Friends of Dentro

Welcome to the third edition of Friends of Dentro! 🎉

We set up this newsletter to keep you in the loop of what we’re up to at Dentro. Our work with clients, our own products and everything along the way. A wild mix of learnings, thoughts and things we’re working on. For anyone interested in AI, digital products and Dentro in general.

In this edition we will be talking about the rise of large reasoning models (Deepseek, anyone?) and how they might play out in a business context. We'll also recap our last weeks and share insights on what else we’ve been up to recently.

Thanks for reading along!

After Large Language Models come Large Reasoning Models

It’s been a whirlwind few years in AI. Large language models (LLMs) like the GPT series from OpenAI have transformed how we interact with technology, from chatbots to code generation and more. But as impressive as these models are, they have fundamental limitations: they don’t truly reason. They lack structured problem-solving abilities, struggle with multi-step logic, and don’t always grasp real-world causality.

Enter the next frontier: Large Reasoning Models (LRMs). If LLMs are about fluency, LRMs are about understanding. The goal isn’t just to generate text but to think through complex problems, analyze contradictions, and apply reasoning skills more similar to human cognition.

Example: when you ask a "normal" LLM "What's the capital of France?", it will output:

"The capital of France is Paris."

With a reasoning model however, the model first generates "thinking tokens" that model how a human would think. So the reasoning would output something like:

"<thinking> Alright, I have to respond with the capital of France. France is a country in Europe. The language is french. The capital is Paris. The capital is the same over the last 100 years, it didn't change recently. So I can confidently say that the capital is Paris. </thinking> The capital of France is Paris."

Btw, you might have heard of Deepseek-R1, a Chinese AI model that made waves in the last weeks. That’s a reasoning model as well, another big one being o1 by OpenAI.

What It Means for Companies

If LLMs were step one, think of LRMs as step two – where AI goes from an impressive autocomplete machine to a system that’s one step closer to actual human thinking processes. For companies, this could mean AI moving from an assistant role into more strategic decision-making functions, automating reasoning-heavy workflows in various areas.

At Dentro, we actually see huge improvements in output quality for some use cases already. Things that hardly worked a year ago are now possible close to perfection.

To sum this up, we’re still in the early days of Large Reasoning Models, but we believe the shift from LLMs to LRMs will redefine AI’s role in business and beyond. Excited for the ride!

Tweet of the month

We’ve just explored the rise of AI models capable of thinking more similar to humans. In order to better understand their potential impact, we’d like to give you a practical example.

Let’s say you run a little company that builds software products. User report bugs and feature requests, which go into a backlog – ranked by potential impact, estimated effort, or else. Your job is then to distribute these tasks within your development team. Oversee development. Provide feedback along the way. And eventually approve the result.

With an advanced AI system in place, this would look a bit different. Let’s say users request a dark mode in your application. Instead of a developer manually writing the code, the AI analyzes the existing design, generates the necessary style changes, and submits a pull request for approval. Or if users for example report slow performance in a specific dashboard. The AI detects the bottlenecks, optimizes the code, and suggests an update. All before an engineer even looks at the problem.

In this setup, product managers and engineers shift from building every feature from scratch to curating AI-generated solutions . Reviewing, refining, and approving changes rather than writing them line by line. This could dramatically accelerate product iteration cycles and free up engineering teams to focus on deeper architectural challenges rather than routine fixes.

Apart from this, there’s thousands of other business scenarios that can potentially be optimized in a similar approach!

We may not be there yet, but the foundation is already being built. The real question isn’t if this will happen, but how soon. (Expect very soon!)

Building activities 🛠

• Started working with a new client, building machine learning models from scratch. We’ve got 10+ million rows of data to build a predictive model with. All prep work is done and we are already getting good results. Now we are gradually improving upon those, in order to hopefully reach a sufficient level soon.
  
  

• Thought up a new application called Paolo – a B2C app that assists users with fashion choices. It consists of a chat, a virtual wardrobe and some style quizzes. Frankly, fashion isn’t exactly in our circle of expertise, but we think there could be demand and from a technical perspective it’s an interesting project to build. So far we have a landing page (paolo.chat) with a waitlist and will start building properly once we see sufficient interest on that. Watch a quick showcase video here.
  
