Summary
Here’s what I learned from nine production projects with Claude Code:
- You must use Coding Agents if you haven’t yet!
- “Is AI right?” is the most important question.
- Treat Coding Agents as a non-deterministic, forgetful junior team that you should let write and test.
- Skills automate tasks, Subagents write better code faster.
- AI will steal “writing code” but not yet “AI team lead”.
Please see my slides for an explanation of why I came to this conclusion.
Logistics
Location
I’ll give the talk at JAX 2026.
Talk
My talk is part of the JAX 2026 program — date and time are on the session page.
Abstract
Deutsch
Ich berichte von meinen Erfahrungen mit Claude Code in neun Produktionsanwendungen seit Juli 2025: Cloud-Microservices mit Kotlin, Java-Monolithen, Web-Anwendungen, mobile App und Marketing-Webseiten. Vier dieser Projekte waren neu.
Was habe ich gelernt? Für uns Enterprise-Entwickler bedeutet Agentic AI, viele „Junior-Ichs” anzuleiten, die den Großteil unserer Arbeit übernehmen können. Und das ist mehr als nur Code zu schreiben! Wie alle Junior-Entwickler müssen wir die AI einarbeiten, ihre Pläne prüfen und ihr beibringen, wann sie tatsächlich fertig ist. Anders als wir vergisst die AI aber leider alles, was wir ihr sagen — und lernt auch nichts von ihrer Arbeit für uns.
English
I’m reporting my experiences with Claude Code across nine production applications since July 2025: cloud microservices with Kotlin, Java monoliths, web applications, a mobile app, and marketing websites. Four of these projects were brand-new.
What have I learned? For us enterprise developers, Agentic AI means guiding many “Junior Me” who can take over the bulk of our work. And that’s more than just writing code! Like all junior developers, we have to onboard the AI, review its plans, and teach it when it’s actually finished. Unlike us, however, the AI unfortunately forgets everything we tell it — and doesn’t learn anything from the work it does for us.
Who Made Me the Expert?
I’ve worked on nine production systems with Claude Code. I teach a sustainable, principled approach, not short-term tuning that’s outdated tomorrow. I’m also a Senior Editor in the InfoQ Java Team, which gives me valuable insights into today’s software industry.
Slides
Repo with Skills & Subagents
The Sample Application
The repository with the sample CRM application from the talk is at github.com/atra-consulting/coding-with-ai-lab. The README describes the requirements and how to get it started.
Skills & Subagents
The skills are here, and the subagents are here. Both are for Claude Code and Claude Desktop. There are outdated versions for Google Gemini that are no longer actively maintained.
Please see the talk slides for a description of the agents and skills.
You can use the skills and subagents in your own projects. Copy the agents into your project. Then tell Claude:
Adapt these skills/agents to my technology stack here. Think hard when you do, and review them.
Training
atra.consulting offers AI training, including a hands-on workshop on agent-based coding.
Contact
You can contact the speaker by email or on LinkedIn.
Senior Principal Consultant Software Engineering
Karsten arbeitet seit 26 Jahren als Full-Stack-Java-Entwickler in Europa und den USA. Als Mitgründer leitete er 13 Jahre lang die Produktentwicklung eines Software-Produkt-Start-ups in den USA, das 2016 erfolgreich verkauft wurde. Im Jahr 2020 war er Mitgründer eines SaaS-Start-ups in England. Er ist seit 2003 auch Freiberufler. Seit Mai 2025 arbeitet er für atra.consulting in Stuttgart.