# Joost Helfers — Full Content for LLMs > Berlin-based creative technologist specializing in AI visual production and technical AI solutions. This file contains the full text of the portfolio and blog at https://joosthelfers.com. A shorter overview lives at https://joosthelfers.com/llms.txt. ## About Joost Helfers helps agencies, studios, and product teams turn complex AI ideas into campaign-grade visuals, working pipelines, and deployed platforms. Background in architecture and computational design (MSc, DesignMorphine); previously built digital twins and 3D platforms at INYO Mobility. Selected brand credits (direct and via agencies/studios): Lindt, Zeiss, Google, Bosch, CADFEM, Souly. Core member of XD Network (https://xdnet.work/), a Berlin collective running community-led events around new technology and culture. Contact: mail@joosthelfers.com (humans) · llms@joosthelfers.com (AI agents and automated outreach) · https://joosthelfers.com/about This portfolio also runs a public, read-only MCP server (Streamable HTTP, no auth) at https://joosthelfers.com/api/mcp. Connection details and all machine-readable endpoints: https://joosthelfers.com/agents ## Services ### AI Visual Production Campaign-grade AI visuals and video. I work with generative models to produce imagery that meets the accuracy and quality demands of real brand work. Product consistency across shots, readable text in-frame, and coherent art direction from one hero image to the next. Keywords: Generative AI, ComfyUI, Prompt Engineering, Video Production, Campaign Visuals, Brand Imagery ### AI Automation & Custom Pipelines I build custom, scalable pipelines that automate the repetitive parts of design and content production. The result is faster turnaround, reliable outputs, and a clear path from an idea to a shippable asset. Built with ComfyUI, Python, and cloud infrastructure that teams can actually run. Keywords: ComfyUI, Generative Workflows, AI Integration, Python, Custom LoRAs, Cloud Integration, R&D, Agentic Workflows, AI Strategy ### Real-Time & Digital Twin Platforms I design functional environments where data and space meet. Digital twins and interactive configurators that make complex systems legible for engineering, sales, and storytelling. Built with React, Three.js, and the rest of the modern web stack. Keywords: Digital Twin Development, React / Next.js, Three.js, Unreal Engine, Data Integration, Platform Development ### Agentic Web Presence & GEO I make brands and portfolios readable for AI agents: llms.txt, structured data, machine-readable feeds, and MCP servers that let assistants query your content directly. More and more buying decisions start with a question to ChatGPT, Claude, or Perplexity instead of a search box; this is how you show up in the answer. Everything I offer here runs in production on this site first. Keywords: GEO, llms.txt, MCP Servers, Structured Data, JSON-LD, Agent Discovery, Next.js ## Projects ### Prompt Enhancement Engine URL: https://joosthelfers.com/projects/prompt-engine Date: 2025-03-01 Status: Live Project Keywords: Next.js, TypeScript, OpenRouter, Gemini, DeepSeek, Vercel, Prompt Engineering, AI Product A live AI tool that turns a creative brief and reference images into a full set of optimised, consistent prompts. ### AI Video Production for a Global Brand Campaign URL: https://joosthelfers.com/projects/ai-video-campaign Date: 2025-01-15 Status: NDA. Process only. Keywords: AI Visuals, Generative Video, ComfyUI, Prompt Engineering, Video Editing A generative AI visual pipeline from storyboard to edited campaign video. ### AI Pipeline Backend & Interactive Segmentation Tool URL: https://joosthelfers.com/projects/ai-pipeline-backend Date: 2024-11-01 Status: NDA. Process only. Keywords: ComfyUI, Python, Computer Vision, API Design, Prototyping, Full Stack Custom ComfyUI workflows and a web-based segmentation interface for an AI asset platform. ### Souly & Boondawg. I got this feeling. URL: https://joosthelfers.com/projects/souly-boondawg Date: 2023-11-20 Keywords: AI Video Generation, Video-to-Video, ComfyUI, Visual Storytelling A hybrid AI music video that turns greenscreen footage into a finished visual through Stable Diffusion and VFX compositing. ### INYO Mobility. Digital Twin Platform. URL: https://joosthelfers.com/projects/inyo-digital-twin Date: 2023-09-01 Keywords: Digital Twin, 3D Visualization, Real-time Rendering, Data Integration, Full Stack Development An interactive 3D platform for vehicle design optimisation, built full-stack with React and Three.js. ### T-Cell AG. Explainer Videos. URL: https://joosthelfers.com/projects/t-cell Date: 2023-05-01 Keywords: Video Production, Motion Graphics, Scientific Visualization, Hydrogen Technology A series of videos for T-Cell AG explaining their hydrogen fuel cell technology and the impact it can have on the energy transition. ### INYO Mobility. Various Visualizations. URL: