How to Safely Adopt GenAI Without Leaking Customer Data
- Jörn Menninger
- Aug 13, 2025
- 32 min read
Updated: Aug 20, 2025
🚀 Management Summary
Generative AI is powerful—but for startups, it's also full of privacy pitfalls and model myths. In this guide, Dennis Traub of AWS breaks down a safe approach to GenAI adoption. Founders, product leads, and operators will learn how to evaluate risk, select secure models, avoid legal missteps, and implement use cases that actually matter. This post anchors a full SEO content cluster on startup AI safety and innovation.
📚 Table of Contents
Why Safe GenAI Adoption Matters for Startups
What Founders Get Wrong About AI
The "Artificial Intern" Framework
How to Choose Your First AI Use Case
GDPR, Model Risk & Mailbox Mayhem
Innovation vs Optimization: A Founder Mindset Shift
AWS Bedrock & Privacy-Compliant Model Hosting
Key Takeaways for Startup Builders
FAQs: Safe AI Adoption for Startups
🚀 Meet Our Sponsor
AWS Startups is a proud sponsor of this week’s episode of Startuprad.io. Visit startups.aws to find out how AWS can help you prove what’s possible in your industry.
The AWS Startups team comprises former founders and CTOs, venture capitalists, angel investors, and mentors ready to help you prove what’s possible.
Since 2013, AWS has supported over 280,000 startups across the globe and provided $7Billion in credits through the AWS Activate program.
Big ideas feel at home on AWS, and with access to cutting-edge technologies like generative AI, you can quickly turn those ideas into marketable products.
Want your own AI-powered assistant? Try Amazon Q.
Want to build your own AI products? Privately customize leading foundation models on Amazon Bedrock.
Want to reduce the cost of AI workloads? AWS Trainium is the silicon you’re looking for.
Whatever your ambitions, you’ve already had the idea, now prove it’s possible on AWS.
Visit aws.amazon.com/startups to get started.
1. Why Safe GenAI Adoption Matters for Startups?
Startups are moving fast to implement GenAI—but many are doing so without guardrails. The result? Privacy breaches, hallucinating chatbots, or worse: lost customer trust. The foundation of safe GenAI adoption begins with understanding the actual capabilities and limitations of large language models (LLMs).
2. What Founders Get Wrong About AI
Many founders believe they need the “best” model or the most advanced tech stack to win. Others treat GenAI like a co-founder. Both are flawed assumptions.
Common misconceptions include:
That LLMs understand facts (they don’t)
That more power = better outcomes (not necessarily)
That privacy concerns only apply to enterprises (false—GDPR applies to all)
The first step is realizing that GenAI is not a magical solution. It’s a probabilistic, pattern-recognizing system. Your job is to direct it wisely.
3. The "Artificial Intern" Framework
Dennis Traub introduces a powerful metaphor: AI as your "Artificial Intern".
Just like an intern fresh out of university:
AI knows a little about everything
AI lacks real-world judgment
AI can do research, write drafts, classify data
But you have to check everything before it ships
This mindset shift allows teams to assign meaningful tasks to AI—without over-trusting its output.
4. How to Choose Your First AI Use Case
Your first GenAI project shouldn't be ambitious or glamorous. Instead, follow Dennis's playbook:
Ask your team:
What tasks always get postponed?
What recurring job is annoying but necessary?
What low-risk function could AI draft, classify, or summarize?
Then ask:
Can we safely hand this to an intern?
Would we double-check the output before it goes public?
If yes, it's a great GenAI candidate.
5. GDPR, Model Risk & Mailbox Mayhem
Giving AI access to your inbox is like handing your intern your house keys. It might accidentally email the wrong client or expose confidential info.
Key risks include:
LLMs storing sensitive data
Non-compliant vendors sending data across borders
Accidental exposure of customer or employee data
Dennis warns that even accidental exposure can trigger legal penalties under GDPR—and you, not the model, are responsible.
Checklist for safety:
Read your model provider’s T&Cs
Choose GDPR-compliant tools (AWS Bedrock, etc.)
Avoid giving write-access to mission-critical tools
6. Innovation vs Optimization: A Founder Mindset Shift
Startups are often tempted to use GenAI to optimize what already works. That’s not where innovation lives.
Dennis urges founders to solve problems that were previously unsolvable due to cost, time, or scale. This is where GenAI shines.
"Don’t just automate old problems. Solve the ones no one thought were solvable."
Examples:
Summarizing 100s of emails into strategic reports
Auto-sorting customer complaints across 12 markets
Real-time classification of legal contracts
7. AWS Bedrock & Privacy-Compliant Model Hosting
Public LLM APIs often retain data, use it for training, or fail to meet EU data residency requirements.
AWS Bedrock offers:
GDPR-compliant infrastructure (models run in Europe)
No data training reuse
Air-gapped model environments
Enterprise-grade encryption
This makes it a viable backend for founders who want safe AI without building a private stack from scratch.
