.png){kind=logo}
Most of the public conversation about artificial intelligence is about the models: the chatbots that write essays, the systems that generate images, the race between labs to build something bigger. Far less is said about the machinery that keeps those systems standing up under real-world load: the pipelines, dashboards, automated checks and reliability systems that decide whether an AI product actually works when millions of people use it at once.
Soumya Trivedi has spent her career on that less glamorous side of the industry, and she argues it is where the AI race will quietly be won or lost. A software engineer at Dell Technologies who previously worked as a technical program manager on artificial-intelligence teams at Meta, Trivedi specialises in the infrastructure and automation that sits beneath machine-learning products rather than the models themselves.
"Everyone wants to talk about the model. Almost nobody wants to talk about what happens the day after the model ships," she said. "Reliability, monitoring, the ability to catch a problem before a customer does; that is the unglamorous work that determines whether an AI system is trustworthy or just a demo."
Her recent work underscores that point. By building internal engineering platforms and automating routine checks, Trivedi has delivered clear operational gains: strengthening reliability, reducing downtime, and easing the manual burden on engineering teams. These improvements rarely make headlines, but they compound quietly inside large organisations, where every preventable outage and recurring workflow bottleneck carries a real cost.
She believes one of the biggest opportunities for engineering organizations is to treat internal platforms with the same care and discipline as customer-facing products. The engineers who rely on these platforms are, in many ways, the platform’s customers, and when they have better tools, they can build better products for everyone else.
“The people using your platform are your engineers, and they deserve the same thoughtfulness a paying customer gets,” she said. “That means investing in clear roadmaps, measuring outcomes instead of just technical output, and asking whether you’ve genuinely made developers more productive. When you improve the developer experience, you’re ultimately improving what the organization delivers to its customers.”
If that sounds like the language of a product manager as much as an engineer, that is by design. Trivedi's career has straddled both. She moved from a program-management role in AI infrastructure into hands-on software engineering, and she believes that dual fluency is becoming essential as AI systems grow more complex. "The gap between the person who can write the code and the person who can align five teams around why it matters is where a lot of good ideas die," she said.
Her prescription is deliberately simple, almost a checklist: automate the repetitive work, standardise the patterns that succeed, measure outcomes with data rather than opinion, and scale through shared platforms instead of one-off manual processes. It is a philosophy she summarises in a single line she is fond of repeating to younger engineers: "Automation is not a shortcut, it is a discipline."
Trivedi is convinced the industry's centre of gravity is already shifting in this direction. The next wave of AI progress, she predicts, will come "not only from larger models, but from better operational systems: the reliability, the governance, the developer productivity that lets teams actually deploy safely."
