Agentless reach by workload type: what each platform reads without an agent

Configuration is not the question

Reading cloud configuration through provider APIs, the job of CSPM, is common to all nine platforms. The difference is in the next layer: reading the workload itself, its packages, secrets and files, without installing anything. That is what our agentless workload depth criterion scores, at 25% of the total, the highest weight on this site.

What each platform says it reads

From vendor pages read on 27 September 2026:

• Orca Security: SideScanning reads runtime block storage and covers Linux and Windows VMs, containers and images, and serverless.
• Wiz: states it reaches VMs, containers, serverless, PaaS and AI resources without agents, through an API connection.
• SentinelOne: describes Cloud Native Security as "an agentless CNAPP" for multi-cloud estates; the workload types are not listed on the pages reviewed.
• Upwind: agentless scanning is part of the platform; its design centers on the eBPF sensor.
• Microsoft Defender for Cloud: agentless VM, secrets and Kubernetes scanning, in the paid Defender CSPM and Defender for Servers Plan 2 tiers.
• Palo Alto Networks Cortex Cloud: posture, vulnerability and attack-surface modules work from cloud APIs; the pages reviewed lead with a runtime agent and do not name agentless workload scanning.
• CrowdStrike Falcon Cloud Security: agentless CSPM, with workload protection split between agentless scanning and the Falcon sensor.
• Sysdig Secure: agentless CSPM, CIEM and host vulnerability scanning on AWS, Google Cloud and Azure; in-use analysis and runtime detection need the agent.
• Aqua Security: agentless posture visibility; depth comes from kernel-layer enforcement.

How the scores line up

Orca scores 9.8 and Wiz 9.6: both name a wide range of workload types read without an agent. Upwind (8.0) and SentinelOne (7.8) lead with agentless capability but either center on a sensor or do not list workload types. Cortex Cloud (7.6), Defender for Cloud (7.4), CrowdStrike (7.2), Sysdig (7.6) and Aqua (6.8) either put agentless depth behind a paid tier or rely on an agent for it.

Why the difference matters for speed

A platform that needs an agent for workload depth gives you an inventory on day one and depth only as fast as the rollout goes. In a large estate that can take months, and some workloads never get an agent. A platform that reads depth agentlessly gives you both on the day the accounts are connected. That is the case for weighting this criterion highest when the question is time to full coverage.

Where agentless reach stops

Serverless functions and short-lived containers can appear and disappear between scans. Managed services are read through APIs, not disks. And nothing agentless can see a process start or block an attack. For the limits in more detail, see lesson 2.

How to check it yourself

Ask each vendor for a table: workload type by cloud, with "agentless", "agent required" or "not supported" in each cell. Then test one workload of each type in a trial, as lesson 6 describes. Use Stack up to see the coverage model lines for your shortlist side by side.