---
title: "The Future of Cloud-Native Resilience: The Clumio Vision | SHIFT On-Demand 2026"
type: "WebPage"
language: "en-US"
url: "https://www.commvault.com/resources/video/the-future-of-cloud-native-resilience"
date: "2026-11-16T14:06:40-05:00"
modified: "2026-11-16T14:06:40-05:00"
description: "See how cloud-native, AI-assisted recovery helps teams restore clean data at machine speed and prove resilience at scale."
image: "https://www.commvault.com/wp-content/uploads/2026/11/The-Future-of-Cloud-Native-Resilience-The-Clumio-Vision.png"
breadcrumbs:
  - name: "Commvault Home"
    url: "https://www.commvault.com/"
  - name: "Resources"
    url: "https://www.commvault.com/resources"
  - name: "Video"
    url: "https://www.commvault.com/resources/video"
  - name: "The future of cloud native resilience"
---

# The Future of Cloud-Native Resilience | SHIFT On-Demand

- [Home](/)
- Videos
- The Future of Cloud-Native Resilience | SHIFT On-Demand

ON-DEMAND SESSION

## The Future of Cloud-Native Resilience: The Clumio Vision

See how cloud-native, AI-assisted recovery helps teams restore clean data at machine speed and prove resilience at scale.

[Request a demo](/request-demo)

Play Video      ![Video thumbnail](https://img.youtube.com/vi/tDUY-7srSeQ/maxresdefault.jpg)

---

- [Key Takeaways](#Takeaways)
- [About This Session](#About)
- [Related Resources](#Resources)
- [FAQs](#FAQs)

[Back to all on-demand sessions](/shift#watch-now)

---

## Key Takeaways

**Recovery Is the New Uptime**

Measure how quickly clean data returns, not simply whether backups completed.

**AI Changes Data Risk**

Agents can create, change, or damage cloud data at machine speed.

**Protect Copies Outside Production**

Keep immutable backup data isolated from production accounts and credentials.

**Recover Only What Matters**

Restore affected objects, prefixes, or partitions while preserving healthy data.

**Automate the Recovery Plan**

Use APIs, infrastructure as code, and AI-assisted workflows to reduce manual effort.

**Choose Clean Recovery Points**

Use anomaly context and risk scores to identify trustworthy backups.

---

## About This Session

**AI has changed how cloud applications create data, fail, and recover, making recovery speed a business priority.** This session examines why legacy backup tools struggle with cloud-native scale and AI-driven change. It challenges teams to identify a measurable clean recovery time and explains why backup success alone does not prove an application can return safely.

#### **Recover at Machine Speed**

Learn where recovery time goes: moving data, rebuilding application dependencies, and selecting a trusted recovery point. Clumio uses elastic serverless processing and service-specific optimization to restore large Amazon S3, DynamoDB, and Google Cloud Storage environments efficiently.

**Precision recovery reduces disruption by restoring only damaged data while preserving healthy objects and application activity.** See how granular selection and in-place recovery can target affected objects, prefixes, or partitions instead of restoring an entire petabyte-scale dataset.

#### **Automate Cloud Recovery**

Explore API-first protection, infrastructure as code, policy-driven workflows, and an MCP server designed to bring recovery into the tools engineers already use. **AI-assisted skills can capture recovery expertise, validate requirements, and help teams review and execute a plan rather than reconstruct one under pressure.**

#### **Find a Clean, Trusted Copy**

ThreatDetect analyzes backup data and combines anomaly indicators into a risk score that helps teams select a clean recovery point with greater confidence.

Measure recovery with a real restore, confirm backup isolation, and establish the time required to return clean data.

CONTINUOUS CLOUD RECOVERY

## CloudRecover

Protect and recover cloud-native data with scalable SaaS simplicity.

[Explore solution](/solutions/cloud-recover)

CONTINUOUS CYBER RECOVERY

## CyberRecover

Clean, isolated recovery after cyber disruption.

[Explore solution](/solutions/cyber-recover)

---

## Related Resources

VIDEO

## Blazing Hot Recovery

See rapid, granular recovery for Amazon S3 and DynamoDB.

[Watch the Video about Blazing Hot Recovery](/resources/video/blazing-hot-recovery)

DEMO

## Clumio Backtrack

Restore S3 and DynamoDB data while preserving Terraform state.

[View the Demo about Clumio Backtrack](/resources/demos/clumio-backtrack-and-terraform)

BLOG

## Google Cloud Recovery

Explore immutable, petabyte-scale protection for Google Cloud Storage.

[Read the Blog about Google Cloud Recovery](/blogs/cloud-native-backup-and-recovery-for-google-cloud-storage-now-available)

---

## Frequently Asked Questions

What is cloud-native recovery?

Cloud-native recovery uses elastic, service-aware architecture to protect and restore cloud data at scale without relying on traditional backup infrastructure or manual console-driven processes.

