Harsh Varshney, a machine learning software engineer at Google who joined the company two years ago after roles at Splunk and Amazon Web Services, describes a rapid shift in Big Tech toward faster execution and direct contribution. The traditional long planning cycles and multiweek sprints have given way to greater agility and results orientation. Generative Artificial Intelligence is now a core competency across teams, shaping priorities and redefining what it means to deliver impact.
Varshney says the explosion of generative Artificial Intelligence has changed both the nature of products and the rules for using data. Tasks that once took days are now compressing to hours because engineers use AI tools to debug, experiment, and optimize. Demand for expertise in machine learning systems and emerging areas such as agentic AI systems has increased, and building intelligent, fair, ethical, and secure systems is now central to software development. He advises engineers to build a strong foundation in core computer science areas — computer systems, distributed systems, software architecture, machine learning, Artificial Intelligence, and cloud systems — and to stay engaged with academic research, conferences such as NeurIPS, and influential researchers like Andrej Karpathy.
Practical experience matters. Varshney urges engineers to go deep on Artificial Intelligence both as builders and as power users: understand the full stack behind models, multimodal systems, and move beyond simply calling an API. Be proficient with AI tools for code generation, debugging, and rapid prototyping. Most importantly, turn knowledge into tangible projects that demonstrate the ability to ship end-to-end products. As an example, Varshney built a multi-agent deep research agent using open source frameworks such as LangGraph, with a supervisor agent delegating subquestions to researcher agents and synthesizing results. He stresses that building and launching complex AI systems is the clearest proof of capability. The views expressed are his own and not Google’s.
