Two days, built for engineers.

AI Tokenomics: Principles for Cost-Efficient GenAI

Grzegorz Wasilewski

The explosive adoption of LLMs has introduced an unprecedented challenge for software architects: Tokenomics. Even fully functional GenAI deployments can turn into business failures if they burn budgets on millions of redundant input (Token IN) and output (Token OUT) tokens, destroying their return on investment (ROI). In this session, you will learn where these costs originate and how they can be consciously controlled from the architectural design phase. Drawing on Google's 11 Principles for Token-Efficient Software Engineering and the Accenture Tokenomics framework, I will demonstrate whether and how these principles can actually be implemented in practice. After a brief theoretical introduction, we will ditch "death by PowerPoint" in favor of pure live coding. Using a free notebook environment and free access to models, we will walk through all 11 principles together, examining whether they genuinely make sense in practice and how I personally interpreted them. The goal is not to take token optimization at face value, but to put it to the test: is it just another empty buzzword, or is there genuine truth to it that makes real financial and architectural sense? Attendees will receive access to a private repository containing the notebook and example code from the session. I hope the Q&A will spark a vibrant discussion about your own experiences with Tokenomics and the real cost of burned tokens in production.