Sanders argues that AI demands purpose-built infrastructure, not retrofitted legacy systems. He details how training requires highly interconnected, GPU-optimized setups, while inference is becoming central to agentic applications. CoreWeave's ARIA tool automates experiment analysis, moving researchers away from manual chart comparisons toward agent-driven interactions. The future, he predicts, will see AI-first experiences replace traditional websites and apps within five to seven years. Sanders emphasizes that enterprises must build dedicated AI engineering teams, which will eventually surpass their cloud teams in size and importance. He also discusses CoreWeave's strategy of balancing open-source portability with deep performance optimization, exemplified by their Sunk (SLURM on Kubernetes) tool. The key challenge is democratizing AI infrastructure so that companies without frontier lab talent can still innovate effectively.