Scripted tests can’t keep up with AI-written code. Datadog is betting agents can.

Nearly a year into the AI coding boom, teams are increasingly relying on AI to write software. As a result, Anthropic's engineers are producing eight times more code than before, with some engineers submitting thousands of pull requests each month. While teams may think AI will speed up software development, the reality is that the bottleneck has shifted from coding to testing.

After AI generates code, it still needs to be tested, which traditionally meant human engineers writing the tests. This includes determining which user flows require testing, scripting those tests, and adjusting them when the flows change. Datadog has recognized this issue and acquired Propolis, integrating agent-written, goal- and output-oriented test writing into its platform.

This new approach allows agents to explore applications, follow user flows autonomously, and generate tests from plain-language prompts. By combining Datadog's observability strengths with this innovative testing method, organizations can potentially reduce software development timelines while maintaining system performance and reducing the fragility of traditional hardcoded testing.