  

• The AI phone assistant we’ve built for a client went live. Anyone who calls at our client’s office after hours or when all lines are busy, will talk with the AI assistant. First live feedback has been very good and it’s interesting to see how differently callers react to it. Some are immediately down to give it a try and are having actual conversations. Others mistake it for a automated mailbox and just throw some keywords instead of talking to it like a human. We’re quite happy with the results yourselves and will keep refining and improving upon the feedback we receive.

  
  

• Included encryption at rest and encrypted backups to our chat apps. This means all messages are automatically encrypted on the server and for the unlikely case that things go very bad, we’ve got encrypted backups in place and are able to be up and running in no time.
  
  

• Built a new landing page for ourselves, it’s in German and we use it for various marketing initiatives. As any other landing page, it’s not directly accessible from our website, but you can check it out here if you like.
  
  

• Introduced our new AI team member Paula to the world. We have already created character consistent photos as well as a dedicated voice (of course with an Austrian accent 😉) for her. Now there’s videos as well: for example see Paula explain some AI concepts, rapping an Eminem song or showcasing our new Paolo app.

Background activities 👨‍💻

• Did our yearly review meeting to start the year right. At Dentro, we work with monthly goals and reviews, but once per year we take an even bigger step back, reflecting on the old year and planning for the one ahead. We now have a clear roadmap for the next months to come and are looking forward to execute on it as well as possible.
  
  

• Started recording our internal calls in order to be more efficient. We plan to automatically store and transcribe them, in order to be able to create summary notes or ask questions about it later on. Our challenge: speaking German. Austrian German to be exact. Current transcription solutions (voice-to-text) work best for the English language and not that well for other languages (yet). Luckily, these days we’ve found a provider called AssemblyAI, which does a very good job transcribing our recordings. Will go with them though and will soon finally be able to have our automated setup up and running.
  
  

• About to start a project with a new client who needs a local AI setup. Due to being engaged in a rather sensitive industry, the usual cloud and API providers are not really an option in this case. Which is great for us, as it allows us to get creative and dig deep into what’s possible in terms of running AI activities on physical on-premise servers. In a first phase we’ll mimic different setups in a cloud environment in order to then be able to replicate them on the client’s hardware. (We also wrote a blogpost about this, btw.) All paperwork is in place and we’re looking forward to kick things off in mid-February.
  
  

• Did a data security review of our client applications, looking at data storage and retention policies of all software providers we currently have in place. Nothing new nor critical surfaced, but it’s nevertheless a good a idea to take a step back from time to time in order to have the overall structure and foundation in a good place. We even plan to engage an outside expert to assess our infrastructure setup later this month to be on the extra safe side.
  
  

• Started using Microsoft Outlook and Teams properly. To be very honest, whenever we work with Microsoft products, we experience issues – so we’re definitely not eager using them. Nevertheless, our corporate clients all have Microsoft in place and in order to support them the best way possible, we set up corporate Microsoft accounts from Dentro and are now perfectly able to mimic, replicate and test certain behaviour and setups.
  
  

• Realized how long corporate sales cycles actually are. As a company that is used to move quite fast, it’s a bit astonishing to see how much time corporate clients require in order to make decisions. With one of our latest clients for example, it’s been 9 months between first call and actually starting the project. For us that’s alright though, we can adapt and are ready to get going when they are.
  
  

• Created social media accounts for Paula. You remember, our new AI employee we’ve introduced above. Apart from being a regular at Dentro’s company channels, she’s now got her very own accounts on X and TikTok – go follow!

And this wraps up the third edition of Friends of Dentro. Thanks so much for being a part of it and reading along as Dentro evolves!

Questions, feedback, or anything else you’d like to tell us? We’re happy to read it all
– just hit ‘Reply’ on this email.

All the best,
Paul & Paul (and Paula) from Dentro

PS: You can share this newsletter with anyone who might be interested, just send them this link: https://friendsofdentro.beehiiv.com/subscribe