8. Key Takeaways for Startup Builders
Treat AI like an intern: Train it, test it, but never trust it blindly
Start small: Pick use cases where failure is low-risk
Avoid mailbox-level access: Protect your org's data perimeter
Use privacy-first models: AWS Bedrock is a solid start
Don’t optimize—innovate: Use GenAI where humans couldn't before
🧵 Further Reading
🎥 The Video Podcast
🎧 The Audio Podcast
Subscribe here: https://linktr.ee/startupradio
🚪 Connect with Us
Partner with us: partnerships@startuprad.io
Subscribe: https://linktr.ee/startupradio
Feedback: https://forms.gle/SrcGUpycu26fvMFE9
Follow Joe on LinkedIn: Jörn Menninger
📝 About the Author
Jörn “Joe” Menninger is the founder and host of Startuprad.io — one of Europe’s top startup podcasts. Joe's work is featured in Forbes, Tech.eu, and more. He brings 15+ years of expertise in consulting, strategy, and startup scouting.
Learn More
If you are looking to understand the rise of AI and deep tech startups in Europe, including how emerging technologies like machine learning, quantum computing, and robotics are transforming industries, you should not miss Europe’s Ultimate Guide to AI & Deep Tech Startups. This in-depth resource provides founders, investors, and ecosystem leaders with a comprehensive overview of European AI innovation, venture capital trends, and deep tech opportunities, making it a must-read for anyone aiming to stay ahead in the fast-growing European startup landscape.
✅ FAQs: Safe AI Adoption for Startups
What is safe GenAI adoption?
It means implementing GenAI tools in a way that doesn’t risk your customer’s data, violate privacy laws, or erode trust.
Why is GenAI called an "artificial intern"?
Because it’s smart, eager, and good with tasks—but still needs supervision, like a junior team member.
What kind of tasks should I assign GenAI first?
Start with classification, summarization, or templating—tasks that are low-risk but time-intensive
Do I need the most advanced model to start?
No. Most frontier models are sufficient. Start simple and upgrade based on results.
How do I know if a GenAI vendor is GDPR compliant?
Review their documentation, ensure they don’t store your data, and check for EU data residency options.
What is AWS Bedrock and why use it?
It’s a service that lets you run GenAI tools on secure, EU-based infrastructure, with no data sharing.
Can GenAI fully automate customer email replies?
It can draft replies, but human review is still essential—especially for sensitive communication.
What risks are involved with GenAI integration?
Misuse, hallucinations, data exposure, legal issues, and loss of customer trust.
How can startups innovate with GenAI?
By solving small but persistent problems that were previously too expensive or time-consuming.
Should I build a chatbot as my first AI product?
Not necessarily. Most chatbots are UX-driven but lack real business value. Focus on backend automations first.
Give us Feedback!
Let us know who you are and what you do. Give us feedback on what we do and what we could do better. Happy to hear from each and every one of you guys out there!
The Host & Guest
The host in this interview is Jörn “Joe” Menninger, startup scout, founder, and host of Startuprad.io. And guest is Dennis Traub, AI Engineering Specialist at AWS.
Reach out to them:
📅 Automated Transcript
Dennis Traub | AI Engineering Specialist at AWS [00:00:00]:
Do you know what AI actually stands for? It's Artificial Intern. And that's how you have to work with. It's an intern who strangely knows a lot of stuff. So they're straight out of university and they have studied pretty much everything, but they, they don't have any, any, any real world experience. They don't have the intuition that, that you have as a business person or a developer or whatever you do. They can do a lot of the repetitive work, a lot of the menial work of analyzing stuff, analyzing text, analyzing information, or writing templates, classifying something and kicking off backend algorithms. This is what they're really good at. But in most cases, if they make any decisions or produce any facts, you should always double check and make sure.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:01:01]:
Welcome to Startuprad IO, your podcast and YouTube blog covering the German startup scene. With news, interviews and live events, AWS is proud to sponsor this week's episode of startup raid IO. The AWS team compromises former founders, CTOs, venture capitalists, angel investors and mentors ready to help you prove what's possible. Since 2013, AWS has supported over 280,000 startups across the globe and provided US$7 billion in credits through the AWS Activate program. Big Ideas Feel at home at AWS and with access to cutting edge technologies like generative AI, you can quickly turn those ideas into marketable products. Want your own AI powered assistant? Try Amazon Q. Want your own AI products? Privately customize leading foundation models on Amazon Bedrock. Want to reduce the cost of AI workloads? AWS Trainium is the silicon you're looking for.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:02:14]:
Whatever your ambitions, you've already had the idea. Now prove it's possible on AWS. Visit aws.Amazon.com startups to get started. Dennis Straub is a developer advocate at aws, where he guides companies through the safe adoption of emerging tech. With a deep background in cloud security, developer enablement and generative AI integration, Dennis helps teams test, iterate and learn without putting the data or business at risk. Today we unpack the AWS playbook for starting with gen AI. Even if you're just getting curious. Denis welcome to StartupRate IO and for every podcast aficionado, we may add that you have been the original voice of the German AWS podcast.
