Agentic Engineering & ToolingConference50min
Software engineering in the age of AI agents: who’s in control?
This talk explores how generative AI is changing software engineering—from coding to verification and guidance. It shares lessons, successes, failures, and responsible practices for using fast-evolving AI tools. The key message: even when AI does the work, humans remain accountable for the outcome.
talk.summaryAiDisclaimer
Bert Jan SchrijverOpenValue
talkDetail.whenAndWhere
Thursday, October 8, 13:50-14:40
Room 5
An AI deletes a production database, and then says it panicked. Welcome to modern software engineering.
Agentic coding assistants are writing more code than ever, creating a flood of code that needs review. But are we producing better software faster, or just more software faster? In a 2025 randomised trial, experienced developers were 19% slower with AI while believing they were 20% faster. By 2026, the follow-up had become unreliable: developers increasingly refused to take part if it meant working without AI.
The real shift isn't about writing more code faster. It's about moving from writing code to writing specs, from building systems to generating them, and from trusting output to verifying it. The prompt that starts an agent matters less than the condition that stops it. And whoever writes that condition is still in control.
In this talk I'll share what I learned: some of it the hard way, from real production projects and from talking to engineering leaders. I’ll tell you what happens when an agent gets write access to production, how I got from 45 minutes of specs, spread over three evenings, to a working app for my kids, and what building three small projects in a single hour actually taught me. You'll leave with a practical framework: experiment boldly, codify what "good" looks like, and raise the bar on verification. One rule applies no matter what: if you bring it to a review, it's your code.
Agentic coding assistants are writing more code than ever, creating a flood of code that needs review. But are we producing better software faster, or just more software faster? In a 2025 randomised trial, experienced developers were 19% slower with AI while believing they were 20% faster. By 2026, the follow-up had become unreliable: developers increasingly refused to take part if it meant working without AI.
The real shift isn't about writing more code faster. It's about moving from writing code to writing specs, from building systems to generating them, and from trusting output to verifying it. The prompt that starts an agent matters less than the condition that stops it. And whoever writes that condition is still in control.
In this talk I'll share what I learned: some of it the hard way, from real production projects and from talking to engineering leaders. I’ll tell you what happens when an agent gets write access to production, how I got from 45 minutes of specs, spread over three evenings, to a working app for my kids, and what building three small projects in a single hour actually taught me. You'll leave with a practical framework: experiment boldly, codify what "good" looks like, and raise the bar on verification. One rule applies no matter what: if you bring it to a review, it's your code.
Bert Jan Schrijver
Bert Jan is CTO at OpenValue and focuses on Java, AI, software architecture, Continuous Delivery and DevOps. Bert Jan is a Java Champion, JavaOne Rock Star speaker, Duke's Choice Award winner and leads NLJUG, the Dutch Java User Group. He loves to share his experience by speaking at conferences, writing for the Dutch Java magazine and helping out Devoxx4Kids with teaching kids how to code. Bert Jan is easily reachable on BlueSky via @bjschrijver.dev.