From Coder to Manager: Navigating the Shift to Agentic Engineering with Notion Co-Founder Simon Last
Episode
29 min
Read time
2 min
Topics
Remote Work, Startups, Leadership
AI-Generated Summary
Key Takeaways
- ✓AI Harness Cadence: Rebuild AI system architecture approximately every six months rather than maintaining a single implementation. Model capabilities advance fast enough that a harness designed around last year's models underperforms. Notion is currently deploying a new harness while already designing the next one, treating rewrites as a standard part of the product cycle.
- ✓Agent-Optimized APIs: Default APIs built for humans perform poorly with agents. Notion replaced its verbose JSON block format with a custom Markdown dialect for pages and switched to SQLite syntax for database queries. Identify which data formats models are already trained on and design agent interfaces around those priors rather than repurposing human-facing endpoints.
- ✓Memory-Loop Agent Training: Build agents with a writable memory page, run them in approval mode for several days, and correct their outputs interactively. The agent converts corrections into explicit rules. After roughly two weeks of iteration, Last removed human approval entirely from his email-triage agent, which now autonomously archives messages using hundreds of self-generated routing rules.
- ✓Verification-First Coding Workflow: Effective agent-assisted engineering requires defining the change, the verification method, and the safe deployment path before prompting. Last no longer writes code manually, instead specifying end-to-end tasks and acting as outer verifier. This approach produces larger, more complex pull requests but with full end-to-end test coverage on every submission.
- ✓Retrieval Quality Requires Source-Specific Tuning: Semantic indexing across different data sources — Slack, Google Drive, Notion — demands separate chunking strategies and retrieval pipelines per source. A single universal approach degrades performance. Notion iterates continuously by running real queries daily against each source, treating retrieval tuning as ongoing craft work rather than a one-time configuration decision.
What It Covers
Notion Co-Founder Simon Last describes how Notion rebuilt its AI architecture roughly every six months, launched personal and custom agents in 2024, and shifted the engineering role from writing code to managing agents that autonomously execute, verify, and deploy end-to-end tasks across workspaces.
Key Questions Answered
- •AI Harness Cadence: Rebuild AI system architecture approximately every six months rather than maintaining a single implementation. Model capabilities advance fast enough that a harness designed around last year's models underperforms. Notion is currently deploying a new harness while already designing the next one, treating rewrites as a standard part of the product cycle.
- •Agent-Optimized APIs: Default APIs built for humans perform poorly with agents. Notion replaced its verbose JSON block format with a custom Markdown dialect for pages and switched to SQLite syntax for database queries. Identify which data formats models are already trained on and design agent interfaces around those priors rather than repurposing human-facing endpoints.
- •Memory-Loop Agent Training: Build agents with a writable memory page, run them in approval mode for several days, and correct their outputs interactively. The agent converts corrections into explicit rules. After roughly two weeks of iteration, Last removed human approval entirely from his email-triage agent, which now autonomously archives messages using hundreds of self-generated routing rules.
- •Verification-First Coding Workflow: Effective agent-assisted engineering requires defining the change, the verification method, and the safe deployment path before prompting. Last no longer writes code manually, instead specifying end-to-end tasks and acting as outer verifier. This approach produces larger, more complex pull requests but with full end-to-end test coverage on every submission.
- •Retrieval Quality Requires Source-Specific Tuning: Semantic indexing across different data sources — Slack, Google Drive, Notion — demands separate chunking strategies and retrieval pipelines per source. A single universal approach degrades performance. Notion iterates continuously by running real queries daily against each source, treating retrieval tuning as ongoing craft work rather than a one-time configuration decision.
Notable Moment
Last described running a single coding agent continuously for thirteen days without interruption, feeding it a task queue each night sized to outlast his sleep. He frames waking up to a still-running agent as the benchmark for having correctly scoped and prompted the overnight workload.
