Datadog: How AI code reviews slash incident risk

Integrating AI into code review workflows allows engineering leaders to detect systemic risks that often evade human detection at scale.

For engineering leaders managing distributed systems, the trade-off between deployment speed and operational stability often defines the success of their platform. Datadog, a company responsible for the observability of complex infrastructures worldwide, operates under intense pressure to maintain this balance.

When a client’s systems fail, they rely on Datadog’s platform to diagnose the root cause—meaning reliability must be established well before software reaches a production environment.

Scaling this reliability is an operational challenge. Code review has traditionally acted as the primary gatekeeper, a high-stakes phase where senior engineers attempt to catch errors. However, as teams expand, relying on human reviewers to maintain deep contextual knowledge of the entire codebase becomes unsustainable.

To address this bottleneck, Datadog’s AI Development Experience (AI DevX) team integrated OpenAI’s Codex, aiming to automate the detection of risks that human reviewers frequently miss.

Why static analysis falls short

The enterprise market has long utilised automated tools to assist in code review, but their effectiveness has historically been limited.

Early iterations of AI code review tools often performed like “advanced linters,” identifying superficial syntax issues but failing to grasp the broader system architecture. Because these tools lacked the ability to understand context, engineers at Datadog frequently dismissed their suggestions as noise.

The core issue was not detecting errors in isolation, but understanding how a specific change might ripple through interconnected systems. Datadog required a solution capable of reasoning over the codebase and its dependencies, rather than simply scanning for style violations.

The team integrated the new agent directly into the workflow of one of their most active repositories, allowing it to review every pull request automatically. Unlike static analysis tools, this system compares the developer’s intent with the actual code submission, executing tests to validate behaviour.

For CTOs and CIOs, the difficulty in adopting generative AI often lies in proving its value beyond theoretical efficiency. Datadog bypassed standard productivity metrics by creating an “incident replay harness” to test the tool against historical outages.

Instead of relying on hypothetical test cases, the team reconstructed past pull requests that were known to have caused incidents. They then ran the AI agent against these specific changes to determine if it would have flagged the issues that humans missed in their code reviews.

The results provided a concrete data point for risk mitigation: the agent identified over 10 cases (approximately 22% of the examined incidents) where its feedback would have prevented the error. These were pull requests that had already bypassed human review, demonstrating that the AI surfaced risks invisible to the engineers at the time.

This validation changed the internal conversation regarding the tool’s utility. Brad Carter, who leads the AI DevX team, noted that while efficiency gains are welcome, “preventing incidents is far more compelling at our scale.”

How AI code reviews are changing engineering culture

The deployment of this technology to more than 1,000 engineers has influenced the culture of code review within the organisation. Rather than replacing the human element, the AI serves as a partner that handles the cognitive load of cross-service interactions.

Engineers reported that the system consistently flagged issues that were not obvious from the immediate code difference. It identified missing test coverage in areas of cross-service coupling and pointed out interactions with modules that the developer had not touched directly.

This depth of analysis changed how the engineering staff interacted with automated feedback.

“For me, a Codex comment feels like the smartest engineer I’ve worked with and who has infinite time to find bugs. It sees connections my brain doesn’t hold all at once,” explains Carter.

The AI code review system’s ability to contextualise changes allows human reviewers to shift their focus from catching bugs to evaluating architecture and design.

From bug hunting to reliability

For enterprise leaders, the Datadog case study illustrates a transition in how code review is defined. It is no longer viewed merely as a checkpoint for error detection or a metric for cycle time, but as a core reliability system.

By surfacing risks that exceed individual context, the technology supports a strategy where confidence in shipping code scales alongside the team. This aligns with the priorities of Datadog’s leadership, who view reliability as a fundamental component of customer trust.

“We are the platform companies rely on when everything else is breaking,” says Carter. “Preventing incidents strengthens the trust our customers place in us”.

The successful integration of AI into the code review pipeline suggests that the technology’s highest value in the enterprise may lie in its ability to enforce complex quality standards that protect the bottom line.

See also: Agentic AI scaling requires new memory architecture

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.