Why are production snapshots not enough?

Snapshots stored in the production account may be reachable by the same credentials as production. An attacker or automation error could delete both the data and its recovery copy.

How does granular recovery reduce downtime?

Granular recovery restores only affected objects, prefixes, or database partitions. This avoids moving an entire dataset and preserves healthy data surrounding the incident.

How can AI assist recovery workflows?

AI-assisted recovery can validate configurations, apply expert skills, recommend a plan, and help teams execute consistent workflows instead of rebuilding procedures under pressure.

How do teams choose a clean recovery point?

Anomaly detection can combine multiple indicators and assign risk context to backups, helping teams select a trustworthy recovery point rather than relying only on timestamps. ﻿

---

## Transcript

View full transcript

Hi everyone. Great to be here. I want to start with the question. Please answer it in your head. Honestly, before I go on. If something took down your application right now, it could be an attacker, an AI agent, or a bad code push. How long would it take to get it back? And how would you know that the data you got back was clean?

If you have a number, you are in a small minority. Most teams I meet do not have one. That is not because they haven’t thought about it for the rest of this session. Keep your number or lack of a number in mind. It is because three things change this year. And the tools they inherited were not built for any of them.

The first change your data has a new author. A year ago most new databases were still created by people. That is changing fast. Neon is a cloud database company. Neon found that 30% of its databases were being created by AI agents, not by people. A few months later, that number was over 80%. Agents were creating four times as many databases as humans.

Think about what that means for recovery. The thing that creates your data now works at machine speed. It never sleeps and it never asks whether there is a backup. The second change, your agents do not need to be hacked to do damage. Sale point. Surveyed organizations that use AI agents. 80% of them said their agents had already done something unintended.

The agents read systems they were not meant to reach. They shared data. They should not have shared. Last July, a coding agent ran destructive commands against a live production database. It did this during a code freeze when nothing was supposed to change. Then the agent told its user that the damage could not be undone. That was false, but the user only found out because a person checked there was no attacker.

There was no bad intent. There was only an agent with more access than judgment, and it caused the same damage as a breach. The third change. Attackers now run at machine speed, but recovery still runs at human speed. Here’s a story recently in the news. Unit 42 is the Incident Response team at Palo Alto Networks. In the fastest attacks they investigated last year, the attackers were stealing data within 72 minutes of getting in.

One year earlier, that number was 285 minutes. And the attackers go after backups first. In 26% of extortion cases. They accessed the backups last year. Anthropic reported a state sponsored espionage campaign in which AI did an estimated 80 to 90% of the tactical work. Youmans only stepped in at a few decision points. The attackers have automated recovery efforts.

Still have not. Recovery waits for a person, a runbook and a console.

So my original question is no longer theoretical. The thing that breaks your application now moves faster than the people who would fix it. Here is what customers tell me. Almost word for word. My developers can ship an application in a week. Nobody can tell me how we would get its data back if it were compromised. We backup everything, but we have never tried to restore all of it.

I worry less about attackers than about our own automation. Deleting the wrong thing. Notice that none of them is asking for a better backup. They already have backups. What they cannot get is the number. They cannot get the number because the tools they inherited were built for a different world. Those tools were built for servers and for the people who ran the servers.

They expect an agent installed on the machine. They expect a console, and they expect an administrator who clicks. In a cloud native environment. There is no server to install the agent, and there is no administrator waiting at 2 a.m.. Those tools also keep the backup copy in the same account where the production runs. That means the snapshot is one API call away from anyone or anything that holds the credentials.

That is a copy that is not a recovery. And those tools measure the wrong thing. They are measuring backup success rates. They measure how old the last copy is. Nobody measures how long it takes to get a clean copy back so nobody knows.

AI changed how applications are built because AI is authoring more of your data and application. Your agents do not need to be hacked to do damage. Attacks are running at machine speed now. Now it has to change how they are recovered. This is what CloudRecover is for. Rather than tell you about it myself, I’m going to ask Woon Zung, our Chief Technology Officer of Clumio.

To take my question apart. Where does the time actually go when you recover a cloud application? And what does it take to make every part of the recovery run at machine speed? Woon.

All yours.

Thank you.

Poojan asked. How long does it take to recover? Let’s unpack that question by looking at where the time goes. When you restore an application. I’ll put it into three buckets. The time that it takes to move the data. The time that it takes to reconfigure the application and get it back running, and the time it takes to identify a good recovery point.

And in every one of those buckets, the goal is the same machine speed. Let’s start with moving the data. Cloud applications can be truly massive. Tens of billions of objects in a single S3 bucket. Hundreds of billions of items in a DynamoDB table. Customers take full advantage of the clouds elasticity to build those massive applications. But traditional backup solutions were never built to handle that type of scale.

That’s why we built as a cloud native SaaS platform.

