Agents and infrastructure now appear in the DigitalOcean dashboard

Autonomous agents now enter the same monitoring screen as infrastructure, with the launch of DigitalOcean Insights. The platform arrived in public preview on October 13. It's part of the company's Agent Learning Services portfolio.

The proposal brings together three layers in a single control plane. These include cloud infrastructure, applications, and agents.

Notice one practical detail. Moreover, data collection happens without requiring external instrumentation or third-party monitoring agents.

Telemetry comes from managed Kubernetes, the App Platform, and the company's own AI products.

The diagnosis behind the product

Shridhar Pandey, Director of Product, summarized the shift in the industry. According to him, coding agents have changed delivery speed, and more and more software being shipped includes agents.

He also described the scope of the problem. However, an agent in production spans model calls, routing decisions, tools, and the infrastructure underneath.

Therefore, monitoring only CPU and memory leaves much of the story out.

The OpenTelemetry foundation avoids format lock-in

This choice deserves attention from anyone thinking about portability. The system uses OpenTelemetry standards to collect and store data in an open format.

There's also a contextual advantage. However, since the data originates from the cloud environment itself, each record shares common dimensions.

These include project identifiers, resource tags, and assigned assets. Consequently, correlating anomalous behavior requires no custom mapping layers.

Collection covers three types of telemetry. These are system metrics, structured logs, and distributed traces.

In addition, querying accepts PromQL expressions. Also, there's a visual designer for those who prefer to build without writing a query.

Results appear in time-series charts, gauge panels, bar charts, and tables.

Agents tracked from model to tool

Here's the most interesting point for those operating agents. The system connects to serverless inference engines, batch inference, and the routing layer.

Through these connections, it captures four signals. These include token throughput, latency, operational failure rates, and computational cost estimates.

In agentic workflows, the integration extends to Managed Agents and the Action Gateway. These components interact with the Harness Runtime environment.

Pandey explained the practical outcome. According to him, the Action Gateway adds traces that allow following a single agent action from the model call to each invoked tool and back.

This way, it becomes clear exactly where slowdowns or failures occurred.

Alerts and configuration as code

The administrative side also comes ready. Operators set alerts per metric with designated evaluation windows.

They can also adjust notification retransmission frequency. Incidents can be forwarded to external communication channels.

Access uses role-based control. In addition, there's also a public API that allows configuring telemetry assets as code.

What to evaluate before trusting this preview

First, consider the product's stage. The public preview is available at no additional cost, subject to usage policies and retention limits.

Second, weigh the performance caveat. The company avoids guaranteeing standards for production workloads at this stage.

Third, plan for the transition. The company intends to consolidate current monitoring features under this brand at general availability.

Existing configurations, for now, continue to work.

Also, test portability in practice. An OpenTelemetry foundation helps, but it's still worth confirming the export of your data.

Follow our profile on Instagram!