World's Fair 2026
Notion's Token Town — Sarah Sachs, Notion
Overview
This talk addresses the significant cost challenges in AI development, particularly the escalating expense of using large language models. It argues that focusing solely on token economics is a losing strategy, as model providers often increase prices or deprecate older versions. Instead, the core thesis is to win on product by leveraging data flywheels, sophisticated orchestration, and maintaining model agnosticism to preserve flexibility and avoid vendor lock-in.
Who should watch
- AI Engineers grappling with rising AI costs.
- Product Managers evaluating AI integration strategies.
- Builders seeking to optimize AI model usage and vendor relationships.
- Teams looking to implement effective AI orchestration and agentic workflows.
- Anyone concerned about vendor lock-in with AI model providers.
Key takeaways
- AI model providers often increase token costs or deprecate predecessors, making token economics a precarious strategy.
- Treating AI model vendors as competitors is crucial for negotiating favorable terms and avoiding markups.
- Winning in AI product development relies on building strong data flywheels and effective orchestration, not just on model performance.
- Maintaining model agnosticism is key to preserving leverage and ensuring flexibility in AI strategy.
- Not all AI traffic requires frontier models; route tasks to appropriate models based on capability and cost per second.
- Open-weight models can be effectively utilized for moderate workloads, optimizing cost and performance.
- CPUs can be more efficient than GPUs for tasks that do not require complex LLM processing, such as data transformation.
- A robust security strategy is essential, particularly when dealing with the "lethal trifecta" of AI risks.
Notable quotes
*Your supplier is your competitor.*
*Win on product, not token economics.*
*Optionality is your leverage.*
Unofficial community note. Prefer the recording for nuance.