When Apple announced its AI strategy, it built genuinely good on-device models for Siri, but partnered with OpenAI to answer whatever those models couldn't. The result in the public's mind: OpenAI was secretly doing Apple's job for them. In this episode of The Team Check-In, host Emre Ok talks with Jannik Reinhard, a dual Microsoft MVP in Security/Intune and AI Platform, previously an AIOps technical lead at BASF and now Head of AI at Epic Fusion, about how a solid product with the wrong marketing can hand a competitor a win it didn't earn.
Jannik explains why he thinks the industry is quietly shifting from giant, everything-trained large language models toward smaller, company-specific models trained on higher-quality data, and why results still come down to garbage in, garbage out. He also makes the case for "copilot," not replacement, as the right mental model for AI at work, weighs in on whether AI-written content can ever be original, and shares how Teamflect built its own AI feature directly out of data its users were already creating.
.svg)
.svg)
[00:00:00 – 00:01:54]
Emre: You're listening to the Team Check-In, powered by Teamflect. Hi, everyone, welcome to another episode of the Team Check-In. And with me today I have Jannik Reinhard. Jannik, hi, welcome to the show!
Jannik: Hey, I'm happy to be here, thanks.
Emre: Yeah — I think you're, by far, alongside Emanuel [surname unclear in source audio], one of the most technically proficient people we've had on the show. Usually it's a bunch of HR people, and other literature majors like me — in terms of technical and scientific literacy, we've been pretty much on the lower end of the spectrum so far. So it's awesome to have you here. Can you tell our audience a bit about yourself, what it is that you do — because you are a Microsoft MVP, after all?
Jannik: Yes, yeah, let me introduce myself. My name is Jannik, I'm from Germany. I'm working as technical lead in a large chemistry company. In the end, in my daily business, I work on AI
[00:01:41 – 00:02:56]
Jannik (continuing): solutions — figuring out how AI can help us improve the end-user experience, make our IT more reliable. Beside this, as you already mentioned, I'm also a Microsoft MVP, and with this I'm very happy to share my knowledge through my blog and public speaking with others. Currently I'm also in the process of writing my first own ebook — it's quite new for me, let's see how this goes.
Emre: Awesome. Is the ebook in German or in English?
Jannik: I will write it in English, it's English.
[00:02:19 – 00:03:34]
Emre: Awesome, and we'll plug the ebook further down the line, of course — but what's the topic, what's it going to be about?
Jannik: I'm a dual MVP — I'm an MVP in Intune and in AI Platform. This means I try to combine both, and I'm writing about automation, analytics, and AI with Intune. So I describe a bit of what you can do with out-of-the-box functionality that Microsoft provides, but I also go very deep into how you can build your own solutions, and explain it in a way that's good for beginners, but also has something for experts, so they can hopefully get some benefit out of it too.
Emre: And does being a dual MVP also mean that you don't have a lot of free personal time — sorry, does being a dual MVP also mean you don't have a lot of free time on your hands?
[00:02:56 – 00:04:12]
Jannik: I have one problem, and this is: I have too much fun in what I do. This means I dedicate a lot of my time to my job, to community work, because my hobby is my job — and that's a great thing, but on the other hand it's also sometimes a problem, because you realize, oh, I've invested a lot of time into this.
Emre: Yeah, I mean, I think it's one of those champagne problems — "oh, I love my job too much." Who cries for Jannik? But how did the MVP thing come about — how does one become a Microsoft MVP, let alone a dual Microsoft MVP?
Jannik: In the end, I was new in an environment at my company, and I'm a person who is very motivated, and my goal was to get up to speed very fast on these
[00:03:35 – 00:05:01]
Jannik (continuing): topics, to learn a lot about them. And what I decided then was also to write blogs about it — my goal wasn't only to learn it for myself, I also wanted to share what I learned with others. That was my starting point, and along this journey I had a lot of fun with it too. It got to the point where I was writing more or less weekly blog posts about topics I was involved in, and I also got involved in different community projects, open source, and this was then recognized by Microsoft, and I got an Intune, or what was then called Enterprise
[00:04:18 – 00:05:33]
Jannik (continuing): Mobility, MVP. And then I switched my job within my company, worked more on AI topics, blogged about it and spoke about AI things, and then this year I was recognized in a renewal, also in that category, and now I work on content for both.
