Anthropic Introduces Claude Haiku 4.5: Faster, Cheaper Coding
Today's curated insights on AI product management from X/Twitter across 60+ expert sources and YouTube channels.
Anthropic Introduces Claude Haiku 4.5: Faster, Cheaper Coding
From X
AI Product Launches & Updates
Introducing Claude Haiku 4.5: Claude @claudeai introduced Haiku 4.5, a small model matching Sonnet 4’s coding performance at one-third the cost and twice the speed, available via API, Amazon Bedrock, and Vertex AI.
Major Veo 3.1 upgrade: Demis Hassabis @demishassabis announced Veo 3.1, featuring enhanced realism, richer audio, scene extension, and refined narrative controls, accessible through the Gemini API and Gemini App.
AI Chat Memory feature: Qwen @Alibaba_Qwen unveiled Qwen Chat Memory, a system that stores and recalls personal context and history to deliver richer, more tailored user experiences.
AI Tools & Applications
LaTeX support in NotebookLM: Josh Woodward @joshwoodward shared that NotebookLM now supports LaTeX, enabling formatted equations and technical documentation directly within the tool.
LangSmith debugging UX: LangChain AI @LangChainAI highlighted how LangSmith offers an intuitive interface to explore and debug logs and traces for large language model applications.
Quick custom extraction agents: Llama Index @llama_index demonstrated LlamaAgents, showing how to build production-ready receipts extraction agents in minutes with custom schemas and confidence scoring.
Product Management Insights & Strategies
Prototype in a day workflow: Lenny Rachitsky @lennysan described a team process where ideas generated in the morning become prototypes by day’s end, underscoring rapid iteration.
Essential PM websites: George from 🕹prodmgmt.world @nurijanian recommended three key sites—Loopy for systems thinking, Nautex for agent requirements, and prodmgmt.world for AI PM frameworks.
PM/Eng/Design collaboration seminar: Shreyas Doshi @shreyas announced a session on stress points and actionable strategies for PM, engineering, and design teams on Oct 29, available live or on-demand.
AI Industry Developments & News
Cancer therapy AI breakthrough: Sundar Pichai @sundarpichai revealed that the C2S-Scale 27B foundation model, built with Yale, generated a novel cancer cell behavior hypothesis validated in living cells.
Coral NPU platform launch: Google Research @GoogleResearch introduced Coral NPU, an open-source, full-stack Edge AI platform optimized for low-power devices.
OpenAI & Meta consumer launches: DeepLearningAI @DeepLearningAI reported that OpenAI and Meta unveiled Sora 2, ChatGPT Pulse, and Instant Checkout to boost engagement and revenue.
From YouTube
Figma’s CEO: Why AI makes design, craft, and quality the new moat for startups | Dylan Field
Lennys Podcast • October 16, 2025
Dylan Field recounts Figma’s rebound from a failed Adobe acquisition to its IPO, details the “follow the workflow” expansion into FigJam, Slides, Buzz, Sites, Dev Mode and AI-powered Figma Make, and argues that AI makes design, craft and quality the new startup moat.
Key Takeaways:
- After the Adobe deal fell through, Figma held legal-permitted, increasingly frequent all-hands briefings and launched a voluntary “Detach” severance program (accepted by 4% of staff) to reset focus and accelerate toward IPO.
- Field’s “follow the workflow” playbook guided Figma’s product line by extracting whiteboarding into FigJam, presentations into Slides, marketing assets into Buzz, websites into Sites and code handoff into Dev Mode before launching Figma Make.
- Figma Make uses generative AI to turn user prompts into interactive prototypes or working apps with design-system compliance and round-trip editing in Figma Design, cutting time-to-value and letting PMs and non-designers co-create early prototypes.
How I Use AI as a PRO to Land $5K+ Jobs Organically
All About AI • October 16, 2025
All About AI’s video demonstrates using AI tools to build a web-based Windows XP–style presentation on context engineering, complete with transcribed research, structured talking points, mermaid diagrams, embedded video clips, real-time quizzes, and audience Q&A. This AI-powered workflow streamlines slide creation, deployment on Vercel, and engagement tactics ideal for landing $5K+ speaking engagements.
Key Takeaways:
- Uses YouTube-dl Python script plus OpenAI’s Whisper API to download and transcribe research videos into text files for AI-driven slide content.
- Leverages ChatGPT GP5 to generate chronologically ordered bullet-point talking points, five main topics, and mermaid flowcharts, then refines layout in Claude with a Windows XP theme via Nano Banana.
- Deploys the presentation as a web page on Vercel, integrates a database-backed /qa endpoint for live audience questions, and inserts real-time quiz slides through a /quiz page between sections to boost engagement.
Genspark's Super AI Agent is INSANE
Greg Isenberg • October 15, 2025
Greg Isenberg reviews Genspark’s Super AI Agent end-to-end, showcasing its multi-agent chat, image, video, slides, sheets, Photo Genius mobile app, AI calling agent and tool integrations to evaluate quality, token savings, and workflow impact on a $20/month plan.
Key Takeaways:
- Genspark’s multi-agent chat workflow simultaneously pings GPT5, Code Soneta 4, and Gemini 2.5 Flash with built-in reflection to generate a refined 30-second YouTube intro in one prompt, cutting out manual LLM ping-pong.
- The image/video multi-agent feature uses Nano Banana, Flux, by Dance Seed Dream, SeaDance Light, Gemini V3 and Miniax with autoprompt optimization to produce personalized banana-eating photos and 3–10-second cinematic videos with audio in seconds.
- Photo Genius on iOS applied a subtle smile and corrected eye gaze on Greg’s selfie via voice command in under two seconds, and the AI calling agent let “Greg’s AI assistant” call his editor Ruby for a YouTube metrics update with transcription emailed afterward.
Build live voice agents that listen, reason, and respond, using Google’s ADK
Deeplearning.ai • October 15, 2025
This video teaches you how to build modular voice agents with Google’s open-source ADK—using code, English instructions, or a web UI—to process spoken input, integrate with tools like Google Search, and output synthesized voice, culminating in a project that orchestrates planner, researcher, and writer agents to build a JavaScript-focused, multi-host podcast agent.
Key Takeaways:
- ADK provides modular, model-agnostic building blocks—including models, tools, memory, orchestration, and observability—and supports development via code, natural language instructions, or a web UI.
- It features patent-pending multi-agent workflows that let specialized agents (planner, researcher, writer) collaborate on different parts of a task.
- The video’s hands-on project shows building a podcast agent that accepts voice input, uses tools like Google Search and external APIs to research a JavaScript topic, and generates a multi-host podcast with prompt optimization and callbacks.