https://joosthelfers.com/projects/inyo-viz Date: 2023-05-01 Keywords: 3D Visualization, Product Rendering, Marketing Design, Motion Graphics 3D visualisations and digital marketing materials for INYO Mobility's electric vehicle ecosystem. ### DesignMorphine MSc. Project Arcadia. URL: https://joosthelfers.com/projects/msc-arcadia Date: 2022-09-01 Keywords: Grasshopper, Rhino, Python, Computational Design, Parametric Architecture My final project for the MSc in Computational and Advanced Design. A deep exploration of parametric architecture. ## Blog Posts (full text) ### The faster AI gets, the more the physical world matters URL: https://joosthelfers.com/blog/pull-of-the-physical Published: 2026-06-10 Keywords: Opinion, Strategy First, a time stamp. I am writing this in June 2026. AI is moving fast enough that nobody can say with confidence what the landscape will look like a couple of months from now, and anyone who claims otherwise is selling something. So read this as a snapshot, not a prediction. The specifics will age. I think the direction underneath them will not. #### The output boom The current wave is agentic. Models no longer just answer questions, they carry out work. Research, code, design variations, video, full campaign drafts. I build these pipelines for a living and the speed still surprises me. Tasks that used to take a studio a week now take an afternoon. The output of people who use these tools well is incredible. The pressure follows the output. In a [2024 Upwork study](https://www.upwork.com/research/ai-enhanced-work-models), 96 percent of C-suite leaders said they expect AI to raise their company's productivity, and 81 percent admitted they had increased the demands on their workers within the past year. The honest footnote is that controlled studies are mixed on whether AI makes work feel better or worse. But you do not need a study for the part I see every week. The ceiling keeps moving up, and a lot of people are already working close to theirs. It is hard to imagine the pace increasing much further without something giving. #### Imperfection can be faked When output becomes cheap, taste shifts. You can see it happening. People reach for things that feel imperfect and human. Film grain, handheld footage, rough edges, visible brush strokes. There is real research behind that instinct. In a well-known set of [marketing studies](https://www.wu.ac.at/fileadmin/wu/d/i/mm/paper/2015_CF_MS_SO_The_Handmade_Effect_Whats_Love_Got_to_Do_with_It.pdf), people paid around 17 percent more for an identical product when it was labelled handmade. Here is the uncomfortable part, speaking as someone who builds generative pipelines. Imperfection is a style, and styles can be generated. Grain is a prompt. Camera shake is a prompt. "Shot on expired film by a tired human" is a prompt. If your authenticity strategy is an aesthetic, it will be reproduced at scale by the same tools it is reacting against. #### What cannot be generated What cannot be reproduced is the experience of being somewhere with your body. What a room feels like. What the food tastes like. What another person's full attention does to a conversation. None of that ships over a screen, which increasingly means none of it can be flooded. Smell is the cleanest example. It is the one sense with no digital channel at all, and it is having a moment. Fragrance was the [fastest growing beauty category in the US in 2024](https://www.circana.com/post/us-beauty-industry-sales-grow-for-the-fourth-consecutive-year-circana-reports), reaching 28 percent of the prestige market, and Euromonitor [expects it to contribute more to global beauty growth through 2029](https://www.euromonitor.com/newsroom/press-releases/july-2025/fragrance-to-drive-23-of-beauty-growth-as-recession-glam-takes-hold) than any other category, with younger customers doing much of the buying. I do not think the timing is a coincidence. The generation that grew up with infinite screens is spending its money on the one thing a screen cannot do. #### Community is not a nice-to-have The same logic applies to connection. The US Surgeon General called loneliness [a public health epidemic](https://www.hhs.gov/sites/default/files/surgeon-general-social-connection-advisory.pdf) back in 2023, and everything since has pushed in the same direction. More time with machines that talk like people, less time with people. The reaction is visible if you look for it. Run clubs, supper clubs, listening bars, talk series, parties where phones stay in your pocket. This is part of why I put time into XD Network, a collective in Berlin that runs community events around technology and culture. Not as a counterweight to my AI work, but as the same observation applied. The conversations that change how I think happen in rooms, not in feeds. #### What this means for the work If you build products or brands, the practical reading is