The economics of agent scale: tokens, ROI, and building platforms for AI-first teams (Part 2)
Andi Gutmans, head of Agentic Data Cloud at Google, returns for the second half of his Leaders of Code conversation to talk through the cost and infrastructure side of agentic development. ICYMI, part one covered judgment, code review, and data activation.
In this continuation of their conversation, Peter O'Connor and Andi Gutmans delve into the economics of scaling AI agents. O'Connor believes that the model itself is no longer the primary bottleneck; rather, the challenge lies in determining the minimal context required to achieve reliable outcomes at the lowest cost. He emphasizes that token efficiency is becoming increasingly important, and the goal should not be to "token max" but rather to minimize token processing while maintaining accuracy.
Gutmans agrees that models need to keep improving, but for many tasks, existing models are already sufficient. He thinks that the future will involve fine-tuning the selection of models for specific use cases, optimizing the context to reduce unnecessary token usage, and achieving more accurate reasoning paths. Both agree that data plays a crucial role in this process, but it's not just about curating large datasets; it's about understanding the information and its relevance to the agent's task.
Gutmans highlights that Google's advantage lies in its dual expertise in model development and data platforms, enabling them to collaborate towards optimizing agent performance. He stresses that companies should adopt an AI-first approach, focusing on the agent's needs rather than merely enhancing human-built ontologies.
Written by urgent.news from Stack Overflow Blog's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
