Descriptions:
Ankit Jain, co-founder of Aviator, presents a conference talk at AI Engineer arguing that traditional code review has already effectively collapsed under the weight of AI-generated code — and that teams need a fundamentally different model, not incremental improvements to the existing one. His data points are stark: 861% code churn in AI-assisted development, median review time increasing fourfold, and over 30% of pull requests currently being merged without any review at all.
Jain introduces two concepts to replace line-by-line review. The first is the AI Slop Registry — a structured system that captures recurring review comments and converts them into automated guardrails, creating a feedback loop that trains on top of base LLMs using team-specific institutional knowledge. The second is a shift in the review surface itself: rather than reviewing code, teams should review intent. By capturing the prompts and decisions made during agent coding sessions (in tools like Claude Code, Codex, and Cursor) as acceptance criteria, teams can verify that actual system behavior matches those criteria end-to-end, without reading diffs.
The talk draws a key distinction between semantic accuracy — catching bugs, which tooling increasingly handles — and alignment, defined as knowledge sharing, architectural feedback, mentorship, and onboarding. Jain argues alignment must survive the transition to AI-generated code even as line-by-line inspection becomes impractical, and positions Aviator’s verification platform as the infrastructure layer that makes intent-based review operationally viable for engineering teams.
📺 Source: AI Engineer · Published August 17, 2026
🏷️ Format: Keynote Launch