Emre: Awesome. And when you started writing those blog posts, was there ever some kind of commercial motivation behind it — "oh, maybe I can monetize this somehow" — or was it just, "I like talking about this, I like writing about this"?
[00:05:32 – 00:06:37]
Jannik: I have no commercial motivation to do this — this means everything I blog, everything I do, is for free. My ebook, what I'm currently writing, is also free — I'll have some sponsors who support me in bringing it out, but in the end I have no commercial motivation, as already mentioned. It's my hobby, and I love sharing it with the wider community. And I think the biggest gift you can get is when someone says, "Hey, this helped me, thanks for writing this." That's more or
[00:06:04 – 00:07:06]
Jannik (continuing): less my motivation for why I do this.
Emre: That's awesome, that's awesome. I mean, I had the same feeling when I was writing for this movie website called Screen Rant — I was writing lists like, "Which one is the best Kingsguard in the history of Game of Thrones," or "Which Avengers character coincides with which Invincible superhero" — topics I would normally not shut up about day in and day out. Those things just came naturally, which is awesome. It's so cool that
[00:06:36 – 00:07:58]
Emre (continuing): you have something you're so passionate about that carries over into your day-to-day work life. So — the MVP situation, it is a lot of work, we've had another MVP on the show before, and it's expected of you, if you're going to have that title, to be producing content, right?
Jannik: Yes.
Emre: So what are some of the benefits of being a Microsoft MVP, in your opinion?
Jannik: In the end, as soon as you get this award, it changes a lot for you. What I have as a possibility now, and I really enjoy this, I
[00:07:16 – 00:08:32]
Jannik (continuing): can travel through the world, can speak at events, and this is very awesome, because it's a mix of doing what you love — speaking, creating content — but on the other hand you also meet a lot of people, see a lot of things. You also get to talk with product groups at Microsoft, participate with your feedback on product development, get a lot of insight. And also a nice thing is I've gotten to know a lot of friends during this journey, and it's always
[00:07:53 – 00:09:11]
Jannik (continuing): amazing what I've experienced over the last months and years, after getting this award.
Emre: Perfect. And I think what our audience will appreciate about this episode is — if you haven't noticed, we'll be talking about AI, this is going to be a very AI-centric episode. We talked about it before the show, and for the longest time we've had literature-major opinions on AI, and then, you know, people in leadership roles' opinions on AI in the workplace, which
[00:08:33 – 00:09:44]
Emre (continuing): I'm not discrediting — it's awesome to have multiple perspectives, and people who aren't experts can bring a certain outside point of view to topics. But again, I think for the first time in this discussion of AI, we actually have someone who is very well versed in it. So let's start right there, because it's not just this show — right now, AI is one of the most commonly talked-about things in the world. It is the biggest buzzword. If anyone and everyone
[00:09:08 – 00:10:28]
Emre (continuing): has a little AI icon on their apps, on their websites, everyone's talking about it — everyone's talking about "these are the best practices with AI," and it's like, "What is this guy's background? Nothing tech-wise whatsoever." So what do you make of the rise, the prominence, of AI in public discourse so far?
Jannik: Yeah, as you describe it, it's a very big hype, and I'm very convinced there are a lot of good use cases, a lot of potential — and
[00:09:48 – 00:11:10]
Jannik (continuing): this will change a lot about how we work, how we communicate, how we interact with content. I think that's very clear to me. But on the other hand, it's a hype — everyone uses it to promote their products, to promote themselves — and this is also something we still have to figure out: where does it actually make sense to apply AI, where is it really a benefit, and where is it just hype, a nice marketing thing, but not much real benefit behind it? Sometimes I also see it promoted
[00:10:29 – 00:11:39]
Jannik (continuing): as AI, but if you dive a bit deeper, it's not actually AI, it's just being called that because it's fancy. This is something that will also develop over time, to see where the strong use cases are. Beside that, I think we're currently in a phase where everyone's spamming with it, where everyone's getting a feel for it, trying it out, building up knowledge — and this is something we shouldn't stop, because it's a good trend that everyone gets some hands-on exposure, some learning of their own. But at some
[00:11:04 – 00:12:23]
Jannik (continuing): point this should become a bit more standardized than it is today, because every vendor is spamming out products, and my view is that in the future this will also need to come together a bit more, because no user wants to search through 20 different portals to find information in some custom copilot. I see it more as: you have one central overarching tool — like Microsoft Teams, or something else — where you type something in, and in the background it gets routed to the correct AI solution, the correct copilot, which can do the task
[00:11:43 – 00:12:35]
Jannik (continuing): for you, or provide you the information you're asking for.