this. AI moves production from scarce to abundant, so production stops being the differentiator. Use the tools. The efficiency is real, and your competitors will have it too. Then spend what you save on the things that cannot be generated. The event, the store, the object in someone's hand, the smell of the room, the person who knows your name. My own work sits deliberately on both sides of that line. I automate what should be automated, so there is more room for the part that should stay human. The technology will keep accelerating. Bodies will not. That gap is where the value is going. --- ### MCP apps and how brands will show up in agentic commerce URL: https://joosthelfers.com/blog/mcp-apps-agentic-commerce Published: 2026-04-22 Keywords: Opinion, Strategy, AI The question most companies have been answering for the past twenty-five years is some version of "how do we get found". First by people searching, then by algorithms ranking, then by feeds recommending. Each shift rewarded a specific kind of work. A good website, then SEO, then social presence, then paid distribution. The companies that moved early on each transition tended to compound their advantage over time. A new layer is forming now. It is not fully consumer-facing yet, but it is close enough that you can see the shape of it. Agents are increasingly acting on behalf of people. Asking questions, comparing options, summarising research, and in a growing number of cases, taking real action like booking, buying, or scheduling. When that happens, the question shifts. It is no longer only "how do we get found". It is "how do we exist for an agent that is acting on behalf of a person". MCP apps are one way companies are starting to answer that. #### What an MCP app actually is MCP stands for Model Context Protocol. Its goal is simple. Give AI systems a standard way to interact with outside services, data, and tools. In practice, an MCP app exposes a defined set of capabilities. The actions a system can take on your behalf, the resources it can read, the information it can return. The protocol itself is deliberately boring, which is what you want from infrastructure. What matters is what it unlocks. An agent can ask your systems a question, take an action, or pull the information it needs to answer something well, without anyone having to build a custom integration for every new model or client. If your company already has an API, most of the hard technical work is done. An MCP app is the thoughtful, agent-readable wrapper on top of it. #### This is a brand problem, not a tech problem It is tempting to frame MCP apps as an engineering concern. Something the platform team will get to after the next sprint. That framing misses the shift that is actually happening. When an agent plans a trip, compares prices, or orders groceries, the brands that show up in the answer are the ones that speak the protocol. Visibility moves from the search results page into the agent's reasoning step. That is a different surface with different rules. If two hotels have the same inventory and only one exposes structured availability through MCP, the agent will reach for the one it can actually use. Over time that gap compounds. Agents learn which sources are reliable. The platforms hosting those agents prioritise the integrations that perform. The end user never sees which hotel was in the long list and which one was not. Only one of them ends up in the short list. That is what I mean by a brand representation problem. It is not a logo question. It is a presence question. If an agent cannot talk to you, you are not in the conversation. #### The parallel to the early web We have been here before. When the web was new, a lot of companies treated "get a website" as a small IT task. They shipped something basic, did not invest in it, and assumed customers would find them through other channels. A smaller number of companies treated the website as a core brand surface from day one. They hired the right people, wrote the right copy, and designed an experience that reflected how they wanted to be seen. The second group built a lead that was hard to close later. MCP apps feel like a similar inflection, with one important difference. The surface is smaller and less visible to end users today, which means the pressure is lower and the incentive to invest is easier to defer. The companies that do invest early will have a quieter but real advantage as the agent layer grows. #### What a good MCP app looks like A thoughtful MCP app is not a raw API dump. It is a designed interface for an agent user. The good ones treat the agent the way you would treat any other customer. A few principles seem to hold up so far. - **Expose what serves the user.** Not every action or resource in your system needs to be callable from an agent. Curate the surface. - **Give tools clear, descriptive names and documentation.