AI Builders
Digest

Claude Blog: Claude Cowork and chat are now one Claude. Starting today, Claude Cowork and chat merge into one Claude. Bring a quick question or hand over a report due at noon, and Claude takes it from there even after you close your laptop. Rolling out on Pro and Max plans over the next few weeks. Claude Docs and Claude Slides are new, and Claude Design now works inside conversations. Ask for a document and you write it together; ask for a presentation and Claude drafts the slides. Edit directly, present from Claude, or export as PowerPoint or PDF. All three in beta on paid plans. Users said the frustrating part was deciding where a task belonged and context not carrying over. Claude now figures out what a task needs. Quote: "I could have Claude pull up [my legal research database], and it would pull all the cases, read them, figure out which other cases I might need, download them, and store them in a folder for my personal review." - Andrew Keller, Senior Economist.

Claude Code at Anthropic Boris Cherny notes Claude Code showed AI can do real work—developers hand over a feature and get shipped code. Cowork did the same for knowledge workers. Today chat and Cowork merge into one Claude that carries context across everything. Claude Docs, Slides, and Design live directly in every conversation—no separate tools.

Cat Wu (Claude Code + Cowork at Anthropic) announces the merge: Claude routes based on the prompt for quick answers or deeper agentic work while users stay in control. Alex Albert (Research at Anthropic) says the resulting UX feels far better than chat or Cowork alone, and the new integrations work very well. Official Claude account details continuity of Cowork work and beta availability. Every CEO Dan Shipper declares the end of the tabs era.

Thibault Sottiaux (Codex & ChatGPT at OpenAI) highlights Astra as Fast, Frontier, Efficient, and For everyone. He also quotes that sometimes physics can't be cheated. OpenAI CEO Sam Altman notes the main thing he was excited about launching this week will come next week instead, but in his view it is worth the wait.

Peter Yang (Practical AI tutorials and interviews) explains how he uses 8 AI skills to produce his entire podcast while applying his own taste and judgment. /podcast-prep researches the guest and builds an interview guide. /podcast-edit reviews the raw transcript and helps select engaging quotes and cuts. /podcast-production orchestrates 5 skills to turn the full episode into 6 different assets. He argues the latest models have not made skills redundant—they remain indispensable for following specific edit and browser-use instructions. He links a demo video of all 8 skills.

Thariq (Claude Code at Anthropic) believes Claude Managed Agents strikes the right balance—the sandbox is optional and can be spun up independently of the rest of the agent loop. He recently ported an old bash-tool-calling project to CMA and it worked well. Increasingly one can simply give Claude tools shaped the way one wants rather than adding a layer of indirection; for data storage an API to a database may be preferable to a file system. Bash is no longer all you need for reliable tool calling, though sandboxes plus bash remain good for code generation and execution.

Vercel CEO Guillermo Rauch reports extraordinary results from typesafeai. Default mode in fx is auto, with a safety reviewer analyzing every command. That reviewer runs on GPT Luna today. Jev is up to 18x faster (p95) and more accurate. It is coming to Vercel AI Gateway and likely the new default.

Box CEO Aaron Levie finds it incredibly exciting that entire universes of AI innovation still exist that were not even on most radars. Being able to process information insanely quickly at crazy low costs with high capability is huge for many enterprise tasks—data classification, routing decisions inside workflows, domain-specific judgment calls, quick safety or security assessments, and more. This model and approach could be quite cool in agentic workflows in the enterprise.

Builder Zara Zhang finds herself using Claude less because the way it speaks has become insufferable—constantly trying to showcase how smart or sophisticated it is rather than actually communicating a point.

Madhu Guru (Sr Director, AI at Meta; previously led Gemini, Veo at Google) states that safety and security are features of your AI product and model—not guardrails that need to be imposed from the outside.

Nikunj Kothari (partner at fpvventures) has been tinkering with a “Home” NousResearch agent for a few months and it has finally gotten really good. It is their first “group” bot that he and his wife interact with daily—fun learning especially when it fails. Attempts to recreate the same with Grok Bot and similar tools could not match the granularity and control (e.g., only read certain emails, convert inline attachments to structured data, or maintain logged-in browser sessions).

Y Combinator President & CEO Garry Tan notes he can use any harness he wants and it is the same personality and full memory of his personal AGI.