Reduce Developer Toil: Autonomous AI for Enterprise Testing @ Scale
summary
Claim: a reinforcement learning agent autonomously generates Java unit tests that compile, pass and self-verifying tests without human review, targeting coverage while minimizing test count. Implementation couples fuzzing with RL to explore inputs and context (Spring annotations, mocks, domain patterns) and runs as an on‑prem CLI and CI integration plus IDE plugin. Measured outcomes: 50–250× faster test generation and up to 95% reduction in developer test-writing time across large codebases.
AI-written summary · about
chapters (19)
- 0:00Introduction and agenda
- 1:02Speaker background
- 2:27Modern engineering challenges
- 6:09Developer intelligence debate
- 8:17Developer toil and AI's role
- 12:36Useful testing principles
- 15:46AI for autonomous testing
- 17:30Assistants vs autonomous agents
- 20:21Reinforcement learning approach
- 21:24Fuzzing and test generation
- 26:47Probabilistic search in testing
- 27:30Testing in refactoring workflows
- 30:01Live demo of Diffblue
- 36:49Modern and Diffblue integration
- 39:00AI agents in testing environments
- 39:42Q&A on junk tests and coverage
- 51:09Q&A on test tracking and workflow
- 55:33Q&A on deployment and integration
- 1:01:21Closing remarks
related talks
39mRevolutionizing Test Automation: Generative AI To Empower Prod Mgmt, Accelerate SDLC
47mTDD & Generative AI: A Perfect Pairing
39mRethinking Unit Testing: Automating the Generation of Java
48mTDD & generative AI - a perfect pairing?
16mTransforming Test Automation with AI and Machine Learning
18mLeveraging AI to Improve the Developer Experience
43mAre Your Tests Slowing You Down?
06mTriple Threat: AI Technologies Reduce Testing Costs