Program JavaZone

Foredrag Torsdag 3. september

Trust, But Verify: Skill-Driven Development for the Sceptical Java Developer

Room 5

Norsk 1 t Skill-Driven Development · verification · Java · software craftsmanship · AI tools

Totto - Thor Henning Hetland

Thor Henning Hetland (Totto) has been writing Java professionally for 30 years — before generics, before lambdas, before the JVM had a JIT. He co-founded JavaZone in 2002.

Today he runs eXOReaction, a software architecture practice, and is CTO at Mynder, a Norwegian compliance SaaS. He built Synthesis, an open-source knowledge infrastructure platform used across large Java codebases, and the Knowledge Context Protocol (KCP), a proposed standard for structured agent context.

The lib-pcb project — 197,831 lines of Java, 7,461 tests, eleven days — is the concrete answer to the question this workshop asks. His expertise grew stronger through the process, not weaker. That's the point.

Something uncomfortable is happening to Java developers right now. You're shipping more code than ever. Your PRs are larger. Your velocity metrics look great. And somewhere between the third suggestion you accepted without reading and the afternoon you spent debugging code you didn't fully write, a question starts forming: am I getting better at this, or worse?

We call this Fear-Driven Development. Not the deadline-pressure kind — something subtler. The growing unease that your tools are making you faster while making you weaker.

This talk is built around that fear. And around six systems we built in response to it.

Over the past year, those systems — collectively called Skill-Driven Development — were tested at production scale: a 197,000-line Java library in eleven days, no existing libraries in the domain, 7,461 tests at 99.8% pass rate. And more recently, in production SaaS systems where the same discipline caught AI-generated code that passed all tests but contained subtle GDPR retention logic errors and binary encoding mistakes that would have shipped in any copy-paste workflow.

The intent here is to move to a BOF format where thesession is interactive. We'll work through real examples: skill files, verification checklists, and the actual failure modes — code that looked correct but wasn't. You'll leave with a concrete framework you can apply starting Monday.

Format: workshop, small group, discussion-based. Bring your actual questions — the ones you haven't been able to ask out loud.

Senior Java developers and tech leads who already use AI coding tools and are responsible for code quality and team standards. No introduction to AI tools — this session starts where the marketing ends.

  1. Getting more out of Maven Room 1
  2. "35 Minutes Ago, You Had 20 Seconds to Comply": A Survival Guide to Days of Future Past Room 2
  3. Feature Management: Beyond Feature Flags Room 3
  4. Brukollapsen som bygde en bro mellom fag og IT Room 4
  5. Tilgjengelighet + utvikling = true Room 6
  6. Flyt i AI-ens tid: når det blir lettere å lage, men vanskeligere å bevege seg Room 7
  7. The Decision Layer: Context Graphs for Spring AI Room 8
  8. Rendering 3D shadows in the browser with three.js Room 6
  9. Du har lov til å sørge, men du må gå videre i livet Room 6