Emre: Yeah, yeah — but Jannik, you're forgetting something, this is the internet, right? You can't have a "middle of the road" opinion — "oh, it's good, but it's also not as good as you think" — you're not allowed to have an opinion that's down the center, reasonable. No — we either have to say, "Okay, this is the new best thing, it's the greatest thing ever, it's going to change our lives, it's going to be so good, it's all
[00:12:10 – 00:13:14]
Emre (continuing): going to be AI from now on, we're gonna have AI dogs, it's gonna be all good, we're going to go into a scientific utopia" — or you have to take the other route and say, "No, this is the coming of the apocalypse, this is bad, we should not touch it with a ten-foot pole." You can't have a reasonable middle-of-the-road opinion on this, you're not allowed to.
Jannik: But I have it — I really like AI, but I think you always have to challenge it a bit too, because there
[00:12:41 – 00:13:55]
Jannik (continuing): are also other, still valid, technologies out there. I also see that sometimes a lot of pre-work has to be done before the AI can work well — tools have to be adapted, knowledge has to be made available — because with AI you always have the "garbage in, garbage out" problem, and this is often the challenge: you have to work on the input side to have high-quality data, and then you also get high-quality answers.
Emre: Yeah, exactly — and of course, with — I mean, with large language models, that's what "AI" turned
[00:13:19 – 00:14:31]
Emre (continuing): into in the public discourse, right? Right now, when people say "AI," we're actually just focused on the large-language-model side of it — that's just become the umbrella term for it. And you mentioned managing datasets effectively, garbage in, garbage out — there have been some recent controversies around OpenAI and where they're getting their data from, and I'd really love your opinion on this, because from the outside looking in, it looks like things are shaking up, and
[00:13:55 – 00:15:18]
Emre (continuing): everyone's so reliant on using these tools now — most people, regardless of their opinions on ChatGPT or other large language models, find them useful for automating everyday tasks, and I think everyone's now using it day in and day out for small things. But now, from the looks of it, some news is coming out saying that it's running out of training data — there are some reports saying ChatGPT is running out of internet to train itself on. So is this the end — is
[00:14:36 – 00:15:41]
Emre (continuing): it, is this as good as that particular product gets? Because a lot of opinions are on the side of, "This is just the Wright Brothers' airplane version of it, and it's going to evolve into something magnificent," but on the other hand, some say, "No, development is going to slow down, because we don't have any more data to feed it." So what's going on there?
Jannik: Yeah, that's a very good question, and I think this is exactly the challenge for the future. In the end, what we
[00:15:09 – 00:16:14]
Jannik (continuing): have today are very big models trained on a lot of data available on the internet — but I think, in the future, such companies will also have to buy information to make their models smarter, or, especially if we're talking about Europe, with data privacy concerns and intellectual property considerations, maybe the trend will also move toward company-specific GenAI models. This means if you're working at a production company,
[00:15:41 – 00:16:53]
Jannik (continuing): maybe you use all your own data and train a custom model that understands your company, understands your data, and can give more targeted answers than a generic model. But what I also see is that the trend is moving away from these large language models trained with a lot of data — but also with good and bad data mixed in — toward small language models, trained with less data, but higher-quality data. These models can then also be hosted on your smartphone, or your smartwatch, and can
[00:16:17 – 00:17:36]
Jannik (continuing): work without connecting to the cloud, to some data center outside — they can process directly on your device, and give you a similar quality of result to these big models, but without any outgoing traffic to some black box. And I think this is more or less what Apple did with their recent operating-system announcement. I think Apple is very good at marketing, but on this particular point, I think they did something wrong. Apple did
[00:16:57 – 00:18:08]
Jannik (continuing): a very good job with the model they developed — they have models hosted on the phone, it's go—
Emre: — before we go into analyzing this for listeners who haven't seen it, what was the announcement, and what's wrong with it?
