CHAPTERS
- 0:00 – 2:01
Defining Notion: one workspace built from software “Legos”
Sarah asks the deceptively hard question—what is Notion—and Ivan frames it as a unified tool for personal and company work. He explains Notion’s core approach: provide flexible building blocks (text, databases, permissions, etc.) so users can assemble workflows without being trapped in fragmented apps.
- •Notion aims to be one place for notes, docs, tasks, knowledge base, and more
- •Market pain: SaaS fragmentation across many specialized tools
- •Building-block/“Lego” philosophy vs cramming every use case into one rigid app
- •Modern spin on earlier computing ideas using cloud and AI
- 2:01 – 4:22
Why unified tools are rare: no-code, budgets, and the literacy gap
Ivan explores why more companies haven’t successfully unified tools and data. He contrasts the no-code movement with the deeper need for shared knowledge spaces, and argues that computing should be a kind of literacy—yet most people use computers in limited, non-creative ways.
- •Different attempts at unification: no-code, budget consolidation, and now embeddings
- •Most users treat computers as typewriters/media devices rather than malleable tools
- •Separation between software makers and software users drives economic concentration
- •AI changes the equation by making software and “thinking work” more accessible
- 4:22 – 5:42
AI’s near-term sleeper hit: embeddings/RAG and the end of manual retrieval
Sarah asks what Ivan is most excited about from AI over the next decade; Ivan narrows the timeline to the next year or two. He highlights embeddings and retrieval as underappreciated breakthroughs that reduce the need for human organization and make knowledge recall conversational.
- •Embeddings enable semantic understanding of what you store, not just keyword matches
- •RAG reduces the need for careful organization to enable later retrieval
- •Notion AI-style Q&A becomes a fast path to institutional memory
- •Agents/workflows are exciting too, but RAG is the quieter, high-impact shift
- 5:42 – 8:43
Building the Notion AI team: probabilistic engineering and applied talent
Elad digs into what talent is required to become AI-first at application scale. Ivan describes the shift from deterministic software to ‘baking/gardening’ style iteration, the emergence of “AI engineers,” and why Notion emphasizes applied work and scaling over pure research.
- •AI product-building is stochastic/probabilistic vs deterministic engineering
- •Need for patient iterators who can “massage” systems end-to-end
- •Rise of young “AI engineer” profiles blending prompting, UX, and implementation
- •Notion remains application-layer focused; scaling to tens of millions is its own challenge
- 8:43 – 12:07
Notion as an application-building company: Engelbart, malleable software, and templates
Elad asks about Notion being less ‘productivity’ and more ‘application building.’ Ivan traces the origin to Engelbart and early computing systems like Smalltalk where software was modifiable, then shares a key learning: most users don’t want to build tools, so Notion packages blocks into templates and ready-to-use workflows.
- •Inspiration: “Augmenting Human Intellect” and malleable early software
- •Goal: restore tinkerability/customization in modern cloud software
- •Reality: most users don’t wake up wanting to build their own apps
- •Product shift: templates and pre-packaged use cases as the adoption bridge
- 12:07 – 14:52
Prioritizing AI investment: GPT-4 as the conviction moment and ‘bet the company’
Sarah asks how Notion resourced and prioritized AI amid other product demands. Ivan recounts early exposure to GPT-3 (didn’t fully click) versus seeing GPT-4 (reasoning/workflow “aha”), leading to a decisive organizational push to hire, align teams, and re-evaluate which “Lego bricks” pair best with AI.
- •Early GPT-3 felt mostly like draft-writing; GPT-4 triggered ‘it can reason’ realization
- •Knowledge work as information ‘paper pushing’ is directly targetable by LLMs
- •Company-level commitment: align convictions, hire ML, and redirect effort
- •Ongoing question: which primitives (editor, DB, KB) are most transformed by AI
- 14:52 – 17:48
The rapid evolution cycle: AI is a new material with new trade-offs
Elad asks what’s missing for full advantage; Ivan responds with a trade-offs framework. He describes how LLM capabilities and techniques evolve weekly, forcing companies to continuously discover what’s possible, while also navigating human inertia and adoption constraints—products must be new but not alien.
- •Tech progress forces constant re-evaluation of what can be built
- •Key model dimensions: context window, reasoning, speed, cost/footprint
- •Different tasks need different models (reasoning vs cheap summarization)
- •Adoption is behavioral: users accept incremental change (Jobs/‘3% difference’)
- 17:48 – 19:51
Notion Q&A explained: RAG as perfect memory that reduces Slack/email interruptions
Sarah asks for a plain-language description of Notion’s AI Q&A. Ivan explains RAG as semantic recall that makes organization optional and cuts time spent asking coworkers for information; he notes the scaling challenges and that they’re early in what workplace RAG can do.
