Register or log in to access this video
New York • September 8 & 9, 2027
Loved LDX3 New York? Pre-sale tickets for 2027 are now available.
Human-written code died in 2025. Code review is dying in 2026. Teams with high AI adoption are merging 98% more pull requests, but PR review time has surged 91%. There is no way we win this fight with manual code reviews, and AI code review tools are just buying us time. This talk makes the case that the traditional code review is a historical approval gate that no longer fits the shape of modern software development. We’ll walk through a practical five-layer trust model: from multi-agent competition and deterministic guardrails to spec-driven BDD and adversarial verification that lets engineering teams ship faster without sacrificing quality or control.
Key takeaways:
- Review time goes up 91%, manual code review cannot scale, and pretending otherwise leads to organizational debt.
- AI code review is just kicking the can. If an AI writes the code and an AI reviews it, the post-PR review stage adds ceremony without real value. The review loop needs to move earlier in the development cycle.
- Shift the human checkpoint upstream from code to intent. The most valuable human judgment is applied before the first line is generated: reviewing specs, constraints, and acceptance criteria, not 500-line diffs.
- Trust is layered, not binary. A practical five-layer model for agentic safety: (1) multi-agent competition, (2) deterministic guardrails and linters, (3) human-authored acceptance criteria via BDD, (4) scoped permission systems, and (5) adversarial verification agents.
- Bring back behavior-driven development. Code is an artifact of the spec. Behavior-Driven Development (BDD), once impractical overhead, becomes the primary workflow when agents handle implementation.
