Base44 Releases Agents Integration to Whatsapp
Today's curated insights on AI product management from X/Twitter across 60+ expert sources and YouTube channels.
Base44 Releases Agents Integration to Whatsapp
From X
AI Product Launches & Updates
Magistral model performance leap: Mistral AI @MistralAI reported that its latest Magistral models achieved a massive leap on the Artificial Analysis Intelligence Index, effectively competing with much larger models.
New compute-intensive offerings: Sam Altman @sama announced upcoming compute-intensive OpenAI products, noting some features will initially be Pro-only with additional fees to manage costs.
WhatsApp integration for AI agents: Base 44 @base_44 launched direct integration, enabling Base44 Agents to operate inside WhatsApp chats for streamlined workflows.
AI Tools & Applications
Guide to shipping agentic AI: Aakash Gupta @aakashg0 highlighted a McKinsey framework that deconstructs how to build and deploy effective AI agents in production.
AI agent workflow comparison: Aakash Gupta @aakashg0 shared a deep dive contrasting n8n's visual simplicity with LangGraph's structured flexibility for designing agent pipelines.
Product Management Insights & Strategies
Managing AI like a team: Lenny Rachitsky @lennysan discussed why every PM will become an AI manager, with Julie Zhuo highlighting how traditional leadership skills apply to AI initiatives.
Portfolio strategy for AI PM roles: Aakash Gupta @aakashg0 shared that portfolios are essential for AI PM candidates, differentiating 4/5 candidates and boosting ROI for hiring managers.
AI Industry Developments & News
AI texting as next form factor: Logan Kilpatrick @OfficialLoganK asserted that text messaging with AI will reach 1 billion users, urging companies to prioritize conversational AI channels.
SWE-Bench Pro benchmark released: Alexandr Wang @alexandr_wang announced SWE-Bench Pro, evaluating code models on multi-file edits with complex dependencies, showing GPT-5 leading at 23.3%.
Hidden code in LLM activations: Andrej Karpathy @karpathy revealed that LLMs store generated code within value activations at layers 22–30, sparking discussions on model interpretability.
From YouTube
Full Tutorial: From Design to Code with Claude Code in 40 Minutes | Meaghan Choi
Peter Yang • September 21, 2025
Peter Yang interviews Meaghan Choi, Design Lead for Claude Code at Anthropic, as she demonstrates her end-to-end workflow — from codebase exploration and planning to Figma design and final UI polish — all powered by Claude Code’s integrated AI features.
Key Takeaways:
- Meaghan structures her workflow into three buckets—0→1 explorations within the live codebase, pre-brainstorm deep dives into existing implementations (down to system prompts and architecture), and Figma design followed by code prototyping and final 10% polish in Claude Code.
- She uses Control+B in Claude Code to run background bash commands (e.g., starting a dev server) while chatting, and leverages Claude’s vision to drag-and-drop PNG exports from Figma so Claude can adjust CSS classes for pixel-perfect header and image alignment.
- By maintaining both shared and personal cloud.md files, Meaghan customizes Claude’s behavior—guiding it to reference the component library, flag high-risk changes with emojis, and explain its code edits for designers.
From managing people to managing AI: The leadership skills everyone needs now | Julie Zhuo
Lennys Podcast • September 21, 2025
Julie Zhuo explains how core management principles—setting clear goals, assembling resources, and designing processes—apply to overseeing AI agents, why AI is reshaping teams into small “builder” units, and which leadership skills (self-awareness, feedback, adaptability) are essential in an era of rapid change.
Key Takeaways:
- The fundamentals of management—building a team, supporting growth, and creating smooth processes—map directly to AI by requiring crystal-clear success criteria, understanding each model’s strengths, and providing context and high-level instructions.
- AI tools can boost an individual’s skill from the bottom percentiles to the 60th–70th, enabling two-person “builder” teams that dissolve rigid roles like PM, designer, and engineer in favor of multi-disciplinary collaboration.
- “Diagnose with data and treat with design”: use data to objectively identify where growth stalls, then apply creative design and intuition to solve problems, while building observability frameworks and targeted experiments to guide decisions.