that can read, write, and create files from folder access. 7. OpenAI's GPT 5.2 Codex arrives in the API after being limited to the Codex app, and the hosts say Cursor uses it for agent harnesses. 8. The show also mentions Cursor building a browser from scratch with more than 3 million lines of code using GPT 5.2 Codex. 9. Google Gemini personalization is described as reasoning across Gmail, YouTube, photos, and search with explicit opt-in controls for US Pro and Ultra users. 10. The drama segment covers Anthropic blocking XAI from using Cloud Code and then telling Cursor it could not serve Opus to XAI through that route. 11. The panel says OpenAI and GitHub stepped in so OpenCode could use Codex subscriptions and Copilot after Anthropic's restrictions. 12. Alex says Cloud co-work was written with Cloud Code and that Felix Reisberg said Anthropic built it in a week and a half. 13. The live demo shows Cloud co-work opening a folder, inferring supported models, and handling a message that says it is unable to verify a fetch as safe. 14. Wolfram notes that code agents are now generalized agents, and Ryan compares the shift to why Ralph took off for task completion. 15. Alex argues that non-developers should stop avoiding command-line tools and can learn by having an agent walk through screenshots and setup steps. 16. The healthcare section ties OpenAI's Torch acquisition to Anthropic's Cloud for Healthcare and Life Sciences connectors and HIPAA-ready products. 17. The conversation stays fast and conversational, with repeated live reactions, product demos, and short debates about pricing, access, and usefulness. 18. The episode moves like a news roundup rather than a single interview, but Eleanor Berger's later segment is positioned as the technical anchor for skills. 19. People following AI coding tools, model launches, and agent workflows would likely want this episode. 20. Listeners wanting a single-topic interview or a slow, polished narrative may skip it.

As heard by us

Agent skills get a practical, demo-led treatment, with connectors and open standards giving the discussion real-world traction.

Agent skills emerge as the episode's central thread, and the discussion treats them as a working shift rather than a slogan. Alex Volkov keeps the pace up, while Eleanor Berger grounds the conversation in actual use, which gives the open-standard angle more substance than…

Read the full review in PlayNext →

Why you'd press play

You want a practical read on agent skills, with less hype and more actual workflow.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from ThursdAI - The top AI news from the past week