The org chart of a typical Indian product company in 2022 had a familiar shape. Engineers built. Designers shaped. The product manager held the centre, owning the roadmap, the prioritisation, the cross-functional translation, and the credit. The PM role was the highest-leverage non-founder role in the company, because shipping the right thing was the rate-limiting step. Hiring a great PM made everything around them better.
That shape is changing, and most Indian companies have not noticed yet. In an AI-native product, shipping is no longer the rate-limiting step. The model ships features faster than the team can think of them. The new rate-limiting step is staying right, keeping the system's behavior aligned with what the user, the business, and the regulator expect, across an input distribution that nobody fully controls. The person who owns that is the reliability engineer, and in five years they will be the most powerful individual contributor in the building.
What This Role Actually Does
The reliability engineer is not a tester. They are not a site reliability engineer in the classical Google sense. They are not a QA lead. They sit at the intersection of all three and pull from each, but the work is genuinely new.
On a Tuesday, a reliability engineer might be designing a stratified evaluation set for a customer support agent that needs to handle queries in seven Indian languages. On a Wednesday, they might be reviewing a model upgrade proposal and demanding the regression suite be re-run on a 50,000-example corpus before they sign off. On a Thursday, they might be sitting with a fraud-ops analyst at an Indian fintech, watching her review false-positive alerts, and quietly building a new behavioral test case from what she shows them. On a Friday, they might be writing a memo to leadership explaining why a 1.5 percent improvement on the academic benchmark masks a 4 percent regression on the user-segment that actually matters.
None of this work produces a feature demo. Most of it produces, at best, a graph that did not get worse. The output of reliability engineering is the absence of disaster, which is a thing that human organisations are spectacularly bad at valuing.
Why the Asymmetry Inverts
Right now, the reliability engineer is paid less than the senior backend engineer and reports to the PM. This will not last. The asymmetry inverts the moment the cost of a confidently wrong AI output crosses the cost of a delayed feature, and for most serious Indian AI deployments, that crossover has already happened. It happened the day the RBI started asking pointed questions about model governance in credit decisioning. It happened the day a state health department asked a hospital system how it validated the AI-generated discharge summaries. It happened the day the first class-action whisper started circulating in legal circles about AI-assisted insurance claim denials.
When the regulator asks how you know your system is safe, the PM cannot answer. The engineer cannot answer either, not in a way that satisfies a regulator. Only the reliability engineer can, because only they have done the work of measurement. At that point, their compensation, their influence, and their veto power all rise together.
The Indian Talent Advantage, Again
India has, by accident of history, trained the world's largest pool of people who are skilled at the disciplined, systematic, slightly-thankless work that reliability engineering requires. The same instincts that made the Indian QA workforce indispensable to the global software industry are the substrate from which AI reliability engineers will emerge. The transition is not automatic, it requires upskilling in ML fundamentals, statistical reasoning, and the particular weirdness of probabilistic systems, but the human substrate is here in greater density than anywhere else.
The Indian eval labs beginning to form, often spinning out of academic groups in IISc, IIIT-Hyderabad, and the IITs, or from veterans of the larger product companies, are the early formal infrastructure for this discipline. The independent evaluators-for-hire who have started showing up at AI safety meetups in the last year are the early market for it. Both are still small. Both will be very large within a decade.
What Reliability Engineers Need That They Do Not Yet Have
The discipline is missing three things: standard tooling, recognised credentials, and seats at the architecture table. Tooling is coming, slowly, from a mix of open-source eval frameworks and Indian-built platforms that are still in their first or second year. Credentials are nascent, there is no equivalent yet of the CFA for evaluation engineering, and there should be. Seats at the table are the hardest, because they are political, not technical. The reliability engineer has to be in the room when the architecture is decided, not summoned later to test what was already built.
The Action
If you are an engineer in India and you want to bet on a role whose value will compound for the next decade, become the reliability engineer at your company. If you are a founder, hire one before you hire your fifth product manager. If you are a leader, make sure your reliability engineer has the authority to block a deploy. The role is undervalued today because its work is invisible. The work will not stay invisible for long, because the failures it prevents will not stay invisible for long.
Join the conversation
This essay is part of an ongoing community. If it resonated, the next step is to be in the room.
Join Bharath.club → Read more essays