No Priors

From Coder to Manager: Navigating the Shift to Agentic Engineering with Notion Co-Founder Simon Last

Notion isn’t designing AI agents that just use tools. Their agents can autonomously build their own integrations, as well as write the code needed to finish a task. Sarah Guo sits down with Notion Co-Founder Simon Last to explore Notion’s rapid evolution from a simple writing assistant to a sophisticated platform for custom AI agents. Simon discusses the technical hurdles of indexing disparate data from sources like Slack and Google Drive, as well as the internal shift toward using coding agents to build Notion itself. Plus, Simon elaborates on what he sees as a fundamental transition in productivity: moving from a tool where humans do the work, to one where humans manage a swarm of agents. Sign up for new podcasts every week. Email feedback to show@no-priors.com Follow us on Twitter: @NoPriorsPod | @Saranormous | @EladGil | @simonlast | @NotionHQ Chapters: 00:00 – Cold Open 00:05 – Simon Last Introduction 00:26 – Genesis of Notion AI 04:10 – Challenge of Semantic Indexing and Retrieval 07:16 – The Six-Month Rewrite Cycle 08:12 – Notion’s Coding Agent Era 09:44 – Impact on Team Dynamics 12:49 – Launching Custom Agents 15:39 – Notion as the ‘Switzerland’ for Models 17:33 – Designing APIs for Agent Customers 20:09 – Simon’s Personal Agentic Workflows 24:48 – Notion: Tool for Work is Now A Tool for Agents 27:28 – How Building Has Changed for Simon 29:00 – Conclusion

SGSarah GuohostSLSimon Lastguest

Mar 12, 2026·29m·Watch on YouTube ↗

📖 CHAPTERS

1. 1

   GPT-4 in Mexico: the moment Notion’s AI strategy became real

   Simon recounts first hands-on access to GPT-4 at an offsite and why it immediately shifted Notion’s urgency around AI. Two realizations stood out: the model could follow complex instructions and it had unusually broad/deep world knowledge.

2. 2

   Two-track roadmap: shipping a writing assistant now vs. building a general assistant later

   Notion formed both a short-term and long-term vision right away. The short-term was an in-doc writing assistant that could be delivered quickly; the long-term was a tool-using assistant that can operate across Notion objects (docs, databases, queries) to complete multi-step tasks.

3. 3

   Shipping timeline: AI Writer to workspace Q&A and beyond

   Simon lays out the product arc: AI Writer launched first (Feb 2023), then a more complex workspace Q&A feature shipped later (GA Oct 2023). Q&A required a semantic index, grounding, and stronger evaluation rigor than simple text generation.

4. 4

   The hard part: semantic indexing & retrieval across messy, heterogeneous data

   They discuss why search/Q&A is difficult, especially across sources like Notion, Slack, and Google Drive. Simon argues success comes from AI-savvy iteration plus craft: empirically testing queries, tuning retrieval, and tailoring pipelines per data source.

5. 5

   Why workspace organization matters less (and where it still matters a lot)

   Embeddings reduce reliance on human-friendly folder/page hierarchies, since models mostly need the right textual snippet. But system-level choices like chunking and retrieval remain critical, even if invisible to end users.

6. 6

   The six-month rewrite rule: constantly rebuilding the AI harness

   Notion rewrites its AI harness roughly every six months as models and best practices evolve. Simon frames this as a common failure mode for companies—locking into an early approach rather than redesigning systems around current model capabilities.

7. 7

   From autocomplete to true coding agents: the ‘agent era’ in engineering

   Simon describes successive tooling phases, culminating in end-to-end coding agents (early last year) that can implement, verify, and maintain changes. The key is pairing agents with strong architecture and verification loops—otherwise output devolves into “slop.”

8. 8

   Team dynamics with agents: higher ceilings, more prototypes, and controlled chaos

   Agents increase individual output potential and widen the gap between low and high performers. Internally this creates more experimentation—more prototypes and bigger PRs—while still favoring small tiger teams and strong tool adoption.

9. 9

   Engineering safety in an agent world: PR reviews, testing rigor, and avoiding vibe-coding

   Even with agent-written PRs, Notion keeps review discipline. Simon emphasizes demanding stronger end-to-end testing and thinking through deploy/verification, using agents as collaborators rather than replacing engineering judgment.

10. 10

    From personal agent to custom agents: autonomous background work and new workflows

    Notion’s agent vision became real with a personal agent shipped last fall (Aug/Sept) and custom agents launched recently. Custom agents start with no access by default, can be granted permissions, and can run autonomously (e.g., Slack-connected task filing).

11. 11

    Agents that write code as a primitive: bootstrapping new integrations and capabilities

    Simon argues coding agents are a core primitive—possibly the kernel of AGI—because code enables deterministic logic and capability bootstrapping. He imagines agents that can create missing integrations themselves, deploy them, and then use them to complete tasks.

12. 12

    Notion as ‘Switzerland’ for models: multi-model support, open source, and cost-performance tradeoffs

    Notion positions itself as model-agnostic so customers aren’t locked into a single lab. Simon highlights fast-changing leaderboards, the rise of strong open-source (including multiple Chinese models), and the value of cheaper models for many tasks.

13. 13

    Designing APIs for agent customers: markdown for pages, SQLite for databases

    Agents are treated as a new API customer with different constraints (token efficiency, verbosity sensitivity). Notion reworked interfaces: a Notion-enhanced markdown dialect for page I/O and SQLite for database querying, replacing overly verbose JSON block formats.

14. 14

    Simon’s personal agentic workflows: always-on coding agents and custom agent routines

    Simon describes running multiple CLI-based coding agents continuously (including overnight and a 13-day run). He also uses Notion’s personal and custom agents for operational workflows like email triage and routing internal feedback/bugs from Slack into the right team systems.

15. 15

    From ‘tool for work’ to ‘tool for managing agents’: how Simon’s building role changed

    Notion’s mission shifts from helping humans do work directly to helping humans manage agents doing the work. Simon notes he no longer types code; instead he designs end-to-end tasks and verification loops, acting as the manager and final verifier rather than the implementer.