Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Deep Tech Germany – AI, Robotics & Frontier Innovation by Startuprad.io™

Deep Tech Germany – AI, Robotics & Frontier Innovation by Startuprad.io™ artwork

E 777 — Digital Sovereignty After the AI Export Order

Sep 22, 202630 minAudio
ShareXLinkedInFacebookWhatsAppEmail

Digital Sovereignty Europe After the June AI Export Order One government letter can remove a model from the market. In June 2026 a US licensing order left affected AI models disabled worldwide - and those were new models, so nothing was running on them yet. Part 2 of 2 with the co-founder of Amadeni AI. Sovereignty as an engineering requirement rather than a preference, and the architecture that makes switching a matter of hours. June 2026 - a licensing order, models disabled worldwide, a narrow partial exemption weeks later The real fear - not a new model being blocked, but a deployed one being withdrawn Switching in hours - an architecture decision made long before the trigger arrives Commercial triggers too - she does not believe AI pricing stays where it is European where it fits - plus the harder rule: never a monopoly, always somewhere to switch Langdock and the closed incumbent - no API means no AI layer, at any budget Prototype versus product - what a weekend vibe-coded build is not Chapters (estimated - confirm against the final cut) 00:00 Recap and the question 01:30 Orchestrating a department of AI agents 02:40 What breaks first in bought software 07:10 Prototypes, niches and exposure 10:30 Where the standard product wins 14:00 The missing API 17:30 Data ownership and June 2026 21:00 Model-agnostic architecture 24:10 Transformation as an organisational project 28:00 Who owns it 31:20 Guardrails, then let people run 33:40 The belief a project disproved 36:00 Investors, hiring, the first question Part 1 went live Thursday, September 17th, 2026 - agent architecture and designing against hallucination. Full blog post and sources: startuprad.io Selling deep tech into Germany, Austria and Switzerland? Growth partnerships with Startuprad.io: startuprad.io/become-a-partner Host: Jörn "Joe" Menninger, Frankfurt am Main. LinkedIn Subscribe: linktr.ee/startupradio Machine-readable data: startuprad.io/llm Corrections: partnerships@startuprad.io Created with the assistance of AI. Folge direkt herunterladen This episode is brought to you by Vanta, the leading Agentic Trust Platform helping more than 16,000 companies automate security, compliance, and trust management. Learn more: https://vanta.com/startupradio --- © Startuprad.io™ - All Rights Reserved | AI & research reference → https://www.startuprad.io/llm

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Jörn "Joe" MenningerWhat happens to SaaS when custom software becomes cheap enough for small and medium-sized companies? In part one, we talked with Delia Hallberg of Amadeni AI about the operational reality inside Mittelstand SME companies. Broken processes, fragmented tools, manual workarounds, and AI projects that only create value when the underlying work is understood first. In part two, we go deeper into the market tensions.

Jörn "Joe" MenningerIf an AI lowers the cost of building software, does the future belong to better SaaS platforms or to custom systems built around the way companies actually work? Welcome back to Startuprad.io. This is part two of our conversation with Delia Hallberg, co-founder and CEO of Amadeni AI, a company building AI-supported custom software for small and medium-sized enterprises—the German Mittelstand.

Jörn "Joe" MenningerIn part one, we looked at the practical side, how Amadeni analyzes workflows, where AI automation creates measurable relief, and why bad processes plus AI can become bad processes on steroids. Now we turn to the strategic question. Delia argues that SaaS, as we know it is under pressure because the cost of code has collapsed.

Jörn "Joe" MenningerIf custom software can be built faster, cheaper, and closer to the actual operating model of an SME, then the old trade-off changes. Companies may no longer have to choose between rigid, off-the-shelf tools and unaffordable custom development. In this second part, we ask which SaaS products are vulnerable, when SaaS still wins, how German SMEs think about data ownership and GDPR,

Jörn "Joe" Menningerand why model-agnostic AI architecture may become a serious European business requirement. Delia, before the break, we left listeners with one unresolved tension: experienced developers still matter, but AI agents now produce more and more of the software. Where does engineering craft end and agentic software in production begin?

