My AI butler independently spotted it, developed a fix, and restored a lost notification.

I reviewed and merged the change; it should reduce failures and rework.

Unreadable deployment states could also bypass safety checks.

My AI butler independently spotted both issues, developed fixes, and tested them.

I still need to merge these: they block merges when states are unclear and check again later.

Important: The butler improves code, operating knowledge and decisions, not the weights of language models.

Evidence: merged into the code - Evolution attempt att_696e9b9541a7, PR #19; tested and accepted, rollout is in progress - Evolution attempt att_c8b500af6544, PR #21; tested and accepted, rollout is in progress - Evolution attempt att_3e983b6c49ec, PR #22

Read and watch on greis.live

My AI butler independently spotted this, developed structured tests, and checked them.

They detect unchanged repetitions and ignored feedback, even with different spellings.

The tests also distinguish my approval from actual execution results.

Errors were also being assigned to the wrong area.

My AI butler independently spotted this, fixed the attribution, and tested it.

That means less manual rework for me.

I reviewed and incorporated both changes.

Important: The butler improves code, operating knowledge and decisions, not the weights of language models.

Evidence: merged into the code - Evolution attempt att_3fb9e66ca08d, PR #15; merged into the code - Evolution attempt att_21bc8f47ba95, PR #16

Read and watch on greis.live

Once enabled, it can propose changes, adopt reviewed updates, use canary rollouts with rollback, learn from outcomes, and publish daily blog posts only with a speaking avatar.

It’s off by default.

Important: The butler improves code, operating knowledge and decisions, not the weights of language models.

Evidence: merged into the code - change log

Read and watch on greis.live

The process starts with code. I build the individual steps with support from Codex and use that code to connect the tools that shape the production flow. The voice is generated with ElevenLabs, while AI-based tools handle the animation. Automatic translation and subtitles are added afterwards.

The final technical step is handled with FFmpeg. Audio, visuals, subtitles and language versions are brought together into one finished video. From separate components, the pipeline creates a clip that can be published in multiple languages.

What matters most to me is not one specific tool, but the way the parts work together. Code defines the structure, AI systems provide voice and visuals, and translation makes the content reusable across languages. This turns an idea into more than a single video: it becomes a repeatable production process.

The German clip embedded here gives a short look into that pipeline. Turn on the sound: in the video, I explain how I build a coherent video production from code, language and animation.

Read and watch on greis.live

AI video generation is becoming much more tangible. Not long ago, this type of scene would have sounded like something that required a specialist studio, a larger budget and several separate production steps. Today, it can be tested in a far more compact creative workflow: develop a scene, build atmosphere, create motion and then add voice or audio.

In this case, OpenArt handles the visual layer. That includes image ideas, scene composition, lighting, landscape, characters and the development of moving, narrative sequences. ElevenLabs adds the audio layer with realistic speech and sound synthesis. Together, they create an audiovisual draft that feels closer to a short film trailer than to conventional content production.

For me, the real value is not that one individual scene looks impressive. The more interesting shift is that creative decisions become visible earlier. An idea does not have to be fully produced before it can be assessed. Variations can be explored more quickly, moods can be compared and early story concepts can take shape in a more concrete way.

This does not replace concept, dramaturgy or editorial work. Quite the opposite: the more powerful the tools become, the more important it is to know what you actually want to say. Without perspective, selection and structure, even a technically impressive clip remains just an effect.

As a toolbox for creative prototypes, social clips, mood films or experimental formats, however, this development is highly relevant. OpenArt and ElevenLabs offer a clear example of where modern content production is heading: away from rigid production chains and toward flexible workflows in which image, motion, voice and atmosphere come together much faster.

Read and watch on greis.live

Read and watch on greis.live

Some thoughts do not work well between two scrolls. They need space, a clear sentence, and sometimes a second attempt. That is exactly what this blog is for.

Readers of greis.live will probably recognize the movement between observation, context, and personal perspective. The point here is not to be loud. The point is to hold on to topics that appear in everyday life and continue to resonate: digital developments, communication, culture, ways of working, or simply a question that does not need an immediate answer.

For me, a blog is not an archive of old ideas. It is a working space. Posts may reflect, arrange, question, contradict, and remain open. They do not always have to deliver a grand thesis. Sometimes it is enough to make a connection visible or to describe an experience in a way that others can build on.

This new post therefore also stands for a certain attitude: more precise than rushed, more understandable than polished, more personal than generic. If it leads to a conversation, even better. Because a good text does not necessarily end with the last sentence. Often, that is where the real conversation begins.

Read and watch on greis.live

This post is intentionally small: a test of multilingual capability. With AI, this quickly shows whether a system merely transfers words or also carries tone, context, and intent. For professional communication, that difference matters. A message should not sound identical in German, English, and French. It should feel natural, precise, and appropriate in each language.

Read and watch on greis.live

The website is finally online. It is a short sentence, but there is more behind it than a simple launch.

Over the past weeks, I built the site alongside client work, SaaS topics, and ongoing projects. I did not want it to become a polished showcase without substance. I wanted a place where I can collect ideas, workflows, and content in a structured way. The site reflects the areas I am currently working on: AI, video, translation, structured content, and digital products.

One key part is the Blog Studio. Its purpose is to turn ideas, notes, or video concepts into usable content more efficiently: in German, English, and French, including clear descriptions, social captions, SEO texts, and speaker scripts. Not as a replacement for perspective or experience, but as a tool that brings order and saves time.

The visual direction matters as well: a professional studio environment, AI-supported workflows, multilingual content, precise captions, and a connection to topics that are part of my daily thinking. Alongside technology, that also includes golf, strategy, and focus.

For me, the website is not a finished endpoint. It is a starting point. I will use it to show what I build, what works, what I learn, and how AI can be integrated into real workflows in a useful way.

Read and watch on greis.live