👨‍🚀 Vibe Workflows - What we learned from experimenting

In partnership with

Hi AInauts,

Welcome to the latest issue of your favorite newsletter!

Last week, we tested the new Sora 2 API. Honestly? We had no idea if it would work. The result: a functioning video studio and n8n workflow! Without writing a single line of code ourselves.

Sounds smooth? It wasn't. But what lessons did we learn from this chaos? We're sharing them with you today. Alright, here's what we've got for you:

• 🫶 Vibe Workflows: Our Sora 2 Experiment—and What We Learned

• 🤖 State of AI Report 2025: Less hype, more business, real robots

• 😁 AI fun: Thinking? Delegated.

Let's go!!

Build hands-on AI skills and an MBA-level network.

The AI for Business & Finance Certificate Program is designed specifically for non-technical roles, so no coding or advanced data skills are required.

Over 8 weeks, you’ll learn directly from Columbia Business School faculty and Wall Street Prep instructors who train teams at the world’s top firms, including:

• BlackRock

• Bain

• KKR

• Deloitte, and more

Earn a certificate from a top business school and gain the practical AI skills to move your career forward.

Enroll by Oct. 13 to get $200 off tuition + use code CERT300 for an additional $300 off.

The program starts November 10.

Enroll Today

🫶 Vibe Workflows: Our Sora 2 experiment—and what we learned from it

Last week was all about OpenAI, especially the new Sora 2 Video Generator. The new AI-slop social network even made it to #1 in the iPhone App Store and provided lots of entertaining deepfakes...

But at OpenAI Dev Day, what we had been waiting for finally arrived: access to the API interface. This allows anyone to build video generator apps or integrate video generation into their own workflows. We wanted to test it right away!

Spoiler: We did it. A working Sora 2 video studio and a complete n8n workflow. Without writing a single line of code ourselves!

And you know we're not die-hard techies. We're marketing people. In the past, we would have hired a developer, waited weeks, and paid thousands of dollars. Today, we build it ourselves. In hours (ok, sometimes days 😉). Welcome to the era of Vibe Workflows!

The shift from coding and prompting to context engineering

Most fail because they rush into it too quickly. They open a chatbot and say, "Build me an app with Sora 2." The chatbot builds something half-baked. It doesn't work. Frustration is inevitable.

What many people have not yet internalized: Prompts are no longer the most important thing. The context must be right!

Would you tell a developer, "Build me something with videos"? Of course not. You would write a clear briefing, make sketches, and specify details—so that you get something useful in return and your capital is used optimally. That's exactly what you need for AI!

How we built our app and workflow

What we have learned in recent months: The best results do not come from the best prompts alone. Solid preparation is required to provide the context. You have to show the AI what you mean, not just tell it what you want.

We wanted two things: a web app for video generation and a n8n automation workflow, both of which connect to the new OpenAI Sora 2 API.

This is how we proceeded with the app:

• We wanted to vibe code the app—we've done that quite a few times now.

• First, we asked Claude to create a developer guide from the OpenAI links containing details about the API specification. In other words, a complete documentation with all API endpoints, parameters, code examples, and best practices (=context).

• Then we used our master prompt to generate a Product Requirements Document (PRD) from the rudimentary app idea. You can also do this with this Nauti CustomGPT.

• We then fed this PRD plus the Developer Guide into the Vibe coding app of our choice—in this case, Lovable. Generate. Check. Iterate.

• After a few rounds: working MVP (minimum viable product). After a few more rounds—final app! That was easier than expected.

• You can use the app here, just enter your OpenAI API key (Note: your organization must be verified).

P.S. And if you need help with prompts, GitHub repo with prompt ideas - here you can also find the Nauti Sora 2 prompt generator as CustomGPT *yes, you could also integrate it directly into the app.

The process was similar for the n8n workflow (and that really surprised us!):

• We wanted to prompt the n8n workflow completely in Claude, instead of manually "building" it in n8n.

• To this end, we provided Claude with the Developer Guide containing details about the OpenAI API (=context).

• Then attach the n8n MCP (=context for n8n) ...

