Designing for the Agentic Era.

Open Wonder exists because classic brand guides stop being enough once generative production becomes operational. The platform takes the kind of brand-production service an agency would normally interpret manually and turns it into a product system that can generate, critique, and steer assets for real teams. Idealo, mobile.de, Hartmann, and Kneipp prove the system had to work under real brand pressure.

AURA shows the governed side of the same product logic. Internal AI workflows had to stay reviewable, privacy-conscious, and deployable inside enterprise conditions. Data-flow diagrams, privacy checklists, reliable environment tagging, and build fixes were part of the product because trust in a brand-generation system depends on governance and operational clarity.

Kneipp is the freshest proof that the system is moving closer to real production. In the new Product Studio Creative Placement Flow, the challenge was encoding brand-specific studio language, props, composition rules, colors, and concept space well enough that the result could feel like Kneipp — not a generic model output. I wrote prompt-enhancer templates with memory persistence using Codex: generations that evaluate their own results, iterate on prompts, and store adjustments so past decisions carry forward. Running on Gemini 2.0/2.1. Early tests moved ahead of traditional render and photo production for social use.

The system also had to operate like product. A Sanity integration let editors generate Open Wonder assets from inside a real CMS surface without dealing with tokens or model orchestration. Next.js 15, Supabase, Prisma, Sentry, Vercel, and OpenTelemetry had to survive sharp binaries, lockfile issues, environment confusion, and privacy requirements. A generative brand system becomes strategically useful when teams can rely on it.

EVG is the Eisenbahn- und Verkehrsgewerkschaft. Wo-Mo-Fonds is its member benefit platform for housing and mobility support: commuting costs, job tickets, and related subsidies. The work was to make that regulated process understandable, then replace a brittle external status page with backend-driven progress updates in plain language: what already happened, what step is current, how long it can take, and when to contact the fund team.

Dein WoMo is the internet and equipment voucher flow within the same service family. Here the hard part was conditional application logic: nested fields, uploads, legal help text, and error states that still had to stay accessible. I mapped FAQ content into inline guidance, tightened ARIA relationships, and kept branching validation readable instead of punitive.

The AI assistant extends the same service into multilingual guidance. Built on Cohere, it helps with benefit questions, deadlines, and applications in up to fifteen languages, using grounded sources instead of guesswork. The goal was to make the benefit system easier to understand while keeping privacy and escalation boundaries clear.

The TeamBank Markenportal is the clearest expression of the problem I was solving. In a landscape of multiple portals and agencies, it had to turn brand governance into a usable working surface: searchable assets, understandable navigation, and downloads that felt like part of a system rather than a pile of files.

The editor tooling behind that portal mattered just as much as the interface. We kept the WordPress Media Library separate from a dedicated Download custom post type so not every upload became public-facing content. That created real operational complexity, so I helped shape micro-plugins that checked media against existing downloads, bulk-created missing entries, cleaned filenames, assigned previews, and regenerated collection thumbnails through PHP and ImageMagick when grouped assets changed.

Search and discoverability were treated as shared infrastructure. Algolia was embedded and versioned in a way that let the same logic travel between WordPress and PHP surfaces instead of being reinvented in each portal. File metadata, collection previews, and later even auto-generated descriptions via a user-provided OpenAI Vision setup all fed the same goal: make assets and pages easier to find, and make that behavior reusable across the wider TeamBank and easyCredit landscape.

On teambank.de, the same UI logic had to hold under different constraints. Corporate communication, recruiting, and brand presentation still needed to feel connected to the wider system instead of drifting into their own interface language.