← Back to Case Studies
AEM Delivery · 2026
Faster AEM builds and AI-assisted authoring for a US hotel group
Full-stack AEM work on a team of more than ten developers: a test strategy agreed in an ADR, CI cut from about 16 to 3.0 minutes per run, and AI tooling that took page authoring from 5 to 50+ pages a day per author.
By Laurentiu NichiteanPublished
My Role:
Full-Stack AEM Developer
Client:
US hotel group (via a digital agency)
Core Tech Stack:
Java 21, Sling Models, OSGi, HTL, dispatcher, Claude
The Business Challenge
More than ten developers shared one AEM codebase, and every pull request waited about 16 minutes for CI. Slow feedback makes developers batch changes and makes reviews harder.
Content authors were building pages by hand, at about five pages a day each, while the team was starting to use AI coding tools without shared rules.
What I Did
- Test strategy as an ADR: I wrote the test-strategy architecture decision record, so the team agreed on the approach before the pipeline changed.
- CI from ~16 to 3.0 minutes: a durable 3.0 minutes per run for the whole team, measured over 1,871 runs.
- Rules for AI coding tools: CLAUDE.md and Cursor rules, with AGENTS.md for other tools, plus a rollout plan, so AI-assisted code follows the codebase conventions.
- AI-assisted page authoring: a custom Claude Desktop skill that automates page authoring in Production, used by 16 authors.
- Day-to-day AEM delivery: Java 21 Sling Models, OSGi services, HTL components, Touch UI dialogs and dispatcher configuration, as a top-three contributor over three months.
Before and after
Outcomes
- About 5× faster CI feedback on every pull request, for a team of 10+.
- About 10× authoring throughput per author, across 16 authors.
- Shared rules for AI coding tools, so faster output still follows the team conventions.