• ...and prompt us with what kind of scenario we imagine. Go!

• Then came the moment of truth: the first import into n8n. And lo and behold: the scenario was displayed immediately! *ChatGPT and Gemini failed here.

• Did it work right away? No. But it produced errors that we could then take back to Claude.

Was the process smooth? No, not really. It was tedious, especially compared to the Lovable app. Lots of testing. Lots of minor fixes. Lots of context resets with Claude ("You have to start a new chat..."). But in the end, it worked. And we didn't have to do a single thing in n8n itself, 0.0%!

And that's the point: three months ago, that would have been impossible for us. In three months, it will be a piece of cake.

The tools are improving. We are learning how to use them properly. The process is becoming increasingly efficient.

You no longer need to understand in detail how to marry five n8n nodes together or how webhooks work. But you do need to be able to describe what you want to achieve. Describe it clearly. With context. With structure.

...then you can easily have videos like this made. 😉.

The next few months will be wild—what that means for you

We are at the beginning of a massive transformation. What began with "vibe coding" is now evolving into "vibe automation"—complete workflows from a text prompt, or even with voice.

The skills that will count tomorrow are no longer those of yesterday. Technical knowledge is becoming less important. System knowledge and networked thinking are becoming more important.

In other words, you no longer need to know how to connect an API technically or what exactly each n8n node can do. But you do need to understand which processes should be automated, in what order, and why. You need to see where the levers are.

AI gives you the tools. You need to know what you want to build with them. Think strategically. Build prototypes. Test quickly. Fail quickly. Learn quickly.

🤖 State of AI Report 2025: Less hype, more business, real robots

We feel that AI has come of age. Less bullshit bingo, more real thinking, and new highlights every week that you first have to digest.

The brand-new State of AI Report 2025 provides the latest facts on this topic (and the Time 100 Innovations show what will shape the future).

While last year was still all about consolidation, 2025 was all about reasoning, robotics, and the global race for thinking machines.

Here is a video overview and the entire presentation as Google Slides.

Here is a brief summary of the most important findings:

• Reasoning makes the difference: Instead of telling ChatGPT to "Write me a marketing plan," you can get the model to really think: "Analyze our target audience, think through 5 strategies, evaluate the pros and cons, and recommend the best one with reasoning. " Reasoning models save you hours of work!

• China's open source models are overtaking Meta: DeepSeek, Qwen, and others are open source and not only beat Meta's Llama, but in some cases even commercial models. Specifically, you can host them locally, your data stays with you, and they cost nothing. Perfect for business cases where you don't want to send data to the cloud.

• Robotics is becoming relevant to business: Department store operators report that robots are already reducing error rates by 40%. They think first, then act. Just like a good employee. And it's no longer a dream of the future: Google's Gemini Robotics 1.5 is commercially available, robots can be ordered online, and Elon and Co. want to start mass production.

• AI is actively involved in research: Stanford's Virtual Lab tested more hypotheses in three months than a team could in a year! In the pharmaceutical industry, this means going from idea to validated protein design in weeks instead of years. Even if you're not in biotech, this pace is coming to all areas of R&D. We agree with Brian Johnson's motto: Don't die! Those who survive the next few years have a good chance of benefiting from the new AI discoveries. Longevity and all that. …

For us, it's clear: the difference between an AI user and an AI power user isn't talent, but 3–4 hours of experimentation. This investment pays off every day! The fastest way to go from "interesting" to "holy shit, this changes everything" takes one afternoon. So: start small, think big—and have fun doing it. At least that's how we do it.

😁 AI Fun: Thinking? Delegated.

Well, that's it for today. We wish you a great week—AI bubble or robots or whatever, it'll be fine. Somehow. Probably. Right?

You made it to the end—thanks for reading! We’ll be back soon with even more updates.

Reto & Fabian from the AInauts

P.S.: Follow us on social media - it motivates us to keep going 😁!
X, LinkedIn, Facebook, Insta, YouTube, TikTok

Your feedback means the world to us. We read every comment and message—just hit reply and tell us what you think!