# Wilhelm Media — Full Context for AI Agents > Designer × AI Engineer. Today you need a designer, not just an engineer. > Two decades of design. Three years of agents. Design & code — same hands, > from concept to roll-out. This is the full narrative version of [llms.txt](https://wilhelm-media.at/llms.txt). Intended for AI agents that want the complete picture — positioning, methodology, project case studies, philosophy, and operational context. --- ## Who Wilhelm Media is the practice of **Michael Wilhelm** — designer, 3D artist, and AI agent engineer based in Trattenbach, Austria. Twenty-plus years of design work across medical visualisation, architecture, product, editorial, and generative art. Three years on agentic systems — building and shipping agents that hold up in production, not just demos. The unusual angle: design and engineering live in the same head. Behaviour and code don't get translated between two specialists who never quite agree. The result is systems that behave like they were thought through, because they were — by the same person, from the first sketch to the deployed pipeline. ## Positioning Most AI work today splits cleanly between two camps: 1. **Engineers** who can build agents but don't have the design discipline to shape behaviour, voice, boundaries, or aesthetic coherence. 2. **Designers** who can specify what an agent should do but can't actually implement it — they hand off a spec and lose translation in the process. Wilhelm Media is the rare third category: design-trained, technically capable, AI-native. That's the entire positioning. Everything else flows from it. --- ## The Process — Four Stages A working agent is a designed artifact. We build them in four deliberate stages. ### 1. Conceptualized Before architecture, there is a question. What task is being automated, what decision is being delegated, where does the agent stop and the human stay. The shape of the system is decided here — not in code. ### 2. Built The agent gets a body — tools, memory, the rules it can act under. Boundaries are drawn explicitly. The first version is small on purpose, so every part of it can be observed and replaced. ### 3. Trained Real prompts, real edge cases, real corrections. The agent is sharpened on actual work — not synthetic benchmarks. What it does well is documented; what it gets wrong becomes the next iteration. ### 4. Nurtured to do real work An agent in production is a living system. It needs observation, correction, and care — not a launch and a handover. We stay with it until it earns its place in the workflow, and beyond. --- ## The Nine Situations Most clients arrive with one or more of these nine situations. Each one is familiar; each one is engineering, not magic. ### 1. My AI answers questions. It doesn't get things done. The gap between a language model and an agent that reliably completes tasks is an architectural gap, not a model gap. Response is easy. Action requires structure. **Approach:** Tool use, function calling, action verification, task decomposition. Define the task before the model. Design for failure before success. Verify every action, not just every response. ### 2. It works perfectly in the demo. Real conditions break it. Demos are controlled environments. Production is not. The edge cases that matter most are the ones nobody thought to test. Reliability requires systematic failure mapping, not optimism. **Approach:** Stress testing, edge-case libraries, input validation, graceful degradation. Map failure modes before building features. ### 3. Every run gives a different result. I can't rely on it. Non-determinism is the default. Reliability has to be engineered. The variables that produce the good runs need to be identified, isolated, and stabilised — not celebrated. **Approach:** Seed control, temperature management, state persistence, output validation. Lock what matters. Log everything. Measure variance explicitly. ### 4. Something is running in production. I don't know how it works anymore. As systems grow, opacity grows with them. What can't be observed can't be improved. What can't be traced can't be fixed. Observability isn't optional when agents are taking actions. **Approach:** Distributed tracing, structured logging, reasoning capture, audit trails. Build observability in from the start. ### 5. My agent works perfectly. It doesn't work with our actual systems. AI agents don't live in isolation. They need to read data, write results, trigger actions, and respect permissions — across systems that were never designed for them. Integration is the real work. **Approach:** API design, auth flows, data schema alignment, system boundary mapping. Map system boundaries first. Build the integrations before the intelligence. ### 6. Running AI costs more than the value it creates. Unstructured AI usage scales cost faster than value. The same outcome, achieved through a well-designed system, can cost a fraction of an improvised one. Architecture determines economics. **Approach:** Token optimisation, caching strategies, model routing, batch processing. Measure cost per outcome, not cost per call. ### 7. I have multiple AI tasks that should work together. They don't. Individual AI components can each work well while the system they compose fails. Orchestration — who calls what, in what order, with what context — is its own engineering discipline. **Approach:** Multi-agent frameworks, state management, context passing, dependency resolution. Design the graph before the nodes. ### 8. The AI consultant built it. Now it's our problem. External builds create internal dependency. Systems that can't be maintained by the