Participants: Henrik, Malte, Benedetto
Henrik: Hello boys. We are back.
Malte: Finally.
Henrik: Yeah. So, the long awaited, much requested season 2 of What's Going Down is starting up now. Very just a few months into the new year, but we're as ready as ever. We are looking forward to I think this season get some more interesting guests on for the first episode here. We'll we'll just have a quick discussion here in particular around what a lot of people are talking about of course being an AI company for six years. It's interesting to see everything that's happening around AI agents vibe coding and all of that. So, we're just going to have a small discussion around where we see that in the the field of energy management. But first of all, maybe just a little update on on what's been going on on our side. And I think one of the reasons why the podcasts have been have been delayed a little bit is that things have been a little bit crazy in Ento land in the past six months. We've basically doubled the business as well. So that is of course a good thing but it also means there's a lot of lot of work but fun fun things going on. Maybe first uh Malte just if you have some insights on maybe the highlights of what's changed and what are the things that we've been working on that would be nice.
Malte: Do you want me to make the excuse of why we haven't recorded any podcasts?
Henrik: Yeah. No. Yeah. It has been crazy.
Malte: Yeah it really has been crazy. Uh yeah. So so the this doubling of our of our company in terms of of clients or and the impact that we can bring to these clients has come from of course our core market being in Denmark but also some really large clients in the UK and in Italy. So if I just start outside of Denmark, I think that I don't know which of the markets that is most exciting because they are all exciting in different ways. But in the UK, we are now working with a very large supermarket chain that is just using a lot of energy and are really receiving really good results on our platform. So that of course adds to our growth when hundreds of supermarkets join join our platform. In Italy, it is a mix of many clients, but lately also a multinational retailer that has joined. Starting out with the Italian source, but there's also hundreds.
So, so that is a super good client there. And then in Denmark, the growth has come from commercial real estate. We have added like eight new municipalities, I think, as well. But that is kind of growth that has come that has kind of just existed throughout the past many years is that we are serving a lot of the public sector in Denmark and the retailers but the new segment that has been growing the most is is within these commercial real estate companies so investors and they are both serviced directly so they're using themselves or through asset managers where we have gotten a really good collaboration up and running with both Newsec and Cobblestone as being the kind of the major asset managers that we have started working with and they're using our product on behalf of their end clients. Um and maybe that's also—
Henrik: Yeah, that's a good rundown. Maybe it's also a teaser for next week's episode, one of the ones you mentioned there.
Malte: Yeah. Yeah. So, so next week, well, depending on when this go live and the next episode go live, but in the real world, we are inviting Anes from Cobblestone. Anes is leading their sustainability team and has been a really good client to get in on our end because he's challenging us quite a lot on the product that we deliver and the service we deliver around it. So, we said, "Anes, why don't you join us our podcast?" And he said, "Yes." So, so next time the next podcast will be with him and we'll talk about how they are creating value with energy management as an asset manager. So, super relevant for this growing customer segment that we that we have.
Henrik: Yeah. And speaking of the product, I guess lots of stuff also happening there. And maybe Benedetto, you have a little bit of an overview of what's been going on.
Benedetto: Yeah. Yeah, definitely. So I I think like I like this idea of the flywheel, right? So from the commercial side, the more customers we close, the more references we get, the easier it is to close the following customers. But I believe that on the product side we are almost seeing the same in the sense that when you start a new company or a startup right you begin with a product and you if unless you are directly coming from a specific sector you have almost no idea what will work and what your what is the actual needs of your customers right but then the more you start closing and working with the more feedback you get on your product and the more you like start following those tribes. And so if you imagine like you know kind of like in in video games you have a map and it's almost completely like gray or covered at the beginning and then gradually as you start working with more of these customers you discover more of this map which is like what how does the product need to look like to bring value in the real world and I really like that these large customers that we are starting to work with now they come with strong opinions they come with a lot of feedback they come with a heavy usage of the platform and that's like the best that we could ask. We would almost want to service them for free just for the feedback that they gave on them.
Malte: They can't say that. Can't say that.
Henrik: We erase it from the record.
Benedetto: Yeah. No, it's fine. Yeah. Like if like closing one of these and then learning from them helps me close 10 more later because they just have the same needs or very similar, then it's just so valuable to start working with more of these. And that's like a super exciting working position on the product to be developing close to them.
Malte: I think it has been a strategy from the get-go is that there's no way that you know when Henrik Casper and I when we started the company there was no way that we could find out which product to serve. We had to really ask our clients how can we develop the right solutions for you. So we have just been very centric on their needs and these clients have some different needs. So, so we've from the start built up this community where we're both of course having check-ins with our customers about these things but also meeting in person and and discussing okay so if all of you are following the SBTI or the reporting with GRESB or however whatever moves their business how can we support that so so it's not so different I would say from a conceptual standpoint, but it but their needs are just different.
Henrik: Yeah. And I think it's important from a business perspective to just this notion of compounding, right? A lot of people are talking about that in in the sense of financial compounding of revenue or whatever it is. But I think it it applies just as well to the number of customers you have and can learn from. you can learn, you can make the product uh exponentially better, you know, as you add more customers because you get more feedback and you get deeper into what product is actually useful and delivers value. Okay. So, so I guess on that note, you know, I think one of the topics that of course we're hearing a lot about a lot of people are talking about in this in this sector is, you know