Laurent Gil is president and cofounder of Cast AI. He works on cloud and AI cost management and presents the company’s Kimchi coding platform, including the economics of choosing models for a task.[1][2]
Start with Stop Rationing Tokens: Let the Harness Pick the Model →
Career
Contributions
Explain an adaptive coding platform
Gil presents Kimchi with engineer Žilvinas, who explains the implementation. Their account places model choice, long-running execution and evaluation in the harness, preserving the workflow while changing model providers.[2]
Publish the harness behind the product
Cast AI releases Kimchi’s code alongside its hosted platform. Gil represents the product as the company’s cofounder and president, while the open repository provides the engineering implementation and documentation. This makes the model-selection and workflow approach available for inspection and adaptation beyond a presentation.[1][4][5]
Ideas
In his token-cost article and conference presentation, Gil argues that the price of an individual token can be a misleading measure of agent economics. A lower-priced model can require more attempts or produce work that needs repair. He frames cost around an achieved task and the level of developer use a company can support, rather than treating a token allowance as the outcome.[2][3]
Tools, writing & talks
Writing, research & talks
Start here
Gil and Žilvinas explain task economics and the Kimchi coding harness.
Begin with the comparison between token price and cost per completed task. The talk explains the product motivation before its architecture; running the open harness requires its documented developer environment and model-provider setup.
Co-presenter[2]Cast AI's team explains Kimchi, an open-source coding harness that picks models by task outcome and cost per task. They share internal results from 300 employees and demo Ferment, Teleport and Studio for long-running agent work.
Catalog workOriginal sourceAn argument for treating AI spending as an operational infrastructure problem.
Author[3]