BlogMar 12, 20265 min read

When AI Writes the Code, Who Controls the Agents?

 

AI coding assistants and agents are accelerating development dramatically. Engineers can produce far more code, far more quickly. But traditional review and control systems were not designed to handle this pace.

And recent reports reveal a new, uncomfortable reality for the industry.

Amazon experienced multiple outages tied to software changes. In one incident, incorrect delivery estimates triggered 120,000 lost orders and 1.6 million errors. In another, a configuration change caused a 99% drop in North American orders.

The internal takeaway wasn’t surprising: velocity has outpaced governance—and the gap is starting to show up in production.

 

The Real Challenge Isn’t Building Agents—It’s Managing Them

Building an AI agent is easier than ever. You can spin up an agent using LangChain, Copilot Studio, or direct LLM APIs in days.

But getting that agent safely into production is another story entirely.

Indeed, agents introduce a new category of risk.They operate autonomously as non-human identities with access to enterprise systems and data. And unlike traditional software, large language models are non-deterministic. Even well-designed agents can behave unpredictably.

The result is a familiar pattern across enterprises today: Development teams move fast, experimenting with AI but Governance teams slow everything down trying to understand and control new risk. The result? Innovation stalls somewhere between the two organizations.

What the AI Agent era needs is real-time control built directly into the infrastructure that runs agents.

 

Rubrik Agent Cloud was built to solve this challenge

Rubrik Agent Cloud provides a control plane for AI agents that sits among applications, agents, and the underlying LLMs. The goal is simple: allow organizations to move fast with agentic AI while maintaining the guardrails required for enterprise environments.

The platform focuses on three capabilities:

Continuous Monitoring and Observability: You cannot govern what you cannot see. Rubrik Agent Cloud continuously scans the environment to create a dynamic inventory of AI agents running across the organization. It identifies existing agents, how frequently they are used, the tools and data they access, and the risk profile associated with their behavior. This visibility becomes the foundation for enterprise AI oversight.

Dynamic Agent Governance with AI: Traditional governance approaches rely on static policies and manual review processes. That model breaks down when agents can generate thousands of interactions and actions per hour. Rubrik Agent Cloud brings governance directly into the runtime of agent interactions, enforcing policies for prompts entering an agent, responses generated by the model, and tool calls executed by the agent.

Yesterday’s static rules fail in an agentic world. Rubrik Agent Cloud uses AI to govern AI, allowing policies to adapt to the dynamic behavior of agents in real time. Organizations can start with predefined policies such as PII detection, or define custom governance rules in natural language. 

Remediation with Agent Rewind: Even with strong controls, mistakes will happen. Agents may delete data, modify records, or execute unintended actions across enterprise systems. Rubrik Agent Cloud integrates directly with Rubrik’s cyber resilience platform to provide Agent Rewind. So if an agent causes damage, organizations can instantly reverse destructive changes, restore deleted assets, and recover systems from a known good snapshot.

This creates an essential safety net for autonomous systems operating inside critical infrastructure.

 

AI Velocity Requires AI Governance

The Amazon outages illustrate a broader reality: AI is dramatically increasing the speed of software development and automation. But the operational controls that protect enterprise systems were designed for a slower era.

Organizations now face a new challenge: How do you allow teams to deploy agents at speed without exposing the business to unpredictable risk?

The answer is not slowing down AI innovation.

The answer is building governance directly into the agent ecosystem.

 


Rubrik Agent Cloud helps enterprises move governance from committee meetings and documentation into the infrastructure where agents actually operate.

And in the age of autonomous systems, the most important question is no longer “Can we build agents?” 

It’s “Can we control them?”

To learn more or get started, please reach out to us at ai-team@rubrik.com.

Related Articles

Blog by This Author