A negotiating strategy that leaders use is to ask, "What would it take?" In light of the weak predictive biomarkers used for immune checkpoint inhibitor (ICI) therapy, it may be time to ask the pathologist "What would it take to provide us with a quantitative tumor infiltrating lymphocyte (TIL) score for each patient with non-smallcell lung cancer?" In this issue of JAMA Oncology, Rakaee et al 1 describe a new biomarker associated with response to ICI based on an objective, open-source, machine learning (ML) method using QuPath software to count TILs in lung cancer. They note that TIL has potential to improve selection of responders to ICI and that ML-TIL is easily obtained, accurate, and reproducible. They created an ML-TIL algorithm and then tested it on a training set to show how TIL count is associated with response to programmed cell death (PD) 1/PD ligand 1 axis singleagent ICI therapy. This was shown in 2 large, retrospective nonsmall-cell lung cancer cohorts in the training set/validation set format. They also showed that the ML-TIL is particularly valuable in PD ligand 1-negative patients and has a stronger association with outcomes than tumor mutation burden.This is an exciting result in that, to my knowledge, it is the best evidence to date that TILs are tightly associated with benefit from ICI. While TILs can be scored by pathologists, 2 it is probably more accurate (and certainly more reproducible) to have this task performed by a machine. Machine scoring of TILs in lung cancer is not novel or unprecedented. 3 But to my knowledge, this is the first effort to show the association with outcome on ICI therapy using an open-source approach. While the work claims predictive value, to be statistically rigorous, prediction requires a test for interaction with an untreated trial arm, which is not present in this retrospective study. However, prediction may never be achievable for any future ICI biomarker because it would be unethical to randomize patients to studies in which they would receive a placebo instead of an ICI. Perhaps this is why the authors were allowed to use the term predicts.However, introduction of an ML-based method creates discomfort. The assay is not approved by the US Food and Drug Administration (FDA), and because it uses open-source software, it is unlikely to ever be submitted for regulatory agency approval. However, I would argue that the use of open-