Dennis Traub | AI Engineering Specialist at AWS [00:03:05]:
Oh, thank you Joe. Thanks everyone for listening. I can't. I don't. Is it. Is it even still true with the AWS podcast? Podcast? It has been. That was during. That was during COVID I started that during COVID I put out I think 50 episodes or so until Traveling started up again and unfortunately I wasn't able to continue, but a few of my friends and colleagues here in Germany actually picked it up and are still continuing it.
Dennis Traub | AI Engineering Specialist at AWS [00:03:30]:
Anyway, thanks for having me on. On the show and right in the, in the introduction, you mentioned something I think that's really, really dear to my own heart and probably to most of your listeners. What's the ROI in AI? I think that's a question that many people have, including myself, quite often. So I'm happy to talk about this today.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:03:56]:
When people talk about AI, what comes to mind is ChatGPT doing everything with it, but it's a chat window. Plus what, what has been in the news on and off is Elon Musk's croc for either very great or very bad answers. So everybody who's only heard about that, how could you get started safely with Gen AI?
Dennis Traub | AI Engineering Specialist at AWS [00:04:26]:
Well, I think, most importantly, first of all, it's important to understand what generative AI actually is, how it works. Not in detail. You don't have to. I don't have a PhD in math. I don't really understand math. But you don't, you don't need to have that. But it's important to have a foundational understanding of how these models work and specifically what they are not. They are not people.
Dennis Traub | AI Engineering Specialist at AWS [00:04:53]:
They are not human beings, even though they talk like human beings. And Andre Karpathy, one of the, one of the people who do a lot of foundational work in AI, he actually said, LLMs are like stochastic simulations of people. So they behave like people in a certain way in terms of putting out text, saying something, but they behave like this friend that some of you may have had in the past. I certainly did. That person who knew every, everything. And when you, when you ask them anything, they, they would have an answer. And they were so convincing with what they said. But once you started questioning, you might have realized, well, maybe, maybe it's not really what they're saying, or maybe they are not as sure as they think they are.
Dennis Traub | AI Engineering Specialist at AWS [00:05:48]:
So I try to, I try to compare AI models, generative AI language models, compare them with this kind of friend who would like to know a lot and probably knows a lot as well, but sometimes confuses things and isn't really aware or doesn't want to, doesn't want to show any weakness and tries to bring across whatever they come up with as convincing as possible. And that's what's really important. AI does not know anything. AI has been trained on, on the entire Internet, basically on a lot of text Material and what they do internally is just whenever you type something, whenever you send something into the language model, it looks at what you wrote and then it compares it to what it has read in the past. And then it comes up with, well, when I had this sentence, most of the time the next word was this. So it learns during training, it learns to relate concepts with each other without actually understanding the concept. Take a cat and the word flurry. And an LLM sees these words together very often when being trained on the Internet.
Dennis Traub | AI Engineering Specialist at AWS [00:07:08]:
And then it knows with I'm doing air quotes here, it knows that cats and flurry somehow relate to each other and may create text that puts these two words together. This is very simplified, but that's effectively how it works. It does not understand anything. It is extremely well trained in terms of pattern recognition. And it repeats patterns that it originally saw. And many of these patterns have been scientific papers, lexical articles and all kinds of information where people convincingly describe what they are talking about because they are convinced. Because most of the time it's actually true. And the model just adapted this way of communicating.
Dennis Traub | AI Engineering Specialist at AWS [00:08:02]:
That's why it sounds convinced. It is hard to find any text on the Internet where somebody says, I don't know. This is why most models also do not respond with I don't know. They just come up with stuff. And that's what's really important. AI. AI models know a lot, but have often have a hard time to really put the things together in a way that's that it's really, that it's really factual. And that is something that you should be basically aware of.
Dennis Traub | AI Engineering Specialist at AWS [00:08:40]:
If you know that, then you can deal with it in a certain way, then you know, I shouldn't rely on it. It's not that they're not good enough yet. The way these models work, they will never have actual understanding of what they talk about. They will always be pattern recognition recognition algorithms. And if you understand that, you can work with them. Like I like to think about them as another thing is like, do you know what AI actually stands for? It's artificial intern. And that's how you have to work with. It's an intern who strangely knows a lot of stuff.
Dennis Traub | AI Engineering Specialist at AWS [00:09:26]:
So they're straight out of university and they have studied pretty much everything, but they don't have any real world experience. They don't have the intuition that you have as a business person or a developer or whatever you do.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:09:43]:
I have the exact example for this. For example, I'm using a lot of chatbots for many, many different functions and I've come to use them for evaluating pitches, guest pitches, for the very simple reason. I get up to 30 a week during summer, and during winter it can be 60 to 100. That only makes sense for me to reply if it's template, if it's not AI generated.
