There is a meaningful difference between an AI that advises an operator and an AI that acts on the infrastructure itself. The first suggests a fix; a human applies it. The second restarts the service, resizes the cluster, reroutes the traffic and closes the ticket — on its own. Agentic CloudOps is the second kind: agentic AI cloud operations, where the system acts rather than advises. And it is arriving quickly. The productivity case is genuine and we are not here to talk anyone out of it. But at L&T Vyoma, our position is blunt: you can automate the action, you cannot automate the accountability. The moment an agent can change production, someone still has to own what it does — and that ownership has to be designed in, not assumed.
The Shift Is Real, and It Is Fast
This is not a horizon technology. Gartner's Predicts 2026 research on infrastructure operations forecasts that 70% of enterprises will deploy agentic AI as part of IT infrastructure operations by 2029, up from less than 5% in 2025. That is one of the steepest adoption curves the operations world has seen, and it is being driven by a real pain: modern estates generate more signals, more change and more scale than human teams can keep pace with. This is the step beyond traditional AIOps and basic cloud automation services: from flagging an incident to autonomous IT operations that resolve it. An agent that can triage an alert, correlate it across systems and remediate it in seconds is not a gimmick — for the right class of problem, it is simply better than waiting for a human to wake up. The direction of travel is settled. The question is what you let these agents stand on.
And Most of It Will Fail
The same analyst house that sees 70% adoption also expects the bodies to pile up. Gartner forecasts that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. The hype is worse than the failure rate suggests: of the thousands of vendors now claiming agentic capabilities, Gartner estimates only about 130 genuinely offer them — a phenomenon it calls “agent washing.” Strip the marketing away and a pattern emerges. The projects that fail are the ones that bolt an autonomous agent onto an environment that was never governed, observable or permissioned in the first place. Autonomy did not cause the failure. The absence of a foundation did.
The Question Automation Cannot Answer: Who's Accountable?
When an agent takes an action in production and it goes wrong — a bad scaling decision, a wrongful shutdown, a change that cascades — the incident review asks the same questions it always has. What changed, who authorised it, what was the blast radius, and how do we roll it back? An agent does not relieve you of a single one of those questions; it makes them harder, because the actor is now software operating at machine speed. Accountability cannot be delegated to the thing taking the action. It stays with the organisation, which means the architecture has to make every agent action attributable, bounded and reversible. An agent with standing production privileges and no audit trail is not an efficiency. It is an unnamed operator with root access, and no enterprise would hire one of those.
Autonomy Is Only as Safe as the Floor It Stands On
Our view is that agentic operations are a foundation problem before they are an AI problem. Four things have to be true underneath the agent before you give it the keys. It must operate under least privilege, scoped to exactly what its task requires and nothing more. Every action it takes must be observable and logged, so the audit trail is complete by default rather than reconstructed after an incident. It must run inside policy guardrails that define what it may never do without a human, regardless of what it concludes. And every action must be reversible, with a tested rollback path. Together, those four controls are what agentic AI governance means in practice. This is the layer we build and operate. L&T Vyoma provides the governed, observable, permissioned platform on which agentic tooling can run safely, and our managed services team remains the accountable human owner of the estate — not a label in a dashboard, but the party that answers for what runs. We let the agents do the work. We do not let them absorb the responsibility.
|
Dimension |
Agent Without a Floor |
Agent on a Governed Platform |
|
Privileges |
Broad, standing access |
Least privilege, scoped to the task |
|
Visibility |
Actions reconstructed after the fact |
Observable and logged by default |
|
Limits |
Acts on whatever it concludes |
Policy guardrails with human gates |
|
Reversibility |
Hope it can be undone |
Tested rollback path |
|
Accountability |
Diffused into the tool |
Owned by a named operations team |
Let the Agents Run — Keep the Accountability
Agentic CloudOps is going to run a growing share of enterprise infrastructure, and that is, on balance, a good thing. The enterprises that get value from it will not be the ones that adopted the cleverest agent. They will be the ones that built the floor first — governance, observability, permissioning and a clear human owner — and then let autonomy operate on top of it. That is the order of operations we would argue for, and the one we are structured to deliver on Indian soil. Talk to our team about running agentic operations on a governed, sovereign cloud in India, or explore our managed cloud services in India — cloud automation services, cloud infrastructure management, observability and managed multi-cloud services for enterprises.
Sources: agentic AI adoption in infrastructure operations — Gartner, “Predicts 2026: AI Agents Will Reshape Infrastructure Operations” (70% of enterprises by 2029, up from under 5% in 2025); project-cancellation forecast and “agent washing” — Gartner press release, 27 June 2025 (over 40% of agentic AI projects cancelled by end of 2027; ~130 of thousands of vendors genuinely agentic).