Jannik: Yes — I think Apple announced, as part of their Apple Intelligence strategy, that Siri is now much smarter. Everyone who's used Siri knows the answers you get are sometimes pretty bad, and here Apple's invested a lot to give you smarter suggestions when you're inside an app, things that
[00:17:32 – 00:18:38]
Jannik (continuing): could help you, it writes emails for you, all that kind of stuff, and Apple did a very good job — they have a very solid approach to this, and the model is partly hosted on your phone, but it's also federated to a data center. This means if your phone isn't powerful enough, the task gets forwarded to their data center. This approach is good — but in their marketing, they had a collaboration with OpenAI, and said, essentially, "OpenAI can answer all the questions we can't answer." And what the
[00:18:05 – 00:19:13]
Jannik (continuing): impression outside was is, "Oh, OpenAI is doing the job for Apple" — but that's wrong. Apple did a very good job, but in the way it was promoted, the outside feeling was that they relied heavily on OpenAI — and that's not accurate. They only forward to OpenAI if, say, you want to know something like the size of a specific company, if you need general public knowledge — because that's the actual problem for Apple. Apple is focused on data privacy, meaning they don't collect data, and without data it's hard to
[00:18:39 – 00:20:04]
Jannik (continuing): train a general-purpose model, the way OpenAI, Google, and all the other companies do.
Emre: And that's very interesting — passing the ball straight to OpenAI. I think, at a time when OpenAI really needed some strong, positive marketing, Apple gave that to them, essentially for free. Moving things along from the general AI discussion to the more corporate side, something we're more used to talking about here — you've studied and worked on digitalization in the corporate world, so how do you think AI will affect
[00:19:22 – 00:20:40]
Emre (continuing): that process? People are moving — we're obviously moving from manual, pen-and-paper processes to digital ones, right — digital transformation is a huge part of today's work environment. So how does AI fit into those digital transformation discussions?
Jannik: For me, the word "copilot" is more or less perfect for this, because I'm very convinced that, in one, two, or three years, your work will not be stolen or taken over by AI — I think it will only support you. This
[00:20:01 – 00:21:12]
Jannik (continuing): won't apply to every job — I'm sure jobs like content writer, or similar roles, could be partly taken over — but for most jobs, it's more that AI relies on you, makes you better, makes you stronger, because you can concentrate on the hard tasks, the tasks where you as a person bring value, and everything like formulating an email — "Hi, how are you, blah blah blah" — that's something AI can do for you, or searching for content. I think everyone knows the feeling of, "Where's this document stored,
[00:20:37 – 00:21:48]
Jannik (continuing): where can I find this information, I have to go do research" — this is all lazy work, why should I have to do this myself? I can concentrate on the stuff where I can bring my own knowledge, my own strengths, instead of doing all this lazy work. And this is why the term "copilot" is very good — copilot takes over everything that's lazy work, and for everything where you need a strong human behind it, that's something we can still do. I like it.
Emre: In the end, I think there's a very nuanced discussion to be had around which jobs AI
[00:21:12 – 00:22:23]
Emre (continuing): can do for us, which jobs it will do for us — I mean, in the simplest terms, yes, like you said, the busy work, the lazy work you were talking about — it's perfect for that, it's amazing for that. I think, being in the content-writer space, the first two to three months of the ChatGPT craze were terrifying — it was incredibly scary for me, thinking, "Oh no, it's doing a darn good job, it's really good." But then, what we started noticing, especially in the search-engine-
[00:21:48 – 00:22:58]
Emre (continuing): optimization department — what we started noticing is that if a gajillion blog writers put the same, similar prompts into ChatGPT, come up with outlines, and start publishing, what you end up with is just variations of the same post being published everywhere, and suddenly originality scores start going down, search engines start punishing AI-generated content, and it becomes a whole thing. And I was thinking — if we're all putting
[00:22:22 – 00:23:24]
Emre (continuing): the same prompts into this thing and letting it generate the content, the results aren't going to be so different from each other, right? They'll be pretty similar. But then, if you think about it — unless you're writing content for a very specific, very technical niche — if you're writing evergreen content, "five ways to keep your employees happy," that kind of thing, which
[00:22:53 – 00:24:14]
Emre (continuing): I'm very guilty of writing all the time — even before large language models existed, people were pretty much writing the same thing, the same ten to fifteen pieces of advice were being recirculated, because in some situations the "right" answer is pretty obvious. And just like large language models, we're influenced by everything we see — we take in everything around us, put it together, and put it back out. So there's a bit of a philosophical discussion to be had around
[00:23:34 – 00:24:40]
Emre (continuing): whether AI-generated content is truly "original" — well, is human-generated content truly original either, since it's also an amalgamation of everything we've seen, with our personal spin on it — except that personal spin is also influenced by everything else we've seen. And then you just end up going down this ongoing rabbit hole, which is pretty terrifying.