Dennis Traub | AI Engineering Specialist at AWS [00:10:14]:
And so you mean people, people approaching you because they want to be on your podcast.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:10:19]:
Okay, exactly, exactly. And so basically, at first I was copying simply in the email and the AI of choice gave me back a potential reply, but then I told it, okay, I want you to first evaluate what this actually is, how likely is it that it's written by an AI, what percentage is by human, and then give me a pretty fair assessment of X, Y and Z and this and this and that. And then only when I know that there is less than 50% AI involved, I look into the email, tell the AI what to do, what kind of reply I want to send, and then it goes out.
Dennis Traub | AI Engineering Specialist at AWS [00:11:07]:
And that's the thing. One of the things that these language models are really good at is classifying text, because they have seen so much text during their training that it's very easy for them to classify text if you help them understand what the premises are, what the conditions are, what the requirements are that you're looking at so they can classify. What they're really bad with is coming up with facts because they do not understand the concept of facts. They just, they're internally, they're just numbers. And texts can be represented as numbers, which is why summarization, classification and similar tasks are very easy, but there is no actual understanding. And the use case you just described is, I would be interested in understanding how many, how many false positives you have in terms of how many supposedly AI generated pitches you unfortunately sort out. Because maybe the model or maybe the person actually writes like an AI and it's, it's getting, it's getting harder to actually distinguish between well trained AI language models and actual human beings. There are, there are some indicators right now they may be out of date by tomorrow because things evolve, and especially people who build systems based on AI, AI to actually obfuscate the fact, to hide the fact that they're using AI, they're working on this as well.
Dennis Traub | AI Engineering Specialist at AWS [00:12:45]:
But let's get back to the intern, the artificial intern, like every intern, especially if they know a lot, they're fresh from university, they are really excited about the job, and they are really excited about learning about what they can do about your industry, about your use case, your products, your customers. At the same time, they have no Intuition and they don't have any real world experience that they can reflect on whenever they approach a new task or have a new challenge to reflect on and say, well, I saw something similar sometime in the past and it worked this way. And they can start abstracting and mapping and matching and coming up with solutions. They just know their textbooks, they know what they studied, they don't have the intuition. So it's your job. If you work with, with an AI, it's your job. You can give them a lot of tasks, but you always have to double check. They can do a lot of the repetitive work, a lot of the menial work of analyzing stuff, analyzing text, analyzing information or writing templates, or kicking off, kicking off classifying something and kicking off backend algorithms.
Dennis Traub | AI Engineering Specialist at AWS [00:13:58]:
This is what they're really good at. But in most cases you should really have a look at what if, if they make any decisions or produce any facts, you should always double check and make sure.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:14:15]:
What I found is they're really, really good in for example, you give them like 10 bullet points and a lot of keywords and say, okay, write me a text from that. They're excellent. They, they spare. They saving me a hell lot of time by doing that. I see an email, I say, okay, I want this, I want this, I want that. Because when you get tired and English is not your native language at 10pm it gets really, really difficult to formulate a straight, easy to read email. And that's when it comes in quite handy. But I wouldn't necessarily hand over my mailbox to Gemini or ChatGPT or Claude.
Dennis Traub | AI Engineering Specialist at AWS [00:14:56]:
And that's the point.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:14:58]:
I just want to know. We cannot hand over the mailbox. So how do we get to really start safely as a company with AI?
Dennis Traub | AI Engineering Specialist at AWS [00:15:10]:
Most importantly, don't give any AI system the keys to the kingdom. And your mailbox, especially if you're a founder, most likely is your key to the Kingdom. There are many, there are many agentic orchestration systems out there that allow to just connect to your Gmail account or to Outlook, whatever, or your calendar to connect to your data sources. It's useful as long as you don't let it act on this information or at least not let it act without you double checking before it actually acts on this information because in fact it would be able to delete all your emails, including important information. It may be able to actually send an email to somebody and it may not be the mail that you want this person to get. So there's certain risks whenever you give an AI or anyone like the Intern. Would you give your intern access to your, to your mail account?
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:16:26]:
I just had in mind, as when he said, for example, my wife has a pretty common first name, and instead of sending her an invitation to date, I send it to a potential client. The AI could do that because it's the same first name.
Dennis Traub | AI Engineering Specialist at AWS [00:16:42]:
Right, right, right, the AI could do that. But on the other hand, there's another risk, and that's actually data, data privacy and security. Because if you give a language model access to your email account, it has access to all private information that's inside this email account. This includes private information about you, but it may also include pii, personally identifiable information of customers of other people. It may even include specific sensitive information like health information or, or like, like relationships between a lawyer and client. And there are certain pieces of information that, that, at least in Germany, there's even more protection around it. And you can actually go to jail if you, if you expose your client's health information or your customer's health information, you can go to jail for that. And it's you, it's not your intent, intern who goes, it's you who goes to jail for it.
