Cased vs Harness (2026)

The 20+ integrations include AWS, GCP, Azure, GitHub, Slack, Sentry, Datadog, PagerDuty, Groundcover, Honeycomb and ArgoCD, and agents manage deploys and open pull requests with fixes for Terraform and CloudFormation drift. Further integrations include CloudWatch, PostHog, Render and Pulumi, as well as Atlassian, Linear, Betterstack, Fly.io, Vercel and Notion. Cased CD sends Slack, webhook and email notifications.

SourceCased, docs.cased.com llms-full.txt, cased.com and cased.com/cdread 2026-10-05

AI SRE connects Datadog, Splunk, New Relic, Dynatrace, Slack, Microsoft Teams, Zoom, Jira, ServiceNow and GitHub with Harness's own deployment, feature flag and pipeline records. Automation runbooks execute rollbacks and mitigations.

Sourceharness.io/products/ai-sreread 2026-09-29

Custom agents are defined by a prompt and a trigger, and the API lists and runs workflows. Cased does not say how a workflow is built, whether in steps, with branches or with handoffs between agents. Each trigger starts a new agent session, and the Sentry agent can open fix sessions of its own. Cased advises writing a prompt as if explaining the task to a senior engineer.

SourceCased, docs.cased.com llms-full.txtread 2026-10-05

The AI Scribe builds the incident record while AI root cause analysis correlates deploys, feature flags, infrastructure and monitors. On call routing, escalation rules and automation runbooks for remediation run alongside them within the Harness pipeline platform.

Sourceharness.io/products/ai-sreread 2026-09-29

Agents can be triggered in three ways, on a schedule, through the API, or when Cased receives a webhook from an external system, so alerts such as Groundcover or Sentry start agents without anyone asking. Named webhook events include a new Sentry issue, a deployment starting and a Groundcover alert firing, and schedules run daily, weekly or monthly.

SourceCased, docs.cased.com llms-full.txtread 2026-10-05

Alerts are ingested and correlated from monitoring tools, and inbound webhooks create alerts from any JSON payload. Related alerts are deduplicated into incidents that route to on call by escalation rules, and the AI Scribe records across Slack, Zoom and Teams as the incident runs.

Sourceharness.io/products/ai-sreread 2026-09-29

Teams connect Notion and import their documentation, runbooks and knowledge base articles into Cased's knowledge base, choosing the pages to share and how many to import, and the agents then use those pages when answering questions or investigating issues. Importing again syncs updates from Notion, refreshing existing pages and adding new ones. Teams can also capture their own knowledge in voice interviews with AI.

They talk through architecture decisions, the conversations are transcribed and indexed, and the result becomes a searchable knowledge base the agents query as context. Agents take context from connected integrations and telemetry at run time, and Cased offers runbook and documentation integration that connects data from any source.

SourceCased, docs.cased.com/integrations/notion, docs.cased.com llms-full.txt and cased.comread 2026-10-05

Root cause analysis draws on the change intelligence the platform keeps. That means the deployments, feature flags, configuration and infrastructure events recorded in Harness, which attach to incidents automatically.

Sourceharness.io/products/ai-sreread 2026-09-29

No memory or state carries across agent sessions. Agents draw context from connected integrations each run, and each trigger starts a new session.

SourceCased, docs.cased.com llms-full.txtread 2026-10-05

Incident records, timelines and change history are the application's data. Harness names no memory layer for the AI SRE agents.

Sourceharness.io/products/ai-sreread 2026-09-29

Agent fixes arrive as pull requests that the team merges, and Cased CD's RBAC and approval workflows control who can deploy to which environments. Roles grant actions such as sync and override on applications, clusters and repositories. There is no approval step on agent actions themselves, and agents can roll back a deploy automatically when they detect issues.

SourceCased, docs.cased.com llms-full.txt and cased.com/cdread 2026-10-05

Remediation runs as one click automation runbooks for rollbacks and mitigations that a responder triggers, with ownership policies and escalation rules governing who acts.

Sourceharness.io/products/ai-sreread 2026-09-29

Access controls cover SSO through Okta, Auth0, Azure AD, Google, AWS Cognito or Dex (with SAML, LDAP or GitHub behind it), fine grained RBAC per application with administrator, developer, viewer and project scoped roles, and groups that map to roles. Users can be created and deleted from the Cased CD interface. Cased publishes no security attestation of its own and has no security page or trust site. SOC 2 and ISO 27001 come up only as standards that its audit trail and compliance workflows help customers meet.