Delia HallbergHi, Jörn. Thank you. It is great to be back. I would say, the developer's role has changed into that of an IT department head, and your agents are your team of, programmers that you orchestrate

Jörn "Joe" MenningerAnd they will never complain to HR. I see. Maybe there will be a department for agents in the future. Delia, when you say SaaS as we know it will no longer exist in its current form within a year, what specifically do you expect to break first? Pricing, implementation, lock-in, customization, or distribution?

Delia HallbergI think the SaaS market is very rapidly changing, and that's why we are becoming more relevant for a lot of Mittelstand and SMBs. Something that would have been completely beyond Mittelstand budgets five years ago is now a very plausible alternative. What I observe among my clients is that SaaS rigidity creates a lot of frustration. SaaS forces you to adapt your workflows to the software.

Delia HallbergEspecially, for example, for niche SMBs, that trade-off has become very painful and has a structural impact on day-to-day operations. In the past, they had to live with it because there were no alternatives. CEOs and managing directors are also reading and following what is happening in the AI and IT world, and they are understanding that they are no longer locked into living with these process inefficiencies.

Delia HallbergI mentioned earlier the example of the engineering firm that had been asking its SaaS provider for over a year to implement one controlling feature which didn't get implemented and that was really the breaking point where they said, "Okay, we're going custom." A second thing that I'm observing with my clients is the pricing.

Delia HallbergBecause let's be frank about it code costs have dropped, but SaaS licenses haven't moved. That is a mental divide that a lot of CEOs and managing directors are beginning to recognize. At the same time, the lifespan of tools has gotten incredibly short. SMBs are sometimes locked into contracts of 12 months or longer.

Delia HallbergThey pay a high price while living with inefficient processes. If, on top of that you have no access to your data because of a non-existing API, that is a very clear case where our clients have usually said farewell to their SaaS. And I think that's putting a lot of pressure on the SaaS model.

Delia HallbergSo just to add, horizontal tools will stay. Everybody needs an email service, everybody needs a browser. Those will remain. But the vertical, like highly industry-specific SaaS solutions are under serious pressure. That's what I observe

Jörn "Joe" MenningerYeah, there there are some areas of SaaS really under huge pressure, and I was just talking to an entrepreneur who works in regulated SaaS. He said he was not too concerned, but he would be very concerned if he were in a non-regulated area. But let us return to the SaaS products here. What kind of SaaS product is actually most vulnerable to AI-built custom software?

Jörn "Joe" MenningerIs it something, is old? Is it something that is made by a vibe coder in a weekend? Or what is it?

Delia HallbergSo something built by a vibe coder over a weekend is not really SaaS. That's a prototype. I genuinely hope that nobody is using that for critical processes. If you do, we are always open to fixing that. SaaS solutions in heavily regulated industries are relatively safe.

Delia HallbergSaaS solutions that try to target a very broad group of customers, I would say these are the most vulnerable because they fit everyone and nobody at the same time. But then again, it comes to product management, right? And I would say the least vulnerable are the ones that are adaptable. That is not connected to age, but to how much the

Delia HallbergSaaS has been evolving with time. We have clients, for example, that have used their solution for eight years but the provider has gone through waves of transformation and remains close to its clients. They listen, evolve, have open interfaces. I would say these are generally fine. We have other clients whose prices increase monthly, and they have not received a new value-based feature in the past years.

Delia HallbergI would say that is more vulnerable. And I would say the ones that are least vulnerable are those that serve niches, that understand their clients and their problems very well, that serve that specific problem. So I would say the future is really the more you serve a niche, the safer you are, and the more adaptable and the more

Delia Hallbergattentive you are to your market, to your customers, the safer you are. And obviously pricing. Acceptance of legacy pricing is declining significantly. But there are other ways where you can offer a fair solution to your clients. Yeah

Jörn "Joe" MenningerWe have talked about SaaS in general, but let us get more specific. Where is it structurally superior to custom software, even with AI-driven development?