Jannik: Yeah, in the end it's also a bit of a chicken-and-egg problem, because the model can only learn if a person generates
[00:24:07 – 00:25:28]
Jannik (continuing): content, but once a person no longer generates content, because the model helps you in most areas, then why should I still write this content myself? So it's also a hard question, because it's chicken-and-egg — they both need each other going forward.
Emre: Yeah — which one came first, the tech bros or the AI apocalypse — it's just back and forth. So how do you see companies changing their mindsets as this AI trend keeps rolling on?
Jannik: Yeah, I think this will be one of the hardest challenges we have to
[00:24:48 – 00:26:04]
Jannik (continuing): tackle in the coming months. I think what we have out there already are quite good solutions — I won't say they're amazingly good, because I also use Microsoft 365 Copilot, and it's good, but expectations are sometimes higher than what the products actually deliver. But the goal now is to enable your employees — meaning, you can buy the product, so the products are quite good, but someone actually has to use it, and without knowing how to use it, and without actually using it, it brings you no benefit, only more cost, more licenses. And this is something
[00:25:26 – 00:26:54]
Jannik (continuing): companies have to work on — enabling users to change the way they work, but also educating them on what counts as a good prompt, when it makes sense to use it, having to check what comes back — is it a 100% answer, or only 80%, and do I have to refine it a bit? And this, I think, is the whole challenge of user enablement around AI, getting these tools out onto the street, so to speak — what all these tools can actually deliver. And I think, especially in Europe, but it's more or less a global problem too, it's also about
[00:26:09 – 00:27:30]
Jannik (continuing): regulation around this — AI is very good, very helpful, but wherever there's something good, there's also something bad, and this is also something that should be a focus, to build regulations so AI doesn't cause too much harm.
Emre: Yeah, I think the user-enablement part you just talked about is incredibly important, and that's where many organizations will need to adjust their mindsets, because sure, AI tools right now are a dime
[00:26:50 – 00:27:53]
Emre (continuing): a dozen — like our latest guest on the show, Daria Rodnik, she was working on an AI leadership-coach assistant. Everyone's coming out with great, innovative tools — but no matter what the tool is, it's only as good as the prompt, as the data you feed it, it's only as good as how you direct it. And I think prompt engineering, and this near-"AI literacy" kind of thing, is where people will need to be trained on how to use it,
[00:27:22 – 00:28:35]
Emre (continuing): so that what you get out of it isn't generic filler, but actually sophisticated, genuinely useful output. Because, for example, in the discussion of AI art — you've seen some AI art that, at a single glance, is very obviously, quote-unquote "AI art," and generic. But then you have people doing genuinely incredible, nuanced work with it, using it through, I'm assuming, very complicated and layered processes to come
[00:27:57 – 00:29:09]
Emre (continuing): out with truly incredible results. I think when it's used as a sophisticated tool, and not as this instant-answer, instant-result machine, the end results can be mind-blowing. But I think, down the line, there will need to be more education on how to best optimize what you're getting out of it, how to best use it — that's the direction people will need to go.
Jannik: Exactly — but I also think the technology itself will develop. This means, yes, you have to work on the user side, so they
[00:28:34 – 00:29:48]
Jannik (continuing): know how to prompt, but the technology will also get better and better, and will increasingly compensate for weaker questions, weaker prompts, to still get good results out of it. But it's the same as when I talk with you — the more precise I am, the better the answer you can give me, and the same applies in this world too.