The moment recovery starts, clumsy offers up hundreds of Lambda functions or Google Cloud Functions, divides the work into smaller tasks and schedules them in parallel. When they work, finishes that compute scales back down to zero. But elasticity alone isn’t enough. Amazon S3 DynamoDB. In Google Cloud Storage, each have their own API limits and performance characteristics. So we’ve built optimizations specific to each service from how we distribute the work to how we move the data.

That takes substantial amount of engineering effort. But that is how we turn cloud capacity into recovery. Throughput in one customer recovery restored 700TB of data in just two hours. That is the kind of throughput that this architecture makes possible. But moving the data faster, it’s only one way to save time. Another is to move less of it. Your application.

My hold. Petabytes. The incident. My effect. Only a few gigabytes. Bad code. Overwrite one prefix in an S3 bucket. An AI agent corrupts a subset of records in a DynamoDB table. Why recover the entire data set? To repair that damage with, you can choose exactly what to recover the affected objects, a specific prefix, or a DynamoDB partition. You choose the point in time before the damage, and you choose where to recover it, including another account or region.

And with Backtrack, you can roll back the affected data in place while preserving the good data around it.

We reduce the time in two ways. Execute faster and recover only what’s needed now. The second bucket. Getting the application running again. Your data can be back while your application is still down. The application also depends on configurations, permissions, networking, and the services that connect everything together.

Rebuilding those dependencies by hand can take as long as recovering the data itself, sometimes longer.

Users have always been engineers, so we built the way we want to use it ourselves. Apply first infrastructure as code and policy driven. Our customers already use these capabilities to integrate into their CI CD pipeline and automate recovery through APIs. But there’s a difficult last mile problem. APIs let us automate the recovery steps, but often someone still has to manually drive the recovery process because of configuration drifts and application specific requirements, and that causes the automated process to slow back down to human speed.

This is where we put AI to work. Engineers are telling us that the way they manage infrastructure is changing. Yes, API and infrastructure as code still matter, but a lot of their day to day work is now moving into tools like Cloud Code and Codex. They’re asking us to bring data protection into the same workflow. And the natural first step was an MCP server, which we’re launching today.

But we wanted to do more. We wanted to capture all the expertise and knowledge that we built over the years, helping our customers recover their data. What to check before starting, how different services behave, what can go wrong and how to validate the result. We’re working to capture all of this in skills and workflows that our customers will be able to use.

Our customers should be reviewing and executing a plan rather than reconstructing one. During recovery.

MCP provides access to the tools skills help the agent to use them to carry out recovery workflow consistently and thoroughly. With the help of this expert agent. Our goal is to accelerate both data recovery and application recovery. That’s the second book bucket at machine speed. That brings us to the third bucket and a good recovery point. We’ve talked about how quickly we can recover.

But before we start we need to decide which recovery point to use. The last thing you want to do is to restore data that is still compromised. This is where ThreatDetect comes in. ThreatDetect is our new anomaly detection capability. It analyzes backup data for signs of compromise, rather than treating every unusual spike as a separate incident.

It brings multiple indicators together to assess the risk that matters, because an unusual change isn’t always an attack. Your team needs enough context to decide what deserves attention. ThreadDetect gives every backup a risk score that helps your team and a good recovery point with confidence. So when you’re preparing for recovery, you have more than a list of backup timestamps.

You have evidence to help you decide which one to use. Three buckets. Machine speed in every one of them. That’s the answer to question. Back to you, Poojin.

Thank you.

So let me go back to the question. If your application broke right now, how long would it take to get it back? And how would you know that the data was clean? One just showed you where the time goes. Moving the data, getting the application running again, and choosing a recovery point you can trust. And he showed each of you those steps running at machine speed.

Petabytes recovered in hours, not days. An agent that executes a recovery plan instead of a person building one under pressure, and a risk score on every backup before you ever need it. 20 years ago, the site reliability engineers at Google accepted that failure is certain. So they stop measuring how rarely things break. They started measuring how fast things come back.

Data protection never made that change. And now AI has made data failure both certain and fast. So recovery is the new uptime. The number that matters is how long it takes until clean data is back. So here is what I ask you to do this week three things. First, pick your number. Pick your largest bucket or your most important table.

Ask your team how long a clean recovery would take. Most of you will discover that nobody knows. That is not a failure. It is a gap, and it is the easiest gap in your environment to close. Second, find your copies. Ask where your backup copies live. Then ask whether the credentials that run production can reach them. If the answer is yes, you have a copy, but you do not have a recovery.

Third run one. Not a planning exercise. A real restore of something that matters from your own pipeline. The number you get from that is the only number that counts. Time to recover.

AI made this question harder, but for the first time a also made the answer possible. Recovery that runs at the speed of the thing that broke it. CloudRecover. Recovery built for the cloud. Proven at scale. Ready for AI. Thank you.