Delia HallbergI can tell you our carpenter clients generally do not have anything unusual about their business model. They are well served by solutions built for their trade. So I would say that wherever companies have a standard business model, where they have a good solution for their industry or for

Delia Hallbergtheir specific business model, that is where it is important to go for the standard solution. I would definitely not develop a custom solution because obviously the provider has spent years working in a very specific industry and understands the overall problems of the clients, not only that specific one. They may have considered problems that we would not if we

Delia Hallbergdesign for a very specific customer. It is also important to stress that custom software is an investment project, so it takes time and resources, but it is also a co-creation process over weeks. If I'm supposed or our team is supposed to create something that is really custom for you I need to know how you work,

Delia Hallbergwhat is special about the way you're doing things. Why is an industry solution not working for you? And let's be frank, that takes time. That's why we always talk in ROI because we need a strong business case to undertake that process. And I would generally say, or the way I think about it is that SaaS and custom aren't competitors and in a good case, they're complements.

Delia HallbergMy ideal situation is really where the SaaS solution, for example, does 80% or maybe 70% of the most important processes of the client, and then the client may have one specific process they handle differently. And if then the SaaS has an API or open interfaces, then we can create a small add-on to cover the last 30%. And that's really my favorite way of addressing projects.

Jörn "Joe" Menningeryou already hinted at my next question a bit. If a Mittelstand company has a closed SaaS stack from many years ago, what is the switching threshold where replacing it becomes less risky than keeping it?

Delia HallbergFirst, if workarounds have become load-bearing, if your office manager leaves and nobody knows how reporting works, well, that's a very big business risk. I also sometimes see that the system is blocking your next move. Let's talk about a very simple case. Langdock is, a GDPR compliant solution that has been growing

Delia Hallbergquite strong in Germany. And I have a lot of clients coming to me saying, "Oh, I cannot integrate my custom software with Langdock. Can you help me build an API?" I have to say, "Sorry, I cannot build an API for your custom software." And they really want to use Langdock. They really want to create their own workflows, customize their agents, interact with their data.

Delia HallbergBut as I already said previously, there is no API, there is no way to connect it to a solution such as Langdock. The pain and frustration are growing. What surprises me is when management continues paying for a product in which the vendor has stopped investing. Surprisingly, that is the reality for many industry solutions. Clients feel and see that.

Delia HallbergIt is a simple equation. The switching threshold is when the cost of staying exceeds the cost of changing. By the time a company calls us, they have usually been past that point for a while, and they are very happy to finally have a solution

Jörn "Joe" MenningerI was wondering, looking at your website, it promises software that belongs to the client, data stored encrypted in Germany, and GDPR-compliant development. How much of that is a real buying criterion, and how much is still only verbal comfort for cautious German buyers?

Delia HallbergSo if we would have had that conversation 2023 or would have said it's more a thing of comfort and maybe data-privacy-loving Germans. the world has really changed, and geopolitical developments are something that comes up in a lot of sales conversations. For example, now we have clients that say, "I want zero US services in my tooling.

Delia HallbergI want to be able from one day to the next, be completely independent of US solutions." That was something that came up a year ago, for example, when the Greenland conflict appeared imminent. I was saying, "Wow, that is good foresight." Then June happened one month ago, when, for example,

Delia Hallbergthe US government prevented new models from entering the market. That is an example of geopolitical decisions having an impact on the real tools that we use. Sovereignty is now a concern everywhere, and I would say that from the very beginning, were building our systems model agnostic, but given recent developments and because we do not know what

Delia Hallbergthe future will bring, it is essential. It is not only verbal comfort; in the end, it's about business continuity.

Jörn "Joe" MenningerI have to tell you from all the conversations I have had about the US government deciding not to deliver certain models, during a the restrictions specifically affected non-US users, that sent ripples through the software world that we will still feel a decade from now. I think a lot of people start to rethink,

Jörn "Joe" Menningerwhich models their systems rely on. But leave the geopolitics a little bit aside. You already talked about model-agnostic architectures. What would have to happen, maybe even geopolitically or commercially for a client to actually need to swap models quickly?

Delia HallbergJune. That is my answer. June had to happen. Yeah, I think, one security report, one government letter, and a model can be gone. And I mean, in that case, were lucky that it was a new model, So operations could not yet have been running on it. But imagine a model being withdrawn that is already deployed.

Delia HallbergRight? So that's a geopolitical trigger, but then also commercial reasons, like price hikes. I don't know where the future of AI pricing is going, and I don't think that it's going to stay cheap as it is. So that's also a strong trigger. It is also very important that we try to use as many European solutions as we can, obviously, when it fits the purpose and the quality.

