Suggest First, Act Second: GreytHR's Senior Director of Engineering on Climbing the Trust Ladder

This piece draws from "Deep-Dive 2: From Copilot to Autopilot, How Organizations Climb the Trust Ladder," a whole-room discussion at AISRENext Bengaluru moderated by Sudheer Bhat, Principal Engineer at InMobi. 

Sudheer opened with a question for the room: where does each team's SRE function sit right now, treating AI as an assistant, letting it suggest and wait for approval, or running closer to full autonomy? 

Abhishek Gaurav, Senior Director of Engineering at GreytHR, answered from a company that has pushed AI hard into nearly every part of the business, but kept one part of it on a shorter leash. 

AI Runs Through Nearly Every Team, SRE Included 

GreytHR's HRMS platform holds employee salary data, one of the more sensitive categories of information any system can touch. 

"We use AI as part of how we work across teams. Developers use it heavily in the coding process, mostly through Cursor. Our product team uses Claude to support decision-making and presentations. Our marketing team uses AI to help produce the videos you see on our LinkedIn. And finally, for SRE, yes, we use Aiden by StackGen." 

AI touches code, product decisions, marketing output, and SRE operations at GreytHR. On the specific question of autonomy in SRE, Abhishek's answer was narrower than the rest of that list. 

"We are not at autopilot, I would say, but we are in the co-pilot mode, where AI suggests, and it also makes decisions after approval." 

Where GreytHR Already Lets AI Act Without Waiting 

Three specific behaviors sit inside that co-pilot label:

• Developers connected an MCP server directly to Cursor, giving it access to production observability data. When a production issue gets reported, a developer asks Cursor to pull the relevant logs and metrics and propose a fix, with observability context already built in.
• The SRE agent generates weekly incident reports on its own, in three versions: one for leadership, one for developers, and one for the SRE team, replacing a documentation task most teams handle manually.
• For pods that reliably enter a known hung state, the agent restarts them without waiting for approval. The failure mode is understood well enough that a human check adds no value.

"So it's again assisted and executing something with approval, a human in the loop. And, we need to climb the trust ladder, that's for sure." 

The Principles Don't Change, Even as Access Does 

"I think the first principles remain the same. Nothing changes with or without AI. That's why we have not rolled out that product yet for all employees. It's only available for admins to use right now. We will be rolling it out for everyone very soon, and yes, we have all the guardrails in place." 

How Access Widened Gradually Across GreytHR's AI Rollout 

Every other part of GreytHR's business moved fast on AI: most code, product decisions, marketing content. The SRE function moved just as aggressively toward AI assistance, with access tied to specific, well-understood categories of risk at each stage.

The same AI capability that makes an internal support request faster, someone asking an assistant to handle a routine task on their behalf, is exactly what surfaces the boundaries a team has to hold, regardless of how capable the underlying model gets. Climbing the trust ladder, in Abhishek's account, means building out access carefully, in step with what a team can stand behind.

GreytHR runs Aiden in production using the same staged model: automated where the outcome is predictable, human-approved where it isn't, and guardrailed everywhere sensitive data is involved. Try Aiden for SRE Community Edition for free today!