OpenAI Introduces Workspace Agents in ChatGPT

#1 📝 OpenAI News

Introducing workspace agents in ChatGPT - Announces workspace agents in ChatGPT, a product feature that enables agents tailored to team or workspace contexts to help automate tasks and surface relevant information. The post outlines how workspace agents can integrate with existing workflows to improve productivity for organizations.

#2 📝 OpenAI News

Speeding up agentic workflows with WebSockets in the Responses API - Introduces WebSocket support in the Responses API to accelerate agentic workflows, enabling lower-latency, streamed interactions and more efficient coordination for multi-step automated processes. The post explains engineering changes that improve responsiveness for agent-based use cases.

#3 𝕏

Sam Altman hails ChatGPT’s new workspace agents—shared AI assistants built to tackle complex tasks and long-running workflows across tools and teams—and predicts most companies will want to adopt them.

#4 𝕏

Anthropic launched the Anthropic Economic Index Survey, a monthly poll of Claude users to capture qualitative data on how AI is transforming their work and driving economic impact.

#5 𝕏

Philipp Schmid announced Gemini Embedding 2 now GA: a single embedding model unifying text, images, video, audio and PDFs in one 8,192-token, 100+ language space with native audio embeddings and flexible 768/1,536/3,072-dim outputs.

#6 𝕏

Qwen’s Qwen3.6-27B (27B parameters) outperforms the much larger Qwen3.5-397B-A17B on every major coding benchmark—SWE-bench Verified (77.2 vs. 76.2), Pro (53.5 vs. 50.9), Terminal-Bench 2.0 (59.3 vs. 52.5) and SkillsBench (48.2 vs. 30.

Also covered by: @Simon Willison

#7 𝕏

Jason Zhou built a Crewlet agent that detected referral farming by spotting a spike in temp email signups and tracing them to one referral link in the database. He then flagged the fake accounts and reset their credits via a SQL script.

#8 𝕏

Cognition launched Devin for the Rivian-Volkswagen joint platform, auto-triaging tickets in Slack and generating safety-critical propulsion tests up to 15× faster than manual authoring. It scales to power 30 M vehicles and frees RV Tech’s engineers to focus on new features.

#9 📝 Claude Code Blog

Building agents that reach production systems with MCP - A blog post about techniques and considerations for building AI agents that integrate with and reach production systems using MCP. It covers product and platform context for deploying agents in real-world environments.

#10 📝 Simon Willison

Bobby Holley - A quoted excerpt from Bobby Holley (CTO, Firefox) describing how Anthropic's Claude Mythos Preview helped identify many vulnerabilities in Firefox and expressing optimism that defenders can now decisively win by focusing on security work.

#11 𝕏

There's An AI For That rolled out a research preview of ChatGPT Business, Enterprise, Edu, and Teachers plans—free until May 6, 2026. Next up: new workflow triggers, enhanced dashboards, more integrations, and workspace agents in the Codex app.

#12 𝕏

LlamaIndex 🦙 launched LiteParse, an open-source PDF parser that projects text onto a monospace grid to preserve layout structure without heavy ML models. This grid projection algorithm delivers accurate, layout-aware extraction tailored for AI agents.

#13 𝕏

NVIDIA AI adds FP8 support to NVIDIA NeMo RL, accelerating RL post-training by 1.48× on Qwen3-8B-Base. This enables faster iterations for agentic tool use and multi-step workflows.

#14 𝕏

Santiago released an open-source, local-first memory layer for Mac-based LLM agents that logs your projects, prompts, tools and on-device actions as Markdown files. You can then share this persistent memory layer with any agent.

#15 𝕏

Claude rolled out interactive charts and diagrams in Claude Cowork (now in beta on all paid plans), enabling dynamic data visualization directly within collaborative sessions.

#16 𝕏

Cursor uses @Cursor mentions in Slack to kick off tasks and stream live updates, leveraging context from threads and channels. It auto-generates pull requests ready for you to review and ship.

#17 𝕏

Garry Tan announced that GBrain now supports multiple repos per brain, paving the way to store your GStack code transcripts, plans, and Claude Code artifacts directly in GBrain.

#18 𝕏

Peter Yang: Ryan Wiggs explains why Mercury builds robust APIs before MCPs and how he ingested 5 million words from five years of PM work into Claude Code (via QMD search) to create a “second brain” that doubles his productivity.

#19 𝕏

claire vo 🖤 shares a detailed step-by-step AI workflow on the ChatPRD blog for using Claude Design and GPT Images to build landing pages, slides, and brand kits, with a companion podcast episode available on Spotify and Apple Podcasts.

Also covered by: @Claire Vo, @How I AI Podcast

#20 ▶️

📉 Turn your multimodal data into something you can actually query

Deeplearning.ai

#21 𝕏

Harrison Chase is launching on May 13th at Interrupt a new feature for LangChain that not only provides testing tools but also guides you on what to test, in what order, and when you’re done.

#22 𝕏

DeepLearning.AI launched a one-video Snowflake-backed short course on Building Multimodal Data Pipelines, covering OCR/ASR for images and audio, Vision Language Model–driven video timestamping, and multimodal RAG retrieval, taught by Gilberto Hernandez.

#23 𝕏

DeepLearning.AI: Google launched AlphaGenome, an open-weights model that deciphers non-coding DNA to pinpoint how variants shape gene behavior and disease expression.

#24 𝕏

Jason Zhou questions whether AI should adopt a single, tool-loaded agent using CLI skills and ad-hoc prompts over a multi-agent system. He argues sub-agents are only truly necessary when switching models for different tasks.

#25 𝕏

Mustafa Suleyman notes that training compute for frontier AI models has expanded 1,000,000,000,000× since 2010 and projects an additional 1,000× surge by 2028, calling it extraordinary growth.