Development PracticesConference - Short25min
Teaching Agents How to Code: Iterative Knowledge Capture in Test Engineering
A talk on preventing AI-generated test code from becoming technical debt. It shares lessons from a large e2e test migration, why AI code becomes unmaintainable, what failed in practice, what worked, and how these lessons apply beyond testing. Targeted at engineers, QA, and tech leads.
talk.summaryAiDisclaimer
Bartosz CytrowskiDropbox
talkDetail.whenAndWhere
Thursday, June 18, 18:15-18:40
Room 4B
talks.roomOccupancytalks.noOccupancyInfo
You've asked AI to write the code. Then spent an hour debugging it. Everyone has. Now imagine your entire test suite written that way - unmaintainable code that rots by the month. Multiply that by your team's size.
This was our reality when starting a large e2e test migration. The question wasn't whether engineers would use AI to write tests (they would). The question was: how do we stop this from becoming technical debt?
You'll leave understanding:
• Why AI-generated code becomes unmaintainable
• The mistakes we made that didn't survive contact with reality
• What worked, and more importantly, why it worked
• How the problem generalizes beyond just test automation
• What questions you should be asking about AI in your workflow
This isn't an AI hype talk. It's about the unglamorous work of keeping AI-generated code maintainable when everyone on your team is using it.
Target audience: Engineers, QA professionals, and tech leads using AI for code and worried about what happens next.
This was our reality when starting a large e2e test migration. The question wasn't whether engineers would use AI to write tests (they would). The question was: how do we stop this from becoming technical debt?
You'll leave understanding:
• Why AI-generated code becomes unmaintainable
• The mistakes we made that didn't survive contact with reality
• What worked, and more importantly, why it worked
• How the problem generalizes beyond just test automation
• What questions you should be asking about AI in your workflow
This isn't an AI hype talk. It's about the unglamorous work of keeping AI-generated code maintainable when everyone on your team is using it.
Target audience: Engineers, QA professionals, and tech leads using AI for code and worried about what happens next.
Bartosz Cytrowski
Bartosz Cytrowski is a Senior Software Engineer at Dropbox Sign, specializing in LLM-augmented development workflows designed to compound learning rather than create technical debt. Working in web development since 2006, he brings experience across engineering, product development, and mentoring. He’s a big believer in human-in-the-loop systems, sees AI as a force multiplier for engineers, and is always exploring what’s emerging on the IT horizon