Observed Signal · May 11, 2026 · Technical Release · Source: Lennys Newsletter · Impact: 2/5 · Sentiment: Neutral
Notion's Spec-Driven AI Engineering Workflow
Ryan Nystrom, a software engineer at Notion, describes the company’s spec-driven AI engineering workflow and agent-driven development practices. Nystrom — who joined Notion after its acquisition of Campsite in December 2024 — was a core builder of Notion AI and led work on Notion Custom Agents (launched February 2026). He runs Project Afterburner to reduce CI times and manages a small engineering team. The piece details internal tooling such as “Boxy,” a VM-based background agent system that lets engineers prompt Codex from Notion comments to generate pull requests, and a spec-first workflow that uses Whisper to capture ideas, Codex to format specs, and autonomous agents to implement and verify changes. The article highlights the importance of fast CI, prompting agents to defend reasoning, subagent/MCP integrations, and why engineers and managers should continue coding.
Describes agentic development practices and an internal agent/product release (Notion Custom Agents) that illustrate evolving engineering workflows; relevant to AI/product engineering but not directly industry-shifting for AdTech.
Track Notion Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Ryan Nystrom is a software engineer at Notion and joined after Notion acquired Campsite in December 2024.
- Nystrom was a core builder of Notion AI and worked on Notion Custom Agents, which Notion released in February 2026.
- He manages a team of six to seven engineers and leads Project Afterburner, an initiative to cut Notion’s CI time to a quarter of its current duration.
- Notion uses an internal VM-based background agent system called "Boxy" that lets engineers @mention Codex from comments to generate pull requests with screenshots.
- Notion follows a spec-first workflow: dictate ideas into Whisper, have Codex format specs, commit specs to the repo, and let agents implement and verify changes autonomously.
Connected Companies & Entities
3 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Notion Launches Custom Agents, Builds Agent-Native Workflow
Notion detailed its multi-year effort to build Custom Agents and an agent-centric product architecture in a Latent Space podcast featuring Sarah Sachs and Simon Last. The company rebuilt its agent harness multiple times before shipping, citing early model limitations, context-window constraints, permissioning complexity, and tooling standards as reasons for repeated iterations. Notion described product and org choices: credits-based usage pricing that abstracts tokens/model/serving tiers, an eval-first approach with dedicated Model Behavior Engineers, progressive tool disclosure, and agent composition using Notion pages and databases as primitives. The discussion covered MCP vs CLI trade-offs, retrieval and ranking investments, Meeting Notes as high-signal data capture, and the longer-term vision of “software factories” where agents spec, code, test and maintain systems collaboratively. Notion said it prefers fine-tuning and retrieval investments over building a foundation model in-house.
Notion launches developer platform for AI agents
Notion announced a Developer Platform that turns its workspace into an orchestration hub for AI agents. The release adds Workers — a cloud sandbox for running custom code and webhooks — plus database sync that can pull data from any API-backed database (examples: Salesforce, Zendesk, Postgres). The platform enables teams to connect external agents, chat with and assign work to them, and exposes an External Agent API for internal agent integrations. Notion says customers have built over one million Custom Agents since launching them in February. The Notion CLI will provide developer access on Business and Enterprise plans; Workers usage is free through August to encourage experimentation. Ivan Zhao, Notion co-founder and CEO, framed the shift as making Notion a programmable platform and orchestration layer for people, agents and tools.
How I AI: Sendbird & Notion on Internal AI Workflows
This newsletter summarizes two podcast interviews about internal AI adoption and developer workflows. Sendbird CEO John Kim describes a gamified internal marketplace called "Automators" that lets non-engineers request automations, token-usage tracking with tiers (up to an "AI God" >100M tokens/day), marketer-built Stripe-integrated swag commerce, secure vetted templates, and a cross-functional AI task force. Notion engineering manager Ryan Nystrom outlines Notion AI practices including auto-generated standup notes, spec-driven development as the source of truth for agents, a Boxy system that ships PRs from Notion mentions, and an effort to drastically speed up CI to unlock agent productivity. The newsletter also notes a breakdown of Anthropic’s "Code with Claude" announcements.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.