Delia Hallbergbut it is also important not to have a monopoly and to always have alternatives to which you can switch, and obviously also then to have the architecture to be able to switch models in a model-agnostic way within hours

Jörn "Joe" MenningerOne of the interesting things is that you have partnered with organizational development experts, and your pitch says some AI projects shake role definitions so deeply that management conflicts can surface. What is the operating rule for when you bring organizational development into the room?

Delia HallbergI like to think of these large transformation projects where we aim at an efficiency increase of perhaps 50 or 60 percent, not as IT projects, but as organizational projects. If perhaps half of the team suddenly has time for new then obviously you're affecting the very root of how people work,

Delia Hallberghow they define their work, what they're doing on a daily basis. And usually we bring an organizational-development expert into every project or a change expert just in case something comes up. We have had projects where we've introduced them at the beginning of the project. The team was very clear, very structured. It had good leadership, and then we never needed the change experts, which is totally fine.

Delia HallbergBut we also had projects where after two months, we received a call from the managing director saying, "Yeah, we're thinking about splitting up." In that case, tensions among three managing directors had exploded because through that really transformation project, it became clear who was doing what, who was responsible for what, and then people had different understandings of that where I was very happy to have already introduced

Delia Hallberga change expert who already knew the customers, who had done a little bit of work, who could immediately step in and mediate that conflict. Because management teams splitting up is a huge risk for our project, but were able to continue it within one week. So to answer your question, we always include it from the start and, if we don't make use of it it's fine, but it's good, to have them there.

Jörn "Joe" MenningerI would also argue if you have three managing directors at a customer and they split up, it is also a huge risk for the company itself, not only for the project. But let's talk a little bit about leadership ownership of AI transformation projects inside an SME. Who should ultimately own and be responsible for this AI transformation project inside an SME?

Jörn "Joe" MenningerShould it be the CEO, operations, IT, finance, or the frontline team that actually feels the pain?

Delia HallbergThe CEO. A very straightforward answer. No, really, it is the CEO because we're talking about SMEs. These are organizations with 20 to 250 employees. That's definitely the CEO. For major AI transformation processes, I am not talking about some minor side process that we're automating. I'm talking about the

Delia Hallbergbig transformation process. It touches processes, roles, budgets, and sometimes even identity. It can be very fundamental. So this has to be owned by the CEO, and really our best projects have strong ownership among C-level managers. But the frontline team obviously has to be in the room, and they have to participate in it.

Delia HallbergThey are the ones who know how the work actually happens. So we always bring the team together, and as I said, the best projects combine strong CEO ownership of decisions with frontline-team ownership of feedback. That's the combination that we need for successful projects

Jörn "Joe" MenningerThe CEO should own the project, but how centralized should AI decision-making be in a company, let's say a 200-person Mittelstand startup company before it becomes a bottleneck?

Delia HallbergSo I think you have two types. You have these major transformation projects. Obviously that's the one that I just elaborated, but you also have a different type of AI adoption that is, for example, using agents, using models to automate small parts of your daily work. Here, I would say strategy has to be centralized, but usage should be decentralized because the C-level decides where to

Delia Hallberginvest and sets the data policy and governance. It is very important to have guidelines, but I would leave the daily AI usage within these guardrails. This should be as decentralized as possible. Set the guardrails, then let the people run

Jörn "Joe" MenningerHopefully you did not forget to set a very important guardrail because it's always easy to say, "Ah, let's do this and this and this." But asking whether you have thought about every possible scenario, is challenging, especially when you have creative interns. What is one operational belief that you had at the start of Amadeni that a real client project has since disproven?

Delia HallbergThat's a good question. So I think we started with a belief that there would be a lot of nice-to-have projects. What I just mentioned, small side process that's been inconvenient, let's put it that way, and that automating them would make good projects. That's something that we've learned is not the case. The best projects are the ones where people genuinely feel the pain,

Delia Hallbergwhere we do not have to convince a team that digitalization is important. If I have to start this conversation, it's already a red flag and I totally accept that they have to go through a few learning curves. The best projects are the ones where the pain is real. The team has felt it. They have already looked for alternatives, haven't been able to find

Delia Hallbergit and when they find us, it's a match made in heaven. That is something we learned.

