Notion's Token Town — Sarah Sachs, Notion

2026-07-23 · Source: AI Engineer · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Advanced, extended

Summary

Notion's AI engineering lead, Sarah Sachs, outlines the company's strategy for building sustainable AI-native products amidst a volatile market characterized by escalating costs and vendor lock-in. Notion observes that 88% of companies struggle beyond AI as an assistant due to siloed data. To counter this, Notion advocates for a model-agnostic approach, where an "auto model" handles approximately 75% of traffic, allowing flexible switching between state-of-the-art and open-weight models. The strategy emphasizes evaluating "cost per capability per second" over raw token pricing, leveraging open-weight models for moderate tasks, utilizing CPUs for deterministic jobs, and implementing robust AI governance. The presentation also highlights the critical role of secure, multi-agent orchestration for future "software factories."

Key takeaway

For AI Architects and Directors of AI/ML building AI-native products, prioritize a multi-model, vendor-agnostic strategy to mitigate escalating costs and avoid vendor lock-in. Focus on understanding your specific "cost per capability per second" needs for different tasks, leveraging open-weight models and CPU-based solutions where appropriate. This approach ensures long-term sustainability and provides the necessary leverage to navigate the opaque and rapidly changing AI vendor landscape.

Key insights

Sustainable AI product development requires vendor optionality and deep understanding of cost-capability trade-offs.

Principles

• Your AI supplier is often a competitor.
• Optionality provides crucial negotiation leverage.
• Model pricing doesn't always reflect capability growth.

Method

Implement a model-agnostic playbook by building for multimodal interoperability, evaluating "cost per capability per second," switching models frequently, and offering use-case expertise to labs.

In practice

• Utilize open-weight models for suitable tasks.
• Prefer CPUs over GPUs for deterministic code execution.
• Establish AI governance for data visibility and control.

Topics

• AI Cost Management
• Vendor Lock-in
• Model Agnosticism
• Open-Weight Models
• Agent Orchestration
• AI Governance

Best for: CTO, VP of Engineering/Data, Machine Learning Engineer, AI Engineer, Director of AI/ML, AI Architect

Related on AIssential

• Locked, stocked, and losing budget: AI vendor lock-in bites back — AI vendor lock-in and rising costs are challenging enterprise AI adoption, debunking executive assumptions about easy model switching.
• SAP's second quarter: The AI takeaways — SAP is aggressively embedding AI into its enterprise offerings and internal operations, prioritizing efficiency and cost control.
• MLWhiz Weekly AI/ML/Recsys Newsletter # 5 — AI model commoditization is driving vendors to build service moats and compete in deployment.
• How Open Source AI Is Shifting Power Back to the Buyer — Open source AI orchestration layers are shifting enterprise procurement power by enabling modularity and cost control.
• The Regime-Agnostic Enterprise — The winning-coalition question is the wrong question for buyers; architect for regime-agnosticism to avoid lock-in.
• Open Models Will Absorb Most of the AI Spend — Open models offer superior economics, control, and resilience, driving a shift in AI spend away from proprietary frontier models.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Engineer.