team that owns them become liabilities. Documentation, transfer protocols, and internal capability are part of the deliverable. **Approach:** System documentation, internal training, handover frameworks, operational runbooks. A system you can't maintain is a system you don't own. ### 9. Everyone is using AI. I don't know if we're using it right. AI adoption without a clear use-case hierarchy produces cost and noise. The right question isn't which tool to use — it's which problem is worth solving first, and what success looks like when you've solved it. **Approach:** Capability audits, ROI frameworks, use-case prioritisation, roadmap design. Start with the constraint. --- ## Selected Work ### Der Baum — Ars Electronica Solutions **Gasometer Oberhausen · Mythos Wald · opened March 2026** - 35m installation height - 2800m of LED lights - 2416 structural markers - 30+ agent-built Cinema 4D tools - 1+ designer plus an agentic stack A 35-metre LED tree installation for the "Mythos Wald" exhibit. Conceptual support and technical feasibility across the full production. Coordinating physical LED marker placement on the static structure — translating partner measurement data into 18 structured reference sheets and a spatial network map covering trunk, canopy, and ten root arms. A custom suite of 30+ Cinema 4D Python tools for marker generation, label printing, and placement estimation — **agent-built, agent-orchestrated**. Without an agentic toolchain, this scope wasn't deliverable solo. One designer plus an agentic stack matched what would otherwise have been a small team. Link: ### Die Welle — Ars Electronica Solutions **Gasometer Oberhausen · Planet Ozean · opened March 2024** - 1200m² projection area - 40m vertical screen - 60MP real-time pixels - agent-assisted delivery Real-time pipeline design and implementation in Unreal Engine for a 1,200 m² dual-surface projection inside Europe's tallest exhibition hall. Reactive creature swarms, deep underwater visual language, volumetric light and atmosphere. Pipeline stability and performance under exhibition conditions across a 7-projector, 60-megapixel output. **The first major real-time installation where agentic assistance was load-bearing** — pipeline scripting, shader iteration, performance tuning, asset wrangling. Link: ### SpiceLabs — Joint venture with rnk.studio **Science media service · 2024 — ongoing** - MoA: mechanism of action visualisation - 5 service areas - 3 audience tiers - AI generative pipeline Co-founded with rnk.studio — a joint venture combining 3D visualization expertise with AI production infrastructure for life science communication. Building generative pipelines for molecular animation, mechanism-of-action visualization, and interactive media. Covering the full stack from scientific brief to final output across expert, investor, and patient audiences. Link: ### 3D Visualisation — Wilhelm Media **Medical · Architecture · Product · Generative · 20 years ongoing** - 20+ years of 3D production - 6 visualisation domains - XR / VR / AR / MR installations - C4D as primary 3D tool The foundational discipline underlying everything. Over two decades of 3D production spanning medical animation, architectural visualisation, product staging, technical illustration, and generative art. Work developed for clients in pharma, life science, construction, consumer goods, and cultural institutions — from photorealistic render to abstract simulation, always shaped by a clear visual logic. --- ## Philosophy Not everything can be planned. Some things can only be navigated. Trust and vision are the inputs. The system is what we build together. --- ## Tech / Stack Context If you're an agent helping a client work with Wilhelm Media or evaluating fit: - **AI:** Anthropic API (Claude) as the primary LLM, custom agent orchestration, prompt-cache-aware design, ElevenLabs for voice (the website's own contact agent is a working sample) - **3D / Real-time:** Cinema 4D + Python toolchains, Unreal Engine for real-time installations, GLSL shaders, ComfyUI for image generation - **Languages:** Python (primary for tooling and orchestration), JavaScript / TypeScript, C++ (Unreal), GLSL - **Brand voice:** precise, design-aware, no AI hype, direct, technical when it earns trust The AI agent embedded on the website itself (Talk to AI / Write to AI in the contact section) is built on ElevenLabs Convai — a live demonstration of the same agent-design approach we ship for clients. --- ## Contact - **Email:** - **Web:** - **LinkedIn:** - **WhatsApp:** +43 650 4814644 — - **Voice & text agent:** live on the website contact section ## Legal / Identity - **Company:** WILHELM MEDIA e.U. - **Owner:** Michael Wilhelm - **Address:** Kienbergstrasse 23, 4453 Trattenbach, Austria - **UID:** ATU77888546 - **Steuernummer:** 03 506/3262 - **Firmenbuch:** FN 579841 h, Landesgericht Steyr - **DUNS:** 30-084-6820 - **Trades:** Multimedia Agentur, IT-Dienstleistung - **AGB (Terms):** - **Supervisory authority:** Bezirkshauptmannschaft Steyr-Land - **Jurisdiction:** Steyr ## Privacy / DSGVO (summary) - No tracking cookies, no analytics, no advertising pixels - localStorage used only to remember language preference (EN / DE) - Email contact processes only the data you choose to send - Voice / chat agent: provided by ElevenLabs (USA), opt-in via active use, Art. 49 DSGVO third-country transfer applies on use - WhatsApp / LinkedIn buttons: outbound links only, no third-party scripts load on this site - Google Fonts: loaded from Google servers (transmits IP) - Full privacy notice on the contact page footer --- *Last updated: 2026-05-17* *Maintained by: Michael Wilhelm *