Verifying AI-authored code: Teaching your agents how to correct their own output
AI coding has fundamentally changed how software is written and improvements are real. Tools like Claude Code and Cursor speed up developer velocity and change how we approach software engineering, leading to better designs.
However, as AI is writing more code than ever, testing and verification is the new bottleneck.
Join this workshop, if you are running into issues like:
Are you and your team stretched thin because you are dealing with more and larger PRs, and spend most of your time doing code reviews?
Are your agents writing tests that mostly cover the obvious path, but fail to find the real bugs affecting your customers?
Are you and your team still hunting for subtle, hard-to-find bugs despite the use of AI?
Are your code reviews taking longer because you no longer hold the mental map of the codebase built through manual development?
In this workshop, we will show a fundamentally different approach to solving the verification gap.
Instead of trying to guess every failure mode up front, we define the invariants that should always hold true for the system, and then put your software through a rigorous environment that systematically and deterministically looks for ways to break these boundaries.
We will cover how this approach helps surface the kinds of failures that are still painful in modern systems, including crashes, data correctness bugs, fault-tolerance bugs, and other hard-to-reproduce issues.
Once we find a failure, we will pinpoint the root cause and give you and your coding agents detailed debugging information to iterate and fix issues faster.
This can integrate with existing workflows and evolve with the way your team builds software today.
To make this a productive workshop, we also encourage you to share your day-to-day realities in your own engineering workflows and foster an open discussion on how AI-assisted engineering can and should evolve.
If you are looking to increase your productivity and want a more rigorous approach to finding, reproducing, and fixing bugs that slip past conventional testing, we would love to have you join us.
Space is limited. Please register soon!