Dennis Traub | AI Engineering Specialist at AWS [00:17:52]:
So what's, what's really important to acknowledge is the fact that in your emails, there's a lot of personal information, whether it's sensitive or not, it is personal information. And as soon as you send it to a model, you need to be aware of how the provider of this model interacts with that data, whether they store it anywhere, whether they send it somewhere else, whether they keep it secure, whether they just throw it away and not store it at all. This information is really important. So as soon as you give a language model access or an agent access to your mailbox, you give the model provider access to personal information. You give your model provider technically access to all the private information that your customers, that your family, that you yourself have entrusted your mailbox with. And I mean, even if your customers wouldn't care, it would be a GDPR headache because as soon as you sent that to OpenAI or any other public provider that provides like a public interface, maybe even without any price tag, you send that information somewhere else. And you as a company have no information about what happened to this data. You lose control over this data.
Dennis Traub | AI Engineering Specialist at AWS [00:19:32]:
And, and by losing control over the data, you effectively violate GDPR and maybe other laws too. So it's really important to understand that if you build an AI system that has access to your MA box or any other proprietary or private information, you need to make sure that you understand the terms and conditions of, of the model provider that you're working with. You need to understand how, how, what do they do with your data? Do they, do they use it for model training? Do they store it somewhere to make it available to authorities when they ask for it, for instance, or maybe even outside the European Union? This is a very important piece of information, especially if you work with publicly available APIs and even more so if you work with APIs or providers that don't charge you. I mean, it should be a fairly well known fact nowadays that if you don't pay with money, you usually pay with data, especially with services on the Internet. And that's most likely what happens with many providers. I'm not going to name any names. It is up to you to have a look at the actual conditions and there are ways to work around that. Most major model providers provide ways to use their models that are GDPR compliant or that allow you to use these models.
Dennis Traub | AI Engineering Specialist at AWS [00:21:13]:
GDPR compliant? The models themselves are never. Or a tool that you use is never GDPR compliant by itself. It's always about the way that you use it. But most commercial model providers actually have options that allow you to build GDPR compliance systems with their models. But it's usually not their chat interface.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:21:37]:
That is exactly because what I was going for, because why wouldn't you start with a chatbot? And how, how would you look in a company, like on a meta level for the first real project they can use, they can do with AI. And I have to admit that made me a little bit nervous because currently I have somebody coding an AI based chatbot from a website here.
Dennis Traub | AI Engineering Specialist at AWS [00:22:06]:
So there's two questions. The first question why I wouldn't use a chat, why I wouldn't build a chatbot. And the second question is what I would look at when I would go into a company or I would think about a use case that would actually make sense. And that's an interesting question. First of all, I would not necessarily say I wouldn't build a chatbot. A chatbot can be a good use case, but most of the time it isn't. The thing is, let me step back just one a little bit. We, as in everybody who's trying to use AI or trying to figure out how to use AI in a useful way, are making, we're falling in a certain trap and it's completely normal to do that.
Dennis Traub | AI Engineering Specialist at AWS [00:22:56]:
We are trying to solve problems that we have. Most of them we have already solved or the problems are inside of what we can imagine fairly easily. And that reminds Me of back in the day, back in the 90s when the world Wide Web became a thing. I don't know if you had been around, I mean, Joe, you probably have been around. I have been around. I don't know about the listeners, but if you have been around. Back in the day, when we started building websites or web pages or home pages as we called them back then, back then we were mostly trying to replicate what we already knew. So we had yellow pages on the Internet where you could find web pages like yellow pages with classification systems that were based on traditional libraries.
Dennis Traub | AI Engineering Specialist at AWS [00:23:55]:
Everybody had a homepage which was more or less a business card. And we tried to replicate print material onto the screen, which was really hard because most MySpace, even before that, even before that, most screens only had 800 by 600 pixels. Early HTML didn't really allow you to position stuff. It's still hard nowadays, but back then it just didn't work. The lineup lines, so the connections to the Internet were so slow that you couldn't really use images. And it was really hard. So everybody said, every company said, well, we know we need to be on the Internet now, just like they say nowadays, we need to use AI, but it doesn't really work. And where's the ROI in this? Where's the ROI in putting my brochure, my print brochure onto the screen? And it's terribly slow for, for customers to load and display, but.
Dennis Traub | AI Engineering Specialist at AWS [00:24:54]:
Or what sense does it make?
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:24:58]:
Do you know what came to mind when you talked about terribly slow? The sound of a dial up connection.
Dennis Traub | AI Engineering Specialist at AWS [00:25:04]:
Right, exactly, exactly. And that's the thing. We tried to solve traditional problems with this new tool with this new technology. We tried to use traditional means and just map them onto the screen. And that didn't work because the screen is not made for printed stuff. The screen is not a thick yellow pages, a tome of addresses, of phone numbers for businesses. That's not what it is. And over time, more and more people, and took a few years, more and more people started realizing there's a completely new way of thinking about things.