Episode Transcript
Hi, listeners. Welcome back to NoCriers. Today, I'm here with Simon Last, cofounder at Notion. We talk about their new vision for Notion in the AI age as a platform for humans and agents to collaborate, how the engineering and product org at Notion is changing, and these new tools for thought. Welcome, Simon. Hey, Simon. Thanks for doing this. Yeah. Of course. Yeah. It's really fun to be here. Notion's at scale, amazing platform, lots of users. You did start quite a while ago. I think of Notion as one of the companies that has really, like, braced AI quite aggressively. I was told you first got your hands on GPT four, at a company off-site in Mexico. Is that true? What is the origin story of, like, starting to work on this stuff? Yeah. I think yeah. That year, that was 2022. I I've been watching, you know, what's going on. In general, I've just been, like, super curious about the technology and fascinated to to try everything and think about, like, like, how we can apply it. It wasn't until I played with GPT four that it it became really, really real. So, you know, we and when we got access to it, it it was sort of like a a proto chat gbt like interface. And, my cofounder Ivan and I both a book on access, and it was just immediately clear, like, I would say two big things. One is that it was just pretty smart. It it it could follow reasonably complicated instructions. It could write things for you. It could edit things. And and the second big thing was that, of of the scope of its knowledge was extremely interesting. Super, super deep, like, and and broad world knowledge. When we play with it, it became just instantly clear to both of us. Like, okay, the the time is now to start, but we think about how to apply this, it's only gonna get better. We're talking about Mexico, GPT four. You guys saw it was, like, clearly the time. Did you start with, like, a particular vision of, like, what you should obviously be able to do with AI and Notion? Or just start pulling people from different teams or recruiting people and say, like, let's experiment? How did you begin? I think we immediately had a long term and a short term vision. I would say the the I'll start with the short term one. The the thing that was immediately obvious was, oh, it could be like a writing assistant. Mhmm. So it it could be in your document. You can, like, select some text, have it rewrite it. You can have it write text for you, maybe look something up, and then, you know, give you, like, like, sources or more information. So that was the thing that we immediately, like like, got to work on, and, you know, we sort of started a tiger team around it. And …
Get the full transcript (6,174 words) + summary by email — free
One-time email with the complete transcript and AI summary of this episode. No account needed.
One email, no spam. We’ll also show you what SignalCast does.
Browse all No Priors: Artificial Intelligence | Technology | Startups transcripts →
You just read a 3-minute summary of a 26-minute episode.
Get No Priors: Artificial Intelligence | Technology | Startups summarized like this every Monday — plus up to 2 more podcasts, free.
Pick Your Podcasts — FreeKeep Reading
More from No Priors: Artificial Intelligence | Technology | Startups
Redefining Chip Architecture with Arm CEO Rene Haas
Sep 3 · 37 min
Latent Space
Retrieval After RAG: Hybrid Search, Agents, and Database Design — Simon Hørup Eskildsen of Turbopuffer
Mar 12
More from No Priors: Artificial Intelligence | Technology | Startups
Rethinking Legacy Data Infrastructure with Eon Co-Founders Ofir Ehrlich and Gonen Stein
Aug 27 · 34 min
The Long Run with Luke Timmerman
Ep203: Sri Kosuri on Small Molecule Correctors of Disease
Jun 16
More from No Priors: Artificial Intelligence | Technology | Startups
We summarize every new episode. Want them in your inbox?
Redefining Chip Architecture with Arm CEO Rene Haas
Rethinking Legacy Data Infrastructure with Eon Co-Founders Ofir Ehrlich and Gonen Stein
From Restoring Sight to Reimagining the Brain, with Max Hodak
What Chess.com Teaches US About Superhuman Capabilities, with CEO Erik Allebest
Chasing Trillion-Dollar Companies, Founder Ambition, Token Budgets, and Regulatory Capture with Sarah & Elad
Similar Episodes
Related episodes from other podcasts
Latent Space
Mar 12
Retrieval After RAG: Hybrid Search, Agents, and Database Design — Simon Hørup Eskildsen of Turbopuffer
The Long Run with Luke Timmerman
Jun 16
Ep203: Sri Kosuri on Small Molecule Correctors of Disease
Latent Space
Apr 15
Notion’s Token Town: 5 Rebuilds, 100+ Tools, MCP vs CLIs and the Software Factory Future — Simon Last & Sarah Sachs of Notion
a16z Podcast
Jan 22
Inferact: Building the Infrastructure That Runs Modern AI
The AI Breakdown
Sep 6
How to Build an AI-Native Company Today
Explore Related Topics
This podcast is featured in Best AI Podcasts (2026) — ranked and reviewed with AI summaries.
Read this week's Startups & Product Podcast Insights — cross-podcast analysis updated weekly.
You're clearly into No Priors: Artificial Intelligence | Technology | Startups.
Every Monday, we deliver AI summaries of the latest episodes from No Priors: Artificial Intelligence | Technology | Startups and 192+ other podcasts. Free for one show.
Start My Monday DigestNo credit card · Unsubscribe anytime