SourceCased, docs.cased.com llms-full.txt and cased.com/cdread 2026-10-05

Access controls include SSO with the corporate identity provider, two factor authentication, Advanced Role-Based Access Control over users and service accounts, and an account and organization level audit trail. Compliance covers SOC 2 and ISO 27001, 27017 and 27018, and Harness runs a trust center.

Sourceharness.io/securityread 2026-09-29

Every action in Cased is logged with who, what, when, source IP and the state before and after, a complete action history that covers what agents changed. Audit events are kept indefinitely within the storage allocation, where the default 10GB holds about 20 million events. Deploy Monitor reports on each deployment's health as it runs and analyzes it afterward.

SourceCased, docs.cased.com llms-full.txtread 2026-10-05

The AI Scribe keeps an automatic incident record and live timeline from Slack, Zoom and Teams for postmortems, which records what people said and decided. Harness names no record of the steps or tools the root cause agent used. The platform also keeps an account and organization level audit trail.

Sourceharness.io/products/ai-sreread 2026-09-29

Cased CD Enterprise deploys into the customer's Kubernetes cluster. The agent platform is hosted, with no region or customer environment option. Cased CD can run on premises or in the customer's own cloud, installs in a few minutes by script, Helm chart or kubectl on Kubernetes 1.19 or later, and works with an existing ArgoCD installation. A free community edition is on GitHub.

SourceCased, docs.cased.com llms-full.txt and cased.com/cdread 2026-10-05

AI SRE is sold as an Operations module on the Enterprise plan of the Harness SaaS platform. The Self-Managed Enterprise Edition supports Cloud Cost Management, Chaos Engineering, CD and GitOps, CI, Code Repository, Database DevOps, Feature Management, IaC Management, Platform, Supply Chain Security and Security Testing, but not AI SRE, so AI SRE cannot be self hosted. No region selection or data residency option for AI SRE is published.

Sourcedeveloper.harness.io/self-managed-enterprise-edition/new-to-self-managed-enterprise-edition/smp-supported-platformsread 2026-09-30

Default Agents ship for separate jobs, such as analyzing Sentry errors, monitoring deployment health and analyzing Kubernetes alerts (the Kubernetes Error Analyzer), beside custom agents, each doing its own job. A fourth default agent adds the Cased Deploy Monitor GitHub Action to a repository's deployment workflow, and agents working with Cased CD handle drift detection, cost optimization and automated remediation.

SourceCased, docs.cased.com llms-full.txt and cased.com/cdread 2026-10-05

Two agents ship, the AI Scribe and AI root cause analysis. The second reads the first's timeline, so the two work together as one incident product. Runbooks are the customer's own.

Sourceharness.io/products/ai-sreread 2026-09-29

The agents are built on Anthropic's Claude Agent SDK and use Anthropic's Tool Search API to pick tools. Cased chose that single provider, and customers cannot choose a model.

SourceCased, docs.cased.com llms-full.txtread 2026-10-05

No model provider is named for the AI SRE agents, and Harness does not say whether customers can choose a model.

Sourceharness.io/products/ai-sreread 2026-09-29

A public API, authenticated with a bearer API key, has endpoints to create deployments (POST /api/v1/deployments/), start agent sessions (POST /api/v1/agent/sessions/), list workflows and start workflow runs. The base URL is app.cased.com/api, and the workflow endpoints also return a workflow's details, list its runs and fetch each run.

SourceCased, docs.cased.com llms-full.txtread 2026-10-05

AI SRE data is released into the Harness Knowledge Graph and available through MCP for cross module analytics. The Harness MCP server has an endpoint, OAuth or API key authentication and create tools across 139 resource types. None of its tools covers AI SRE incidents, on call or runbooks, and Harness publishes no AI SRE API reference.

Sourcedeveloper.harness.io/harness-ai/3.0/use-harness-ai/ide-and-integrations/mcp-server/harness-mcp-serverread 2026-09-30

The agents have no evaluation harness, scored tests or feedback loop. Deploy monitoring and anomaly detection watch the customer's releases, not the agents, checking each release for anomalies, new bugs and performance issues.

SourceCased, cased.com; docs.cased.com/llms-full.txtread 2026-10-05

Postmortem automation records incidents, not agent quality. Harness names no evaluation or feedback loop for the AI SRE agents.

Sourceharness.io/products/ai-sreread 2026-09-29

Agents act through integrations, pull requests and the API, and they have no browser, desktop or computer control. Cased's open source sandboxes project offers one API for cloud code execution sandboxes, which run code and do not control a browser or computer.

SourceCased, cased.com and docs.cased.com llms-full.txtread 2026-10-05

Agents work through integrations, runbooks and chat. Harness names no browser, desktop or computer control.

Sourceharness.io/products/ai-sreread 2026-09-29