The velocity of AI-driven code generation has outpaced traditional testing approaches. Datadog acquired Propolis to enable agents to construct and execute goal-oriented tests based on plain-language specifications.

The agentic coding wave has fundamentally altered how many organizations approach software development. Engineers at Anthropic now produce eight times more code than previously, with some individuals submitting thousands of pull requests monthly.

Organizations banking on AI to multiply engineering output will discover that agent-generated code alone does not automatically compress the software development lifecycle (SDLC). The true constraint has migrated from code authorship to downstream validation phases.

Testing represents the critical chokepoint in modern software creation. Once agents generate new code, human testers traditionally shoulder the responsibility for validation—identifying which user journeys require coverage, building test scripts, and maintaining them as workflows evolve.

The ease of code generation has shifted complexity rather than eliminated it. Humans now face amplified testing demands. Addressing this imbalance requires structural change in how teams approach quality assurance.

Propolis brings agent-driven, goal-and-output-focused test generation to Datadog's platform. The system permits agents to navigate applications autonomously, trace user interactions, and synthesize tests from conversational instructions.

Combining Datadog's observability capabilities with this approach yields a testing framework that validates code while generating the telemetry needed to assess system behavior—all without the brittleness inherent in manually coded tests.

Datadog will host a discussion with Ruxanda Lueck, group product manager for goal-based testing of AI features, and Matthew White, co-founder and former CTO of Propolis, on Wednesday, November 4, at 9 a.m. Pacific / 12 p.m. Eastern.

The conversation will explore how much test authorship teams can delegate today and whether agentic testing has reached a comparable inflection point to recent advances in code generation. Topics will include whether natural-language test specifications democratize quality assurance across broader teams, and the timeline for agents to independently validate agent-generated code without human involvement.

Agent-generated code has introduced friction across multiple SDLC phases that no single solution can fully resolve. Accelerating the testing phase, however, removes another barrier between conception and production deployment.

Key topics

  • How agents can validate user flows that teams did not anticipate
  • Crafting natural-language test specifications
  • Maintaining goal-based test validity across changing user flows
  • Testing AI-driven features that produce probabilistic rather than deterministic results
  • Scenarios where traditional scripted tests remain valuable