# Notion's Token Town — Sarah Sachs, Notion

## Executive summary

The presentation argues that building sustainable AI-native products requires shifting focus from optimizing token economics to mastering product architecture and optionality. The speaker warns against vendor lock-in due to volatile model pricing (the 'token town' trap) and advocates for strategies like implementing an 'auto model,' leveraging open weight models, prioritizing CPUs over GPUs for deterministic tasks, and building robust multi-agent orchestration systems.

## Key takeaways

- Vendor Lock-In is the Primary Risk: Relying solely on a single AI provider creates significant risk because pricing structures are volatile. The speaker asserts that 'your supplier is your competitor' (7:28), making model agnosticism crucial for business viability.
- Win on Product, Not Tokens: Instead of competing on the lowest cost per token, companies must build data flywheels and orchestration layers that solve unique customer problems. The value should come from the product's workflow, not just its underlying model capability (8:44).
- Implement Model Agnosticism: To maintain optionality, systems must be designed to route traffic across multiple models and providers (e.g., Notion’s 'auto model' handling 75% of traffic) to mitigate pricing shocks or provider deprecations (13:25).
- Prioritize CPUs for Deterministic Tasks: For tasks that do not require complex reasoning—such as turning a CSV into a PDF, running deterministic SQL queries, or simple tool calls—CPUs are often more cost-effective and efficient than relying on GPUs/LLMs (17:00).

## Technical details

- Model Agnosticism & Auto Model: The concept of an 'auto model' involves building a system that automatically routes traffic across multiple models and providers, ensuring continuous service even if one vendor raises prices or deprecates a version (13:25).
- Open Weight Models: These models are recommended for moderate tasks as they lower the barrier to entry on cost and provide negotiation leverage against oligopolistic frontier model providers. The speaker notes that the gap between open weight and proprietary models is expected to close (15:04).
- System Architecture & Governance: The 'lethal trifecta' risk involves combining access to private data, exposure to untrusted content (ingestion/email), and the ability for external communication. Mitigating this requires robust sandboxing and governance (19:00).
- Cost Optimization Metric: The focus should shift from 'cost per token' to 'cost per capability per second,' which accounts for entire workflow trajectories rather than single API calls (13:25).

## Practical implications

- Design systems to be model-agnostic by implementing an auto-routing layer.
- Shift architectural focus from maximizing LLM capability to optimizing workflow orchestration and data handling (data flywheels).
- For non-reasoning tasks, evaluate CPU usage over GPU/LLM services for cost efficiency.
- Treat AI vendors as competitors rather than sole suppliers to maintain negotiation leverage.

## Topics

AI Architecture, Vendor Lock-in, Token Economics, Model Agnosticism, System Design, Notion AI, Opus, Claude Agent, Decagon Agents

Source: https://www.youtube.com/watch?v=-I5W5QVAT8E