India’s AI moment needs more than momentum

By Anupam Kumar Jha, Head of Solutions Engineering – India, Datadog.

Building trust at scale requires achieving transparency on a large scale. That was the founding decision behind UPI: every transaction visible, traceable, and accountable in real time. Not because regulators demanded it, but because the architects understood that confidence cannot be declared — it has to be demonstrated, one transaction at a time. The result is a payments infrastructure processing over 20 billion transactions a month, built entirely on earned trust.

The question facing India’s enterprise leaders today is whether we will apply the same discipline to AI.

56% of Indian organisations have already embedded AI into core business processes, yet only 25% believe their workforce is adequately prepared to use it, a gap that has widened by 12% in a single year. Meanwhile, 81% of Indian business leaders are concerned that AI advancement is already outpacing their workforce capabilities, governance frameworks, and operating models. As organisations move past the ‘why’ and ‘how’ of AI to the ‘what’ — what it is doing, what it is actually producing, and who is accountable when it goes wrong — the real work begins. Ambition is not the constraint. Accountability is.

The organisations that are getting the most from AI are not necessarily the ones deploying it the most. They are the ones that can see what their systems are doing, secure them against drift and failure, and act on evidence rather than assumption. That capability is what turns AI activity into accountable, scalable value. Without it, AI spending is an act of faith. With it, it becomes a manageable, measurable investment.

The productivity prize and the accountability gap

Over 80% of Indian knowledge workers report using AI at work regularly. Randstad’s 2026 Workmonitor survey found 89% say AI improves their productivity. By every surface metric, India’s workforce has embraced AI. And yet India registered the sharpest rise in AI concern of any country surveyed.

The stakes of getting accountability and governance right in AI are not abstract. India’s AI story is one of staggering ambition meeting the hard reality of uneven execution. The numbers are extraordinary: AI adoption is projected to transform 38 million jobs by 2030, contributing a 2.61% uplift to national productivity, and could add nearly $500 billion to India’s GDP. These are not incremental gains. They are structural shifts.

But productivity gains only materialise when AI systems are observable, governable, and accountable. Without that infrastructure, AI doesn’t eliminate inefficiency — it hides it. Realising this prize requires a deliberate shift: from measuring how much AI is deployed, to measuring how well it is performing.

A framework is not a monitoring stack

India’s policymakers have moved in the right direction with the AI Governance Guidelines, built around a single directive — Do No Harm — and set the right tone: voluntary standards, regulatory sandboxes, and human accountability at the centre. But that’s not enough.

Most of what is running in Indian enterprise AI today – credit models, hiring algorithms, customer-facing LLM agents – is live in production without the observability infrastructure to detect when something goes wrong. And the failures that matter most don’t throw error codes. Hallucinations, output drift, and context contamination: these degrade the quality of real decisions silently at scale.

In practice, observability works by capturing every interaction in an AI pipeline as a structured record: what prompt was sent, what context was retrieved, what the model returned, how long it took, and what it cost. These records — called traces — give engineering and business teams alike a continuous view of system health. When a model’s output quality begins to slip, when costs spike unexpectedly, or when a particular query type starts producing inconsistent results, the observability layer surfaces it before it becomes a business problem rather than after.

For a senior leader trying to measure ROI from AI investments, this is the layer that finally makes it possible. It shifts the conversation from “we believe AI is working” to “here is evidence of where it is working, where it is not, and what needs to change.”

Accountability builds trust

As models evolve and new tools are introduced, agents adapt their behaviour, subtle distortions accumulate, and eventually produce conspicuous changes that no one saw coming. Observability addresses this through three concrete capabilities.

Distributed tracing maps every step an agent takes across tools, systems, and data sources, so anomalous behaviour is caught at the source. Continuous evaluation runs automated quality checks against defined benchmarks, alerting teams when outputs fall below acceptable thresholds. Drift detection monitors whether a model’s behaviour is shifting over time, even when individual outputs still appear normal.

Together, these capabilities answer the question that matters most before scaling any AI system: has it earned the right to more autonomy? The organisations that build proper instrumentation first – traces, evals, quality-aware alerts, and drift detection across the full AI lifecycle – are the ones that will earn the right to scale. Everyone else is flying blind and calling it momentum.

Trust what you can see

India ranks fourth globally in AI performance. The talent is unmatched. The enterprise appetite is genuine. None of that is the constraint.

What separates leaders from laggards now is the quality of the feedback loop — the ability to see what AI systems are doing, catch where they are failing, and intervene before errors compound into decisions that matter. In practical terms, that means investing in evaluation frameworks, not just deployment pipelines. It means making model behaviour auditable by design. It means treating governance infrastructure with the same rigour as the AI systems it is meant to govern.

We didn’t trust UPI because it existed. We trusted it because every rupee was visible, every failure flagged, and every transaction accountable. India’s AI systems deserve the same standard.

The route is already live. The question is whether the systems guiding us have truly earned the right to navigate.