Dennis Traub | AI Engineering Specialist at AWS [00:25:47]:
And that's where Google started. And Google started replacing traditional yellow pages. And that's when Amazon started, Amazon started replacing traditional brochures on the Internet where you could read what you could buy and then pick up the phone or send an email to the retailer to tell them, well, I want this as a mail order. That's when ebay came around, when Wikipedia or Wiki as a principle started to come around, when things like Facebook and Twitter came around, when we started to embrace the new medium and actually use it for things that we couldn't even imagine before. Traditional classification systems, like in libraries, they are important in libraries because they have to deal with shelf space. They have to put the book somewhere and they cannot put the book everywhere. But with the Internet, with hyper hyper data, with hypermedia, you can put the book literally everywhere. You don't need classification systems, or you can have adaptive classification systems, you can have dynamic classification systems.
Dennis Traub | AI Engineering Specialist at AWS [00:27:00]:
You can even create something that looks at how many people actually read your book and cited from it, which effectively is Google. How many people actually visit your homepage, your website, your application, and actually link to it from their page. That is what Google looks at. And it's the same situation all over again, in my opinion. With AI, we're still trying to solve old problems with the new tool, and we haven't really figured out what's the exciting new thing that we can build with this tool that was prohibitively costly in terms of money or in terms of time, so that we didn't even think about doing it. Imagine back in the day before we had the Internet and mobile phones. Imagine back in the day. You live in Germany.
Dennis Traub | AI Engineering Specialist at AWS [00:27:58]:
I do. When I called my relatives in the US or when my family called relatives in the US that happened once a year on Christmas. And every family members had about five seconds to talk to them because it was an intercontinental call, which was so expensive. So we never talked to our relatives except on Christmas when I went on vacation and I sent a postcard back home. Most of the time the postcard arrived two weeks after I arrived back home. And right now, with mobile devices, with the Internet and everything, we talk to people all over the planet all the time.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:28:42]:
Yes. I vividly remember what a revelation it was when I was studying in the US or working in China, when I could use Skype to call people for, for local, for local rates. So how could a company really identify what should be the first project? Because I wouldn't necessarily recommend to have like this really big hairy goal for the first AI project, but rather something small, something that really makes sense, takes a lot of maybe repetitive work out of the job of the employees.
Dennis Traub | AI Engineering Specialist at AWS [00:29:23]:
Right? And that's the most important question. What use case? What workflow? What item on your to do list gets never done? What are the painful things in your business that nobody ever took care about because it would take too much time, or because it would take too many people to work on it, or because it would be just too costly to do it? What are the things that if you had a magic wand and you could make them go away. So the daily tasks, the menial things, the things that bother you all the time, but they need to be done or they should be done, but I don't get around to doing them because I have so much to do. What is that thing that you would like to be to get done and it never gets done because there's no time for it. Make a list of these things, write down your most painful things that you have to deal with every day because they don't get done and they don't get done. And then have a look at it at them and think about, is this something I could hand over to an AI safely?
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:30:52]:
Hand over to an AI, right, safely.
Dennis Traub | AI Engineering Specialist at AWS [00:30:54]:
Maybe not in its completeness, maybe just a small part of it. And if you want to build a startup for a certain industry, talk to the people in this industry, talk to them, ask them this, this question. What is the one thing that has been bothering you for the last 30 years since you started in this industry? What is the one thing that's bothering you but nobody ever took care of it? And then think about whether this could be something that you could hand over either in part or maybe even completely to an AI and well, of course, don't start with the big hairy goal, don't start with the big thing. Try to find a small painful thing, create a solution for it and then go back to the customer, go back to the market, go back to the person you talk to in the industry or go back to yourself if it's for yourself and see does it actually.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:32:02]:
Do you know, Dennis, what my consultant mind was making of what you say? Basically you put a lot of your employees into brainstorming session, they come up with 20 problems, you cut it down to 10 problems that are really efficient if you could automate or partially automate them. And then you start with the easiest.
Dennis Traub | AI Engineering Specialist at AWS [00:32:24]:
Yes. And it's not only about they would be most efficient, but we could finally address them through automation. It wasn't possible before. That's the thing. If we try to address problems that we already automate and we could make them more efficient and that's not innovation, that's optimization. That's a good thing. I'm not saying we shouldn't optimize, we should optimize, but that's not innovation, that's not the breakthrough, that's not the next Google, that's not the next unicorn startup. The next unicorn startup will solve a problem that everybody has, but nobody even knew that they had it or nobody even thought of solving it because actually solving them, it wasn't even possible before.
Dennis Traub | AI Engineering Specialist at AWS [00:33:12]:
And this can be a small thing. This can be a tiny, small thing. It doesn't need to be a big thing. It can really be a small, tiny thing. And if you have something like this, you effectively have a money printing machine because everybody's going to tell, whoa, I didn't know that's possible. Right. I don't have the solution for you. So I cannot tell you it's this or that thing.
