AI Agents Make Slow CI a Death Sentence
For AI-powered engineering teams, slow CI bottlenecks thousands of PRs. Ryan Nystrom and Claire Vo explain why fast CI isn't optional.
Spec-driven development: the AI engineering workflow at Notion | Ryan Nystrom
With Claire Vo, Ryan Nystrom · Sunday, May 17, 2026
Claire Vo and Ryan Nystrom, an engineering manager at Notion, discuss how AI has revolutionized their engineering workflows. They cover an AI agent that automates meeting preparation, a system for AI-powered code generation and bug fixes, and a novel spec-driven development process where AI builds from natural language specifications. The conversation emphasizes the increased efficiency, enjoyment, and the critical role of fast CI in this new AI-driven engineering landscape.
Key takeaways
- Your engineering team's comfort making frequent changes directly links to CI speed; slow CI creates “monster changes” no one wants to review. Read more →
- Automate Low-Value Prep: Notion's "Hot Potato" AI agent autonomously gathers updates from Slack, tasks, and pull requests, generating a meeting pre-read daily at 9:00 a.m. Read more →
- Notion’s internal “Boxy” system integrates an AI agent called Codeex to automate code generation and bug fixes directly from Notion tasks. Read more →
- Notion engineering manager Ryan Nystrom treats AI coding agents like Codeex as junior engineers: demanding explanations and forcing them to defend their work. Read more →
- Notion's engineers, led by Ryan Nystrom, are ditching traditional code-first development for a spec-driven approach. They start with comprehensive, plain English markdown documents, not a single line of code. Read more →
5 articles from this episode
Notion's 'Hot Potato' AI Agent Cuts Standup Prep to Zero
Learn how Notion uses its 'Hot Potato' AI agent to automate daily standup prep, shifting focus from status updates to high-quality discussions.
Notion's AI Agents Write and Fix Code in 20 Minutes Flat
Notion's Ryan Nystrom reveals how their internal 'Boxy' system uses AI (Codeex) to generate full PRs, including UI verification, from natural language prompts. Automate grunt work now.
Ryan Nystrom's AI Prompt: 'You're Wrong. Defend Your Argument.'
Notion's Ryan Nystrom shares his aggressive AI prompting strategy: admitting ignorance and telling the AI 'you're wrong.' Get cited evidence, not just code.
Notion's AI Engineers Write English Specs, Not Code
Notion engineering manager Ryan Nystrom reveals their AI-driven workflow where features are built from plain English markdown specs, shifting engineers to architecture and verification.
More How I AI episodes
- Claude Code for creators: AI editing is hereOct 25 · 4 articles
- Meta Muse: the AI agent people actually loveOct 18 · 5 articles
- Is Meta’s Muse Wonder model any good?Oct 18 · 5 articles
- How the OpenAI team uses ChatGPT Sites dailyOct 11 · 5 articles