OpenAI Launches Codex App
Today's top 10 insights for PM Builders, ranked by relevance from Blogs, X, YouTube, and LinkedIn.
OpenAI Launches Codex App
đź“ť OpenAI News
Introducing the Codex app - OpenAI has launched the Codex app, enhancing user interaction with AI.
đť•Ź claire vo đź–¤ @clairevo
Claire overhauled Maplewood’s architecture by migrating to Inngest workflows and persisting stories/actions in NeonDB, added infinite scroll for event feeds, and squashed an auto-scroll bug.
đź“ť Doug Turnbull
Check twice, cut once with LLM search relevance eval - Highlights the importance of checking both directions in LLM pairwise evaluation of search relevance.
▶️
How To Train Your AI Agent's Skills (Claude Code / OpenClaw)
•
How To Train Your AI Agent's Skills (Claude Code / OpenClaw)
All About AI • February 01, 2026
All About AI demonstrates how to iteratively create and refine a LinkedIn skill.md for a Claude Code / OpenClaw agent—teaching it to draft posts, search profiles, and send direct messages—reducing task execution from six minutes to 40 seconds.
Key Takeaways:
- Created a LinkedIn skill.md file that uses Chrome CDP and JavaScript injections to compose draft posts, cutting posting time from six minutes to 40 seconds.
- Trained the agent to search LinkedIn for Antropic employees, extracting profiles such as Tyler Quinn, Software Engineer at Entropic, along with their roles and locations.
- Extended the LinkedIn skill to send direct messages by identifying and clicking the send button, enabling autonomous greetings and follow-up messages.
▶️
How this PM uses MCPs to automate his meeting prep, CRM updates, and customer feedback synthesis
•
How this PM uses MCPs to automate his meeting prep, CRM updates, and customer feedback synthesis
How I AI Podcast • February 02, 2026
Reid Robinson shows how he uses Zapier MCPs to give Claude access to apps—building custom tool collections and Claude Projects—to automate meeting preparation, CRM record updates in KOD and HubSpot, and ongoing FAQ generation from support tickets.
Key Takeaways:
- Zapier’s MCP server supports over 8,000 apps and lets you assemble custom tool sets (e.g., Slack, Evernote, Glean, Koda, Google Calendar) with scoped permissions into a single connector URL for AI clients like Claude, ChatGPT, or Cursor.
- Using Claude Projects, Robinson provides step-by-step instructions on tool-calling order and CRM field mappings, enabling automated sequential lookups and record creation for daily planning and post-meeting note entry in KOD and HubSpot.
- He built a Zap that triggers on closed support tickets or chatbot transcripts, uses AI to extract core FAQs and solutions, and then allows human review before updating the internal knowledge base used by customer-facing chatbots.
in Carl Vellotti
Carl Vellotti breaks down his team’s Claude Code workflow: they spin up 3–5 parallel terminal sessions (tied to quick-jump shortcuts) to tackle different tasks at once, always kick off big requests in “Plan Mode,” and maintain a living CLAUDE.md file to encode preferences and correct mistakes for ever-smarter prompts.
đť•Ź Julien Chaumond @julien_c
Julien demoed how to extend an AI agent to call the Hugging Face CLI as a built-in skill, using the new `hf skills add --opencode --claude` command to inject up-to-date model knowledge directly into its context. This technique makes it trivial to integrate real-time HF CLI interactions into your agent workflows.
▶️
Screensharing Kevin Rose's AI Workflow/New App
•
Screensharing Kevin Rose's AI Workflow/New App
Greg Isenberg • February 02, 2026
Kevin Rose screenshares his AI workflow by demoing “Nylon,” a personal Techmeme-style news engine that ingests 63 RSS and social feeds and enriches stories via framely, firecrawl, and Gemini. He explains how he generates TLDRs and vector embeddings with GPT5 mini, clusters content in Postgres, and manages durable, retryable tasks using trigger.dev orchestrators.
Key Takeaways:
- Nylon pulls in 63 sources (RSS and social) into a Postgres database, then uses framely, firecrawl, and Gemini to select the richest summary and main content per article.
- Each article is processed by GPT5 mini to produce a human-readable TLDR, key bullet points, spicier title variants, and OpenAI vector embeddings stored via pgvector for nuanced clustering.
- The end-to-end enrichment and clustering pipeline runs on trigger.dev TypeScript functions with automatic retries and monitoring, achieving 99.8% enrichment of 2,288 stories in 24 hours at under $100/month.
đź“ť Simon Willison
Introducing the Codex app - OpenAI released a new macOS app for their Codex coding agent, featuring a nice UI and new functionalities.
đť•Ź claire vo đź–¤ @clairevo
Claire Vo upgraded her AI-driven simulation Maplewood.zone with automated death notifications, a morgue feature, more compelling dream sequences, and support for running multiple simulations after a 12-hour, $50-token test revealed a bug where characters weren’t informed of deaths.