Dennis Traub | AI Engineering Specialist at AWS [00:33:37]:
That is something that you need to look into with your specific expertise, with your intuition, with your background, with your creativity. Creativity. But what we've been doing most of the time with AI in the last two or three years really was just trying to, trying to, trying to use AI to solve problem that we are already solving and making them more efficient, making them least less costly, reducing cost, unfortunately, firing people and replacing them with AI to then figure out, well, maybe.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:34:13]:
That was needed those people.
Dennis Traub | AI Engineering Specialist at AWS [00:34:15]:
Right. Maybe it was the best idea. Maybe we shouldn't have listened to the promises of AI now replacing everyone. AI is something that can help you solve new problems and AI shouldn't be used to solve a problem that's already been solved unless you are in the optimization stage. Big enterprises may be in that stage and big enterprises may be doing the right thing when they're looking at their processes and workflows and everything and think about well, where are the bottlenecks? Can we apply this to individual bottlenecks in here to make the overall process more efficient or more scalable or whatever. But especially in the startup space, you want to innovate and innovation. Innovation is creating something new or solving a problem that everybody thought was not solvable or didn't even think about solving because it didn't. Yeah, we didn't.
Dennis Traub | AI Engineering Specialist at AWS [00:35:13]:
We didn't think about calling our relatives in the US every day because it just wasn't possible. It was too expensive.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:35:23]:
Right, I see. I would be wondering what for our audience would be their first gen AI use case idea and what's holding them back from trying it. To top you, drop your comment or DM us on LinkedIn. We'll be back after a very short ad break. So let's talk a little bit more specific about the problems here. What do you think is the biggest misconception of non technical founders they have about AI, especially around model choice and privacy?
Dennis Traub | AI Engineering Specialist at AWS [00:36:07]:
There's two misconceptions, one I mentioned earlier that is that we mistake these things for humans because they use human language and that's the way our brains work. If somebody talks to it or to us or uses human language, our brain automatically thinks it's a human being. And by thinking this starts to make assumptions. And many of these assumptions just aren't true. And these assumptions lead us down a path where we get disappointed, where we get frustrated, where we feel like, well, it just doesn't work for me. AI just doesn't work for me. It isn't there yet. It will never be a human being.
Dennis Traub | AI Engineering Specialist at AWS [00:36:52]:
It will never be able to actually replace a human being in that sense. However, its capabilities are incredible, but they are slightly different. So that's the first misconception. And one of the things that we tend to do is give it names, which makes it even harder for us. Like I remember we had. I can say it because I don't have a device in here. There's Alexa from Amazon. Alexa, the device which uses human language, it talks to us.
Dennis Traub | AI Engineering Specialist at AWS [00:37:26]:
And when I talk to Alexa, at least the original version, not Alexa, when I talked to the original person and I asked it something and it didn't know or it didn't understand me because it was just rule based, more or less, it would say, I can't answer that question. And I would get annoyed. I would get, I would feel frustrated because due to the fact that it was speaking to me with a human voice, something inside of my brain thought, it's a human being. And the next thought was not even consciously, but probably unconsciously. How stupid are you? Why don't you understand? And that is, that is there's a break in communication happening because my brain makes some assumptions that the technology doesn't fulfill. So I was frustrated. The technology doesn't care, but I was frustrated. I felt like that doesn't really work until I really understood.
Dennis Traub | AI Engineering Specialist at AWS [00:38:26]:
Well, it works in a different way. I cannot, I cannot project human consciousness into it. And that's a misconception. Projecting human consciousness into the thing is a misconception. This is something to be really aware of. The other misconception, and that's an entirely different, different thing, is that you need the most capable model. That you need to really make sure that you get the most capable model to get started. That is a way, and that's a kind of procrastination because all the models are really capable nowadays.
Dennis Traub | AI Engineering Specialist at AWS [00:39:05]:
Sure, if you look at the benchmarks, the models are different and every day there's a new one which beats some specific capability over all the others. There's a lot of progress going on. But if you wait for the perfect model you'll never get started. You can use literally any of the frontier models nowadays. It could be one of the open weights models like Llama or Mistral. It could be one of the commercial models like GPT or Claude or Nova. It doesn't really matter. These models are capable enough to experiment with the first use cases.