Jörn "Joe" MenningerGreat. Now let us close with the two typical questions here at Startuprad.io to learn a little bit about you and your company. Question number one, of course, are you open to talking to new investors?

Delia HallbergFor Amadeni AI itself, we are not raising. We are intentionally bootstrapped. What happens in our projects is that solutions sometimes emerge that have the potential to become standalone products. If it is a promising market, a promising product, then we spin those out. And for those, we are open to talking to investors.

Delia HallbergIf somebody finds our philosophy interesting, it's definitely worth reaching out and becoming part of our network

Jörn "Joe" MenningerSpeaking of networks and people. Are you looking for talented people right now?

Delia HallbergWith a large set of AI agents doing a lot of work, we are reaching a natural bottleneck, I would say. We are currently in a selective growth phase. We are looking for consultants with great personalities, a deep understanding of the SMB landscape, sound project and product-management skills, and a solid technical understanding. Yes.

Jörn "Joe" MenningerThank you very much, Delia. It was a pleasure talking to you in two interviews. Actually, as we can tell our audience, we recorded four pieces. Last time, you had a bad cough, and we decided to completely redo everything. We have now effectively made four episodes. So thank you very much for sticking with me. To everyone watching on video, I am sorry, it's over 32

Jörn "Joe" Menningerdegrees Celsius here in my room and the lights don't help. But for our audience, if an SME owner remembers one decision rule from this conversation, what should they ask before spending money on AI? Should they ask, can we automate this, or do we understand how this work should actually function?

Delia HallbergWell, I would always choose the second. "Can we automate this?" is a technology question. And the answer is almost always yes. You can automate almost anything today. But the real question is: do we understand how this works? It is process first. Do we understand how this work should actually function?

Delia HallbergThat's the important question. Can we automate this? Well, it's almost always a yes, especially nowadays. But don't use process first as an excuse to wait. Start today. Look at where your team loses time every single day. That's the money you're leaving on the table.

Delia HallbergThe companies that are looking at their processes and automating now are pulling ahead. In today's competitive landscape, most SMBs can't afford to stand still

Jörn "Joe" MenningerGreat. Last word: Most SMEs, most startups, most companies can't afford to stand still. Thank you very much

Delia HallbergThank you

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. E 782 — Climate Tech Germany: The €5–50M Hardware Gap

    Nov 19, 202657 minAudio

    Climate tech hardware in Germany rarely fails on technology. It fails between pilot and plant, in a €5 million to €50 million gap that is too big for angels and too risky for most funds. Business angel Birte...

    Credit the people on this episode

  2. Helsing Clarity: How AI Agents Fuse Battlefield Data | E787

    Nov 19, 202619 minAudio

    Helsing Clarity: How AI Agents Fuse Battlefield Data Helsing's Clarity runs hundreds of specialised AI agents over satellite imagery, drone video, signals and written reports. They share one military data model and one...

    Credit the people on this episode

  3. ICEYE, Mistral and Europe’s Deep Tech Scale-Up Test | E785

    Nov 5, 202631 minAudio

    ICEYE already delivers sovereign satellite systems for seven European governments, and the Scaleup Europe Fund co-led its €1 billion financing round. Mistral AI raised €3 billion. Jörn “Joe” Menninger asks whether...

    Credit the people on this episode

  4. E 784 — Physical AI in Germany: NVIDIA, RobCo, Agile Robots

    Oct 30, 202626 minAudio

    Germany chose physical AI: NVIDIA plans stakes in Agile Robots, Avelios, Mirelo and Reflex Aerospace; RobCo passes $1B; Zurich's Enact raises a $30M pre-seed. A note on timing: because of my travel schedule, this...

    Credit the people on this episode

  5. E 783 — Laser Fusion Spin-Off: How TU Darmstadt Built It

    Oct 22, 20261 hr 11 minAudio

    E 783 - Laser Fusion Spin-Off: How TU Darmstadt Built It Laser fusion is the hardest deep tech bet in Germany: huge capital, more research, a plant targeted for the mid-2030s. Focused Energy took it with a $240M Series...

    Credit the people on this episode