- •RAG replaces brittle keyword search with semantic understanding
- •Users can ‘throw anything in’ and still retrieve later via questions
- •Transforms internal knowledge discovery and reduces back-and-forth comms
- •Operational reality: hard to run at scale, hence gating/waitlists
- 19:51 – 22:44
Calendars, agents, and workflows: from scheduling to delegation
Elad links calendaring to the broader agentic future and asks how AI applies. Ivan separates the ‘knowledge/RAG’ bucket from the ‘workflow/agent’ bucket, positioning calendars as a simpler workflow domain and hinting at a future where AI reduces the need for meetings and synchronous communication.
- •Two buckets: knowledge retrieval (RAG) vs workflows (agents)
- •Calendars are a ‘baby step’ toward broader agentic coordination
- •AI could pre-load meeting context and handle scheduling reshuffles
- •Long-term question: do we communicate more or less when AI mediates work?
- 22:44 – 24:23
Moving past the need for organization: dump-first UX and the declining role of the sidebar
Sarah asks whether people can become more disorganized if AI can retrieve anything. Ivan argues embeddings make manual indexing less necessary; he envisions ‘dump and retrieve’ as the default, potentially challenging iconic UI elements like Notion’s left sidebar and forcing teams to avoid innovator’s dilemma.
- •Organization historically served retrieval; embeddings shift that burden to machines
- •Future capture: quick dumps (text/photos) with AI-driven structure later
- •UI implications: even core paradigms like the left sidebar may become optional
- •Analogy: brains aren’t explicitly organized, yet recall works—software may follow
- 24:23 – 27:21
History of SaaS doesn’t repeat, it rhymes: bundling/unbundling and AI-driven re-bundling
Sarah asks how history informs Notion’s AI strategy. Ivan frames the industry as alternating bundling/unbundling cycles—from early PC fragmentation, to Microsoft bundling, to SaaS unbundling—and argues LLMs create a new pressure to re-bundle information because models work best when data and endpoints are connected.
- •Business cycles: ‘long divided must unite’—bundling vs unbundling
- •SaaS era amplified fragmentation due to web and cheap capital
- •LLMs push toward connected endpoints and shared information spaces
- •Bundling now is about both macro economics and model/data efficiency
- 27:21 – 30:07
Bundling as strategy: distribution bundles vs information bundles (and enterprise value)
Elad adds that founder behavior has shifted toward building integrated bundles that are ‘good enough’ because cross-sell and integration win (e.g., HubSpot/Rippling). Ivan distinguishes bundling distribution (Microsoft-style) from bundling information (LLM/embedding-space-driven), and Sarah notes historical parallels like Oracle’s acquisition-and-sell playbook.
- •Modern bundlers accept 80–90% parity in exchange for an integrated suite
- •Two bundling types: distribution bundle vs shared data/information bundle
- •Embedding-space benefits amplify the value of unifying tools under one roof
- •Enterprise appeal: fewer vendors, lower cost, CFO-friendly consolidation
- 30:07 – 34:26
Design at Notion: centralized systems thinking and ‘Apple for software’ ambition
Sarah asks about Notion’s design obsession and how it scales. Ivan defines design as system integration and trade-offs, arguing Notion must be more centralized—like an OS or programming language—so pieces work together cohesively, aspiring to an Apple-like experience in software.
- •Design is about how systems plug together, not just visual polish
- •Bundled products require horizontal, holistic coherence
- •Centralized design decisions resemble OS/language creation constraints
- •Positioning: more Apple-like integration, less Amazon-like decentralization
- 34:26 – 36:37
Office aesthetics and culture: no-shoes lore, furniture longevity, and cohesive vibe
Elad and Ivan shift to Notion’s office culture and physical design choices. Ivan explains the ‘no shoes’ tradition and how furniture, lighting, and timeless design classics reinforce a long-term product mindset; the segment also touches on consistency of vibe and the practicalities of scaling culture.
- •No-shoes policy across early offices; later dropped due to uncomfortable rug choice
- •Office design intentionally avoids ‘corporate’ feel, using home-like furniture
- •Preference for enduring design classics mirrors building software to last decades
- •Cohesive environment (music, aesthetics) as part of company identity
- 36:37 – 42:07
Model strategy and roadmap implications: multi-model usage and RAG reshaping core UX
Elad asks about using one LLM vs many and how AI impacts different Notion templates/use cases. Ivan notes using multiple providers/models depending on needs (reasoning vs speed/cost) and reiterates that RAG most fundamentally changes knowledge work—potentially removing the need for manual structure and redefining how users navigate Notion.
- •Multi-model approach: select models based on reasoning vs fast/cheap tasks
- •RAG is the biggest shift for knowledge-base and retrieval-driven workflows
- •Potential UX disruption: navigation/organization paradigms may become secondary
- •Notion positions itself in ‘front office’ work while vertical back-office AI also grows