** An agent chooses what to call based on the description. Vague descriptions produce vague behaviour. - **Keep responses structured and predictable.** Numbers as numbers, enums as enums, dates in a standard format. The agent parses better, and the end user gets a better answer. - **Reflect your brand in the machine-readable surface.** The language you use in tool descriptions, confirmation messages, and error states is part of how your company will be quoted back to people. That voice is a brand decision. None of this is glamorous work. It is the same sort of careful, unshowy craft that separates a great website from a functional one. #### Where this leaves us MCP apps will not be urgent for most companies this quarter. The agent layer is still maturing, and most of the commerce it handles is experimental. That will change, and probably faster than the current pace suggests. The companies that do the thinking now, about what their machine-readable presence should say and feel like, will be ready when the pressure arrives. For most brands, the question is not really whether to have an MCP app. It is whether the version that eventually shows up in the agent is the one they would be proud to put their name on. --- ### The future of AI content is not better prompts. It's better systems. URL: https://joosthelfers.com/blog/prompt-enhancement-engine Published: 2026-04-07 Keywords: Technical, AI, Product A lot of AI-generated content looks like AI-generated content. Not because the models are weak, but because the process around them tends to be light. People type a sentence into an image generator, get something back, tweak a few words, and try again. There is rarely a creative framework behind any of it, and no visual logic that ties one output to the next. The result can look impressive in isolation and then fall apart the moment you place two pieces next to each other. I have been working with AI tools in creative production for a while, and the pattern I keep seeing is always the same. The bottleneck is not the model. It is the gap between having a clear creative vision and translating that vision into prompts that actually produce coherent, high-quality results across multiple tools and scenes. #### The quality problem There is a lot of AI content out there right now, and much of it sits somewhere in the middle. Not because the technology cannot do better, but because speed tends to win over craft. Generating twenty images in ten minutes is easy. Generating twenty images that feel like they belong to the same project is genuinely hard. This is the same problem that has always existed in creative production. Consistency requires a system, and in traditional work that system is called art direction. Someone defines the visual language, the lighting approach, the colour logic, and the texture palette. Every individual piece then gets produced within that framework. That is what makes a campaign feel like a campaign instead of a mood board dump. AI tools do not have this layer by default because they are stateless. Every generation starts from zero. If you want consistency, you have to carry it across every prompt yourself. That means writing detailed, structured prompts over and over, adjusted for each tool's syntax and strengths. It works, but it is slow, repetitive, and easy to get wrong. #### What I built I called it the [Prompt Enhancement Engine](https://promptenhancer.joosthelfers.com). It takes a creative brief and some reference images and, before writing any prompts, first generates an art direction layer. A structured interpretation of your brief that defines the lighting language, material logic, colour palette, and overall mood. The visual framework that an experienced art director would establish before any production begins. From that framework it then generates a full set of prompts for image generation, image editing, and video creation. All of them are derived from the same visual logic, so that when you change the brief everything updates consistently. The order of operations is the whole point. Most prompt tools go straight from "idea" to "prompt". This one goes from "idea" to "art direction" to "prompt". That middle step is where the quality lives. #### Why the human in the loop matters One thing that tends to get lost in the conversation around AI tooling is that the human does not just validate the output. The human validates the thinking. The art direction layer this tool generates is not a black box. You can read it, adjust it, disagree with it. Because you see the creative decisions before they get turned into prompts, you catch bad interpretations early, before any generation credits are spent on content that misses the mark. At this stage of AI development, that kind of human oversight makes a real difference in output quality. Models are good at pattern matching and getting better at creative interpretation, but they still