WeMakeDevs Founder Kunal Kushwaha: As AI Takes on More Implementation Work, Engineers Are Advised to Run Large Models Locally to Deepen Technical Understanding
In an interview with CodeRabbit's podcast "The Merge," WeMakeDevs founder Kunal Kushwaha proposed that as AI coding agents take on more implementation work, the core value of engineers shifts to understanding problems, selecting solutions, and explaining the reasoning behind decisions. He also suggested a practical direction: running large language models locally to access the infrastructure behind the interfaces—how models run, how components communicate, and what deployment choices are involved—to deepen technical understanding.
If you ask a developer about their workflow, they can tell you how much faster the latest model is, what their coding agent can now handle, and which new tool has improved the way they work. Kunal Kushwaha has a follow-up question.
"What about you?"
Kushwaha’s answer starts with the work that still belongs to the engineer. He believes that regardless of the tools used, engineers need to understand the problem well enough to choose an approach and explain the reasoning behind their decisions; they remain responsible for what ships. As AI takes on more implementation work, where does this judgment come from? He also pointed to a practical direction: learning to run large language models locally.
The purpose of this exercise is not to replace API calls, but to expose the infrastructure behind the interface. Kushwaha said that running a model locally and then exploring fine-tuning and what it would take to run at scale allows engineers to access how the model runs, how its components communicate, and the deployment choices involved. He compared it to learning Kubernetes: an engineer may use managed services in production, but still benefits from having built and operated a small cluster themselves—that experience helps them understand what the managed service is doing on their behalf. The same principle applies when a team chooses how to deploy models: APIs, company-owned hardware, and managed platforms each have different tradeoffs, and hands-on local setup provides more context to ask better questions.