Restoring systems and restoring the business are different jobs. A recovery plan can bring every server back on schedule and still leave the company stuck. A restored system doesn’t help much if employees can’t log in or if no one can confirm it’s clean. At its Catalyst virtual summit, Cohesity made the case for recovery plans built around the business, and for putting AI agents in those plans. I think both points are overdue. Agents act on real data and systems, and a bad agent action should be as recoverable as any other incident. Cohesity showed how it plans to deliver on each.

Agent Resilience gives a compromised agent a way back

Agents build up memory and context as they work, and that context helps them deliver better answers and decisions. I expect guardrails to stop plenty of bad prompts, but not all of them. Say a bad prompt poisons an agent’s memory, or its settings drift. Without a trusted recovery point, a team may have to rebuild the agent from scratch and lose the context it had built up.

Cohesity Agent Resilience gives teams that recovery point. It protects an agent’s memory and configuration, as well as the databases, file systems, and other services the agent interacts with. When something goes wrong, a team picks a recovery point from before the trouble started and restores only what was affected. The agent comes back with the memory and configuration it had at that recovery point.

Agent Resilience maps each agent, including its memory, configuration, and data resources, to a single application group in Cohesity Data Cloud. I like that design because the group draws a single boundary around the agent and its critical dependencies. The agent’s memory and configuration get the same snapshot architecture, immutable backups, and clean-room recovery that Cohesity customers already rely on.

Cohesity’s Autonomous Cyber Resilience vision starts with the business

Every recovery plan should start with one question: what does the business need to keep running? Cohesity calls the answer a minimum viable company, which includes the critical systems, data, and processes a business needs to survive. Knowing that list tells a company what to recover first when an attack hits.

Cohesity described Autonomous Cyber Resilience as a vision to automate its five-step cyber resilience framework through agentic workflows. Teams would set goals through Copilot, Cohesity’s AI assistant, instead of configuring policies by hand. I think turning business priorities into backup policies by hand is where recovery plans fall out of date. Cohesity’s recommendations would still need a person’s approval before they run.

The vision builds on RecoveryAgent, which Cohesity launched in 2025 to orchestrate cyber recovery. RecoveryAgent is a good example of how building on a product that already ships makes a vision easier to believe in.

Cohesity also shipped a step toward that model. Cohesity DSPM finds and classifies sensitive data. Customers who run it with Cohesity Data Cloud Enterprise Edition can now have newly discovered sensitive data protected automatically. Before applying a rule, customers can preview which objects the rule would add or remove. That preview is the right instinct for automation that changes protection policies.

Gaia Catalog is designed to serve backup data without copying it

Cohesity already backs up file servers and network-attached storage in companies’ own data centers. Cohesity argues that AI tools built for the cloud can’t reach those files without copying them and building pipelines. Every copy is one more place to secure and one more bill to pay.

Gaia Catalog is designed to enable teams to search, classify, and curate documents and other unstructured data within Cohesity backups. A team would pick only the files a project needs and publish them through a read-only S3-compatible endpoint. The data itself would stay in Cohesity. Databricks and Microsoft Fabric are the first platforms Cohesity has named for the integration.

Cohesity picked the right interface. Data teams that already work with S3 storage should have little new to learn. Gaia Catalog is designed to apply Cohesity’s access controls to each dataset and log every access. I think carrying those controls over matters more to a data team than the interface does, because governance is usually what stalls these projects. Curating first should also save money, because a team processes only the files it selects.

Cohesity is moving into the tools customers already run

Agent Resilience discovers supported agents on Amazon Bedrock, where teams build them. Gaia Catalog is designed to make curated data available in Databricks and Microsoft Fabric. Showing up inside tools customers already use should make Cohesity easier to adopt. Easier adoption can get Cohesity in the door. I think what keeps customers there is the new data Cohesity will hold, like agent memory and configuration, and the new uses for the backups it already has. Every agent and dataset a customer adds should make Cohesity more central to how the business runs.

In my view, Catalyst moved Cohesity further from the backup label. Agent recovery and AI-ready backup data should put Cohesity in front of security and data teams as well as backup admins. I see agent recovery specifically as a new market, and Cohesity entered it with recovery technology that its customers already rely on. Next, I’d like to see Agent Resilience reach more agent platforms. I expect most enterprises to run agents in more than one place.

Cohesity made a convincing case at Catalyst that AI agents belong in the recovery plan, right next to the data they touch.