Dedy Kredo is the co-founder and chief product officer of Qodo, formerly CodiumAI, and a co-author of AlphaCodium, an open-source framework for improving AI-generated code through iterative testing. His work focuses on a central challenge of automated programming: ensuring that increasingly easy-to-generate software actually behaves correctly.
Kredo previously managed a product line at VMware, co-founded an analytics business focused on online purchasing decisions, and led customer-facing data science at Explorium. In 2022, he co-founded CodiumAI with chief executive Itamar Friedman and became its product leader. The company rebranded as Qodo in 2024 as its software expanded from automated testing into code generation and pull-request review.
- A critic for generated code. Kredo advocates pairing code-generation systems with an independent verification component that analyzes developer intent, functional behavior, edge cases, and security risks. His AI Engineer Summit demonstration applied this architecture to test generation, iterative repair, pull-request review, and exposed-credential detection. The design borrows the generator-and-critic structure of adversarial networks without requiring adversarial model training.
- Behavior coverage over line coverage. Kredo prioritizes tests that exercise meaningful outcomes and failure modes over raw percentages of executed code. His work on Qodo Cover extends automated testing into continuous integration and regression prevention.
- Test-oriented flow engineering. The AlphaCodium paper, co-authored with Tal Ridnik and Friedman, describes a structured cycle of understanding programming problems, proposing solutions, generating tests, executing code, and correcting failures. Its open-source implementation makes that iterative approach available to other developers.
- Context-aware multi-agent code review. Kredo introduced Qodo 2.0 in 2026 as a review system combining specialized agents, repository context, and organizational standards. His model-evaluation work similarly judges pull-request feedback by its precision, relevance, and usefulness to developers.
