Notion's AI engineering workflow cuts CI time by 75% with spec-driven development

The Gist

• Ryan Nystrom leads Project Afterburner to slash Notion's CI time to 25% of current duration
• Custom AI agents auto-generate standup pre-reads by pulling Slack, GitHub, and meeting transcripts
• Engineers @mention Codex in Notion comments to get full PRs with screenshots in 20 minutes

• Notion's Project Afterburner aims to cut CI time by 75% through spec-driven development.
• AI agents can auto-generate daily standup pre-reads by pulling data from Slack, GitHub, and Honeycomb metrics.
• Notion's internal 'Boxy' system allows engineers to @mention Codex in Notion comments to generate pull requests in 20 minutes.
• Spec-first development involves dictating ideas into Whisper, formatting them as specs with Codex, and letting AI agents implement them autonomously.
• Fast CI is critical in the age of AI coding agents to maintain efficiency.
• Engineering managers and executives should continue writing code to stay connected to technical workflows.

• Adopt spec-driven development to streamline CI pipelines and reduce deployment times.
• Implement AI agents to automate routine tasks like standup pre-reads and PR generation.
• Encourage engineering leaders to stay hands-on with coding to maintain technical relevance.
• Explore integrations with tools like Whisper, Codex, and Honeycomb to enhance AI-driven workflows.

• 75% (Reduction in CI time targeted by Notion's Project Afterburner.)
• 20 minutes (Time it takes for Notion's 'Boxy' system to generate a pull request from a comment.)
• 1,300 PRs weekly (Number of pull requests Stripe's AI coding agents ship, as referenced for comparison.)