Emre: Yeah, exactly, exactly. And there are plenty of different applications of these AI tools in the everyday corporate workplace, I
[00:29:11 – 00:30:07]
Emre (continuing): think — and normally this is where I turn to the camera and do an ad-plug thing for Teamflect, our product, but I just want to tell you about it, because it's something we've been working on and we're pretty excited about it. It's been out for a while, we've been getting great feedback from our customers, but obviously, as a performance-management product, we were thinking, okay, how do we make the most of AI, how do we implement it in
[00:29:39 – 00:30:40]
Emre (continuing): a way that actually makes sense for our users, right — in a way where we're not just slapping "AI" on our web page and saying, "Look, look, look, we have this, we're on the same wagon as everyone else." We didn't want to do that, so we held off on it for a while — we had some ideas, but we were, "this is too gimmicky, this is too gimmicky." What we have, though, is a performance-management solution where employees exchange feedback all the time, give each other recognition,
[00:30:09 – 00:31:12]
Emre (continuing): complete their goals, all built right inside Microsoft Teams — and we realized, hey, look, we have this amazing data that people create inside their own company, on how they work, how they're completing their goals, the 360-degree feedback from people outside the organization — everything. Look, this is really, really valuable stuff. So what we started doing with it, inside our performance review templates, is we added a little
[00:30:41 – 00:31:42]
Emre (continuing): section — we'd always had an employee development-plan section, where reviewers and reviewees could work on, okay, how do we improve, how do we create development plans, set development goals for the employee — and now we've included an AI-based assistant, AI-based suggestions, in that section. What it does is take into account everyone's feedback, all the feedback the employee received throughout the review period, their goal-completion rate, their tasks, and
[00:31:11 – 00:32:17]
Emre (continuing): everything, and uses the data created through all of that to come up with, hey, here are some really solid development-plan suggestions for this employee. We've gotten great feedback on it, and I think the best part is that it's the data you create, the data within your own organization — and when it's organic like that, built in organically like that, I think it can do wonders.
Jannik: Absolutely, and this is also the perfect example of it never being a 100
[00:31:44 – 00:32:58]
Jannik (continuing): percent solution — this means it helps you save a lot of time, gives you good proposals, maybe not 100%, you might have to adjust it a bit, but it's a good example, it saves your time, it takes over some of the work, and it does an awesome job, because it learns from what you already have.
Emre: Yeah, yeah, exactly. And I think the more we use it, the more time we spend with these tools, the more they get used to us, it just keeps getting better and better and better. So — absolutely, Jannik, we're approaching 40
[00:32:21 – 00:33:43]
Emre (continuing): minutes, and this has been an awesome chat overall — it's been really nice having you on the show. Before we wrap things up, let's talk about the book — let's go into detail, when's it coming out, what can people expect from it, and where can they find it?
Jannik: Yes — my goal was that it would already be out, but I didn't reach my own goal, because I got very involved in some topics at my company. I think the release will be early next year, and people can expect, as I
[00:33:02 – 00:34:14]
Jannik (continuing): mentioned before, that if you're new to this space, there's a good on-ramp where you can get started — you'll learn a bit about how things work in the background, how you can build your own solutions — but there's also something in there for people who are already experienced, where I want to give them some concrete things to work with, some solutions, some architectures they can learn from, that they can apply one-to-one in their own company. As for where people can find the book —
[00:33:38 – 00:34:59]
Jannik (continuing): I'll publish it on the Apple Books store, but I'll also make it available on my personal blog, jannikreinhard.com, where you can also download it.
Emre: All right, and what's the working title — what will the main title be for the ebook?
Jannik: To be honest, that's the point I want to figure out at the end. I'll start by writing the content, and then I'll do the creative work of designing the cover, creating the title — though even there, I'll get some advice from AI, because sometimes AI is more creative than I am.
Emre: Awesome, awesome, Jannik, it's been great to have you on the show, and
[00:34:19 – 00:35:14]
Emre (continuing): people can reach you as Jannik Reinhard on LinkedIn. And if you're listening to us on iTunes, Spotify, or wherever you find podcasts, leave us a like, subscribe, it really helps the show out. And if you're looking for a performance management solution built right inside Microsoft Teams and Outlook, you can always give Teamflect a try — all the links you need from this episode will be in the description. That's it, bye-bye, everyone!
.svg)