Datadog bets on AI observability

A lot of companies are racing to build artificial intelligence agents, but most of those agents are still half-blind once they touch a real app or cloud system. Datadog is trying to become the dashboard those agents look at before they make a change. (datadoghq.com) Datadog’s new Model Context Protocol server is the bridge. It lets an artificial intelligence agent pull live metrics, logs, traces, monitors, dashboards, incidents, and security findings from Datadog instead of guessing from stale snippets or pasted screenshots. (docs.datadoghq.com) Model Context Protocol is a standard for giving artificial intelligence tools access to outside systems. Datadog says its server works with coding agents and editors including Cursor, Claude Code, OpenAI Codex, Visual Studio Code, JetBrains tools, GitHub Copilot, and Kiro. (datadoghq.com, docs.datadoghq.com) The pitch is simple: if an agent is helping fix a production outage, it should be able to see the same telemetry a human engineer sees. Datadog says its managed server includes governance controls, records tool calls in Audit Trail, and emits usage metrics such as `datadog.mcp.session.starts` and `datadog.mcp.tool.calls`. (datadoghq.com, docs.datadoghq.com) Datadog is not selling only the pipe. It is also expanding Bits Artificial Intelligence, the company’s assistant first launched in August 2023, into a set of domain-specific agents for operations, software development, and security work. (datadoghq.com, datadoghq.com) One of those products, Bits Artificial Intelligence Site Reliability Engineer, is marketed as an always-on troubleshooting agent. Datadog says it was developed against thousands of real-world incidents and can help restore services 90% faster by narrowing root causes in minutes. (datadoghq.com) The other new leg of the strategy sits far from on-call alerts. Last week Datadog launched Datadog Experiments, a product that combines product analytics, business metrics from a company’s data warehouse, and application observability so teams can test a feature and watch both revenue and system health in one place. (datadoghq.com, datadoghq.com) That product came out of Datadog’s acquisition of Eppo in May 2025. Datadog said Eppo would bring feature management and experimentation into the same platform as observability, and the new Experiments launch is the clearest version of that plan so far. (datadoghq.com, cooley.com) Put together, the strategy is to cover the full loop: an engineer ships a change, an experiment measures whether users liked it, observability shows whether the system stayed healthy, and an artificial intelligence agent helps investigate if something breaks. That is broader than classic infrastructure monitoring, which is why Datadog keeps tying product teams and engineering teams into the same workflow. (datadoghq.com, datadoghq.com) Wall Street is starting to read these launches as new ways for Datadog to make money from the artificial intelligence boom without trying to build the underlying models itself. On April 9, 2026, Guggenheim upgraded Datadog to Buy from Neutral and set a $175 price target, saying the company looks like a beneficiary of rising data volumes, growing information technology complexity, and demand for artificial intelligence-era tooling. (investingchannel.com) The bet inside all of this is that artificial intelligence agents will not be trusted unless they can be measured, replayed, and constrained like any other production system. Datadog wants that trust layer to run through its platform, whether the customer is debugging a midnight outage or testing a new checkout button. (datadoghq.com, docs.datadoghq.com, datadoghq.com)