Date & time

17:00

AI-coding agents produce changes faster than developers can read them, but engineers still own responsibility for the code they ship.

The LeadDev AI Code Review Report 2026 shows the strain. Half of the 377 engineers and leaders surveyed say pull requests per developer are up at least 25% since adopting AI tools. Furthermore, 45% now name code review capacity as their biggest delivery bottleneck and only 26% say they fully understand the AI-generated code they review.

Volume is only part of the problem: 68% say reviewers don’t consistently know which changes need deeper scrutiny before they open a pull request. Just 18% consistently have enough context to judge a change’s architectural impact and blast radius. Where reviewers do have that context, 40% report faster review cycles, compared with 25% overall.

Drawing on the report, this panel explores how your team can validate AI code it hasn’t read, preserve understanding and decide where human judgment belongs across the agentic SDLC.

We’ll look at how teams check changes against developer intent and codebase context, direct engineers’ attention to the most critical changes, and explain how those changes affect the wider system.

We’ll also discuss how security practices must adapt across the codebase, before and after deployment, as agents take on more development work.

Key takeaways:

• Validate AI code changes against developer intent and codebase context
• Prioritize human attention on the most critical code changes
• Understand how AI-generated code impacts the rest of the codebase
• Detect and address security risks across the codebase, before and after deployment