Operations | Monitoring | ITSM | DevOps | Cloud

We are exploring how we can help on-call engineers troubleshoot incidents more effectively by providing the OpenAI Codex agent with access to real-time observability data in terminals. We've developed an integration and new tool visualizations that connect OpenAI's Codex CLI to the new Datadog MCP server. In this post, we'll share what we've been experimenting with: enabling an AI agent to retrieve production metrics, logs, and incidents from Datadog in real time and act on that context.

As organizations bring more AI and LLM workloads into production, the underlying GPU infrastructure that supports these workloads becomes even more critical in ensuring these workloads remain fast, reliable, and scalable. Inefficient GPU resource usage, for instance, can lead to longer runtimes and reduced throughput, negatively impacting overall model performance. Additionally, idle and underutilized GPUs can quickly drive up costs and lead to needless spending.

As development teams adopt AI-powered tools and build services that make use of AI agents, they want to extend their AI capabilities to incorporate familiar tools and observability data. However, AI agents struggle with regular API endpoints and frequently fail when parsing complex nested JSON hierarchies or incorrectly handling errors. As a result, these agents often fail to retrieve relevant results.

Developers lose hours each week to a familiar troubleshooting loop: chase down telemetry across dashboards, decipher vague errors, and juggle alerts to find the signal worth fixing. Production issues, performance regressions, and security vulnerabilities all demand attention, but they often come with little context for taking action.

As your organization grows, you may operate in increasingly complex environments and manage more services and larger teams to maintain them. Evolution like this can lead to an explosion of telemetry data from across your stack, including metrics, traces, logs, and frontend interactions. The benefit of greater visibility is often outweighed by the challenge of acting on the data you collect, and you can easily fall behind on implementing the fixes your services require to operate reliably and efficiently.

As data systems grow more complex and data becomes even more business-critical, teams struggle to detect and resolve issues that impact data quality, reliability, and, ultimately, trust. Engineers have to rely on manual checks and ad hoc SQL queries to catch data quality issues—often after teams relying on the data have noticed something has gone wrong.

Datadog’s Oracle Cloud Infrastructure integration enables you to collect metrics and logs from your entire OCI stack and monitor them within a single platform alongside other third-party technologies. Datadog’s new OCI QuickStart is a fully managed, single-flow setup experience that helps you monitor your OCI infrastructure and applications in just a few clicks.

To efficiently optimize your LLM application before pushing to production, you need a comprehensive testing and evaluation framework. By running experiments, you can optimize prompts, fine-tune temperature and other key parameters, test complex agent architectures, and understand how your application may respond to atypical, complex, or adversarial inputs. However, it can be difficult to manage your experiment runs and aggregate the results for meaningful analysis.

Getting paged pulls engineers away from meaningful work, yet incident response in many organizations remains manual, reactive, and draining. An alert fires and teams scramble to find the root cause, relying on siloed knowledge, incomplete context, and a few on-call experts who are already stretched thin. The rise of AI coding agents has only intensified this challenge: As teams ship code faster with less human oversight, production systems grow increasingly complex and harder to understand.

Migrating to a new logging platform can be a complex operation, especially when it involves both active and historical logs. Observability Pipelines offers dual-shipping capability, making it easy to route active logs to your new platform without disrupting your log management workflows. But migrating years worth of historical logs—which are critical for investigating security incidents and demonstrating compliance with applicable laws—requires a different approach.