How Notion Built the Best AI Agents For Work (Full Tutorial)
Today's curated insights on AI product management from X/Twitter across 60+ expert sources and YouTube channels.
How Notion Built the Best AI Agents For Work (Full Tutorial)
From X
AI Product Launches & Updates
API Key & Projects Page: Logan Kilpatrick @OfficialLoganK announced a brand new API key and Projects page in Google AI Studio, making it easier to create, import, and manage projects with quality-of-life features like naming API keys.
Robot Phone by Honor: Unknown @theresanaiforit announced Honorâs AI-powered robot phone with a mechanical gimbal camera that tracks subjects, analyzes outfits and surroundings, and interacts with its environment.
AI Tools & Applications
Deep Agents Evolution: LangChainAI @LangChainAI shared a breakthrough AI architecture enabling agents to scale from 15 to 500+ steps with advanced planning and memory systems.
Article Explainer Tool: LangChainAI @LangChainAI introduced an AI document analysis tool using LangGraphâs Swarm Architecture to break down complex articles through interactive natural language queries.
AI-Powered Canvas Template: LangChainAI @LangChainAI released a production template for AI canvas apps with real-time UI-AI synchronization built on a Python-Next.js stack powered by LangGraph.
Product Management Insights & Strategies
Junior vs Senior PM Collaboration: George @nurijanian shared how involving engineering early transforms lengthy research sprints and ensures teams build features that match customer needs.
Big Tech PM Reality Check: George @nurijanian highlighted common pitfalls with ignored processes, unread documentation until executive review, and stakeholders ambushing decisions at late stages.
AI Industry Developments & News
RL vs Instruct Debate: Andrej Karpathy @karpathy clarified that reinforcement learning (RL) remains essential, layered on top of base model autocompletion and instruction tuning.
Cognitive Core Metaphor: Guillermo Rauch @rauchg praised stripping large language models to a âcognitive coreâ, combining cognition, knowledge, and skills as ideal ingredients for intelligent agents.
From YouTube
How Notion Built the Best AI Agents For Work (Full Tutorial) | Akshay & Ryan
Peter Yang ⢠October 19, 2025
Peter Yang hosts Notion co-founder Akshay and AI lead Ryan in a deep dive into Notionâs AI Agents, showcasing live demos of natural-language database creation, real-time web data integration, personalized memory pages, and shareable custom Agents with Slack automations. They also detail the two-year development journeyâpivoting to a markdown-based LLM architecture, building rigorous eval suites, and optimizing prompt engineeringâto deliver reliable AI tooling within Notion.
Key Takeaways:
- Notionâs AI Agents turn plain-language prompts into complex databasesâcomplete with custom properties, views, and real-time streamingâand can fetch external data like IMDb critic ratings via built-in web search.
- By representing pages as markdown for the LLM to read and write, then translating markdown back into Notion blocks, the team achieved far higher reliability than exposing internal JSON schemas.
- Custom Agentsâshareable across teams with âmemoryâ pages, autonomous triggers, schedules, and Slack integrationâquadrupled Notionâs internal AI usage, driving discussions around seat-based pricing.
How to measure AI developer productivity in 2025 | Nicole Forsgren
Lennys Podcast ⢠October 19, 2025
Nicole Forsgren explains why traditional metrics like lines of code and unadapted DORA scores mislead in an AI-driven world, how AI shifts focus from writing to reviewing code, and how companies can build a frictionless developer experience with her seven-step framework.
Key Takeaways:
- âMost productivity metrics are a lieâ because AI can generate verbose or low-quality code to game measures like lines of code, so teams must track code survivability, reliability and attribute human vs. AI contributions.
- DORAâs four metrics (deployment frequency, lead time, MTTR, change-fail rate) remain vital for pipeline performance but must be adapted for AI-driven feedback loops and paired with new trust metrics to detect hallucinations and style conformity.
- Forsgrenâs seven-step âFrictionlessâ process to improve developer experience starts with listening tours, securing quick wins, building data foundations, setting strategy, selling the plan, driving change at scale, and evaluating impact.