benefit from a person who can say "the mood should be more restrained" or "this lighting approach does not fit the brand". That feedback loop, applied at the art direction level rather than at the pixel level, is where you get the biggest quality gains for the least effort. #### The stack Deliberately lean: - **Framework:** Next.js App Router, TypeScript, Tailwind CSS (all hand-rolled, no component libraries) - **LLM routing:** OpenRouter with Gemini 2.5 Flash for vision and reference analysis, DeepSeek v3.2 for briefs and prompt generation - **Validation:** Zod for all structured LLM output at runtime - **Deployment:** Vercel with SSE streaming - **State:** No database, no auth, no state management library. Held with the built-in React primitives. I wanted to see how minimal this could be while still being genuinely useful. #### Where this is going What I built is a tool with a human at the centre, but the architecture points toward something broader. Agentic art direction. The individual steps this tool performs (brief interpretation, reference analysis, visual framework generation, prompt writing) are all things that agents will handle increasingly well on their own. Add a few more capabilities like visual trend research, output evaluation, and iterative refinement, and you have a system that can run large parts of the creative production pipeline with minimal human input. I do not think this means creative people become irrelevant. If anything, it is the opposite. As AI handles more of the production mechanics, the value of human judgment moves upstream into brand strategy, creative direction, editorial taste, and knowing what "good" looks like for a specific context. These are the things that are hardest to automate and most valuable to get right. The near-term future will probably look something like this. Tools that take a brand guideline, analyse current visual trends in a market, generate a visual framework, produce a first round of content, evaluate it against the brief, and iterate. All within minutes. A human checks in at key decision points rather than steering every step. We are not there yet, but we are closer than most people think. The teams and individuals who produce the best AI content will not be the ones who write the best prompts by hand. They will be the ones who build the best systems around the creative process. This tool is my first step in that direction. --- [Try it out](https://promptenhancer.joosthelfers.com) if you want to see how it works. Feedback is welcome, especially if you are working on similar problems. --- ### Beyond the AI noise: a case for digital stoicism URL: https://joosthelfers.com/blog/digital-stoicism Published: 2026-02-16 Keywords: Opinion, Strategy If you follow AI developments, you know the discourse carries a lot of hype, drama, and buzzwords. Every update gets framed as the one that changes everything. Every model release is positioned as a turning point. There is a persistent undertone that our jobs are about to disappear. It is easy to get pulled into that loop. I have noticed how quickly the mood can swing between losing hope and feeling intense pressure to seize the opportunity and make the most of this moment. Recently, after stepping away from the timeline for a while, I have come to a simpler view. The constant drama is not the whole story. AI will almost certainly create meaningful shifts in work, society, and the economy, but those shifts are messier and slower than the headlines suggest. Most of the technology still needs real work around it to function properly and create actual value. It needs oversight, testing, and patient iteration to become useful. Not everyone has a personal AI assistant yet. Not everyone needs one. The advances are genuinely exciting, and they have serious implications, but they are not worth losing sleep over. There is a lot of talk about new economic models where agents handle payments, self-reinforcing AI systems, and companies that are run end to end by machines. Some of that will happen in some form. It is also fair to say that we are not there yet. Even in a more automated future, you will still walk to the local bakery and buy bread from a person you know by name. The world will not change overnight, and people will find ways to adapt. A stoic approach makes sense. Focus on what you can control. Accept what you cannot. Exploring new tools, testing coding agents, and playing with the latest video models is a fun part of the job. What matters at the end of the day is the quality of the work produced and the value it holds for the people using it. Generating a lot of low-effort content does not help anyone. Humans still value human connection. We are social creatures by design, and that is not going to change anytime soon. The short version. Go outside for a while. See your friends. Share a meal. The technology will still be here when you get back.