Dennis Traub | AI Engineering Specialist at AWS [00:39:45]:
And once you've experimented and once you've found a product market match, once you found a use case that really works, then it makes sense to think about, well, does it make sense to maybe use a different model that's a bit more capable in this specific use case? Or maybe it makes sense to introduce a second model which is less expensive for part of the use case? Because for instance, for summarization, I don't need any reasoning capabilities, I just need a good summarizer model. And for the actual workflow orchestration, for the agentic workflow maybe that I'm going to build, I need a model that's actually able to do planning and reasoning. These are two very different capabilities and they have very different costs. So I might need, at a point in the future, I might want to look at different models and at their price, structure, at their capabilities. But to get started, just pick one. Just pick one. If you have GDPR or other privacy issues that you need to take into account, pick a model service, a model hosting provider that provides you this functionality that guarantees you, that tells you we don't store your data and all the data is being encrypted and we don't use the data to train our models and we don't send the data to anybody else. If you come to AWS to use a model on Bedrock, and no matter whether you use our own Nova models or you use Claude or you use Llama or any of the other models, we host these models ourselves.
Dennis Traub | AI Engineering Specialist at AWS [00:41:26]:
We're not a gateway to the actual model provider. We're not a gateway to LLAMA or to the Llama API or to the Anthropic API or anything. We host versions of these models in air gapped accounts. Nobody gets into this. These models and these models don't send anything anywhere. Everything that happens is just your request gets sent into the air gapped account, gets handed over to the model. The model itself is stateless, it doesn't store anything. It just takes your data, loads it into its GPU along with the model algorithm, with the model weights and processes that, and then it sends the response back and everything else just goes back to sleep.
Dennis Traub | AI Engineering Specialist at AWS [00:42:15]:
There's nothing that we store and well, we do store, obviously we store telemetry, we store that you actually called the model and how many tokens you use because that's how you ultimately pay for that. But we don't do anything with your data. We even have models running in Frankfurt that you can use so that you don't even have to send your data to the us. We even provide access to these models through our own backbone, so you don't even have to use the public Internet if you want. That's what model provider or non model providers, model hosting providers like AWS provide. It's a little more costly than just going to ChatGPT or to Claude AI or to the Llama API. It's more costly. But on the other hand, we have different terms and conditions and we make sure that you will be able to build GDPR or HIPAA or whatever compliant workloads using these models.
Dennis Traub | AI Engineering Specialist at AWS [00:43:21]:
And that's an important point. If you have pii, if you need to take GDPR into account from day one, make sure you work with one of the providers that actually give you these capabilities, give you access, give you encryption, make sure that they don't use your data in any other way so that you can safely say in your own audit and to your own customers, I know where your data is going and I guarantee that it's not being handed over to somebody without my knowledge or without your knowledge as a customer. That's the important thing. But to get started, I might even really start with a use case that doesn't even need these complexities because it introduces complexities. And that's the thing. As soon as you need to work with sensitive data, you have to think about these things. You may have to think about. Even if you have a workflow that uses data from your database which goes through a model in a GDPR compliant way and gets displayed somewhere to a client, you still need to make sure that the data isn't being displayed by accident to somebody who shouldn't have access to them.
Dennis Traub | AI Engineering Specialist at AWS [00:44:45]:
So you need to be able to ensure authentication, authorization. You need all the security and compliance mechanisms that make sure that not a random person on the Internet or just a random person inside of your company is able to just use the agent and access your customer data.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:45:06]:
I see. I was wondering for our audience, if you could safely test any AI idea without risk, what would you build, tag us or reply to us on substack or with your moochart? I have two final questions for this interview, Dennis, because we are already recording for more than 45 minutes. But. But I do believe they're very important thoughts before you even start thinking about applying AI. And we already know you guys support different models. You are more the infrastructure provider for something like this. But I was wondering, have you seen any clever AI adoption stories where companies started small and then scaled rapidly?
Dennis Traub | AI Engineering Specialist at AWS [00:45:57]:
The first thing really that I would look at is do I even need AI for that? Many of the things that we're trying to solve with AI nowadays, they have already been solved and probably in a good and much less expensive and much less ecologically impactful way. If you have a calculator that can add up to numbers, use a calculator. Don't ask an AI to do it for you. First of all, it isn't very good at it. Well, they're getting better at math. But why would you start up an entire cluster of big Nvidia GPUs to get the sum of two numbers? You shouldn't be doing that. So first of all, don't try to solve already solved problems. And the second thing really is again, look at the painful things that nobody ever tackled.
Dennis Traub | AI Engineering Specialist at AWS [00:46:54]:
Look at something that has been bothering you or your customer for a long time and it hadn't been addressed because everybody said, well, it doesn't just doesn't work and we don't have the time to do it ourselves.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:47:12]:
It's actually pretty good closing words. We will be back for one, the Founders Vault for our premium subscribers on substack and YouTube. And second, you'll be back for a second interview where you get more hands on when you go through all the thoughts you had that you need to think through before you can even get started on AI.
Dennis Traub | AI Engineering Specialist at AWS [00:47:36]:
Great. I'm looking forward to it.
Jörn "Joe" Menninger | CEO and Founder Startuprad.io [00:47:38]:
Me too. Have a good day. Bye Bye. That's all folks. Find more news, streams, events and interviews@www.startuprad.IO. remember, Sharing is caring.
📝 Copyright: All rights reserved — Startuprad.io™
Comments