Anthropic Releases Claude in Excel
Today's curated insights on AI product management from X, LinkedIn, and YouTube.
Anthropic Releases Claude in Excel
From X
AI Product Launches & Updates
Claude in Excel release: Claude @claudeai shared availability on Pro plans with multiple file drag-and-drop, session auto compaction, and cell overwrite avoidance.
Petri 2.0 release: Anthropic @AnthropicAI announced enhancements to counter eval-awareness and expanded behavior seeds in their open-source alignment audit tool.
LlamaCloud SDK & LlamaParse v2: LlamaIndex đŚ @llama_index launched a revamped integration with n8n, featuring stable nodes for parsing, extraction, classification, sheets, and a complete setup guide.
AI Tools & Applications
Under-hyped AI tools: Lenny Rachitsky @lennysan highlighted top under-hyped tools like WisprFlow, Conductor, NotebookLM, and MeetGranola to boost PM productivity.
Devin Review code diffs: Cognition @cognition shared how Devin Review detects copied and moved code lines in git diffs to streamline code reviews without burying small changes.
Product Management Insights & Strategies
AI-accelerated PRD workflow: George Nurijanian @nurijanian shared shipping two complete PRDs in 4 hours using a five-step Claude code agent process covering research, outlining, drafting, validation, and polishing.
PM role spectrum: George Nurijanian @nurijanian noted PM roles range from technical depth to stakeholder management to vision strategy, advising PMs to find roles that fit their strengths.
Higher-order outcomes: George Nurijanian @nurijanian argued PMs should prioritize delivering aspirational customer benefits over commoditized features to drive differentiation and pricing power.
AI Industry Developments & News
MetaâManus AI deal: DeepLearning.AI @DeepLearningAI reported Metaâs planned $2â3 billion acquisition of Manus AI to integrate autonomous agents across Facebook, Instagram, and WhatsApp.
Retriever limits study: DeepLearning.AI @DeepLearningAI highlighted research showing that single-embedding retrievers canât scale to retrieve all relevant document combinations, motivating multi-embedding methods.
AI for societal challenges: Google DeepMind @GoogleDeepMind shared Demis Hassabisâs vision for AI to address climate change and disease in an interview with CNBCi.
From LinkedIn ⢠Deeper Insights
AI Tools & Applications
Accelerating design and deployment with AI agents is on the rise. Peter Yang (@petergyang) highlights Google AI Studioâs ability to prototype multiple UI variations in one interface by simply asking AI to add a toggle buttonâstreamlining A/B testing and speeding up feedback loops. Meanwhile, Tal Raviv (@talsraviv) showcases Render as a âHeroku for coding agents,â where infrastructure and deployments live in a tiny config file and AI-driven agents handle both engineering and DevOps tasksâletting PMs delegate incident response end-to-end with minimal manual intervention.
Product Management Insights & Strategies
Defining clear intents for AI agents demands the same rigor as product specifications. Karthick Nethaji Kaleeswaran (@karthick-nethaji) introduces an Intent Engineering framework, arguing that most âwonkyâ agent behavior stems from under-specified constraints or ambiguous decision autonomy. By translating vague prompts into explicit requirementsâmapping inputs to outputs, outlining decision scopes, and enforcing guardrailsâPMs can architect reliable, predictable AI agent experiences.
AI Industry Developments & News
The ChatGPT App Store is emerging as a major new distribution channel. Aakash Gupta (@aagupta) breaks down why 900M+ weekly users and 26% higher conversion rates make this ecosystem compelling: start with easy-to-build utility apps, enable streaming responses for instant feel, and define smart trigger phrases to surface your app organically. He also recommends a lead-gen modelâusing ChatGPT as top-of-funnel and handing off transactions to your own site to retain full customer relationships and avoid platform fees.
From YouTube ⢠Video Content
Vibe coders are missing out...
AI Jason ⢠January 24, 2026
AI Jason demonstrates how to replace bulky MCP agent tools with lightweight skill+CRI integrations to cut token usage by over 70% while extending coding agent capabilities at scale.
Key Takeaways:
- Combining skill.md files with a CRI tool can reduce token consumption by more than 70% compared to default MCP setups.
- Each skill adds only 10â50 tokens to the context window, allowing up to ~4,000 skills versus hundreds of tokens per MCP schema.
- The open-source MCP Porter enables agents to run MCP commands via CRI, converting existing MCP tools into token-efficient skills.
I got a private lesson on Claude Cowork & Claude Code
Greg Isenberg ⢠January 23, 2026
In this episode, Boris demonstrates how Claude Coworkâs simple Mac appâpowered by the same Claude Code âquad agentââlets users mount local folders to rename receipt files, generate and format spreadsheets in Chrome, and send emails via Gmail, and shares best practices for Claude Code sessions using plan mode, Opus 4.5, and a shared Cloud.md knowledge base.
Key Takeaways:
- Claude Cowork is a Mac-only desktop app (Windows coming soon) that uses the Claude Code âquad agentâ SDK to access user-mounted folders, rename files by date with reverse elicitation, and output new files like spreadsheets.
- With the Cowork Chrome extension, the agent automates browser tasksâcreating and formatting a Google Sheet from receipt data and composing an email in Gmailâwhile enforcing safety via a sandbox VM and deletion protection.
- Boris recommends starting Claude Code sessions in plan mode to refine a task, then switching to auto-accept edits with Opus 4.5 for one-shot code generation, and maintaining a shared Cloud.md (checked into Git) with a GitHub action that lets you tag the Claude app on PRs to automatically update team guidance.
Claude Code Let's Build: The AI Video Oracle (Qwen3 TTS)
All About AI ⢠January 23, 2026
The video walks through building a local AI video pipeline on a MacBook using Gemini 3 Flash for research, Qwen 3 TTS (1.7B) for animeâstyle voice cloning, and the Omnihuman model to generate concise 20-second answer videos.
Key Takeaways:
- Qwen 3 TTS 1.7B runs locally via MPS on a MacBook in under a minute, producing a cloned Vtuber-style voice with surprisingly good quality for its size.
- The six-step pipelineâonline research with Gemini 3 Flash, summarization (â¤50 words), TTS audio generation, and Omnihuman avatar video assemblyâyields a final MP4 in about 5â7 minutes.
- In a demo, the system reported that at Davos 2026, Dario Amade likened U.S. chip exports to China to selling nuclear weapons to North Korea and predicted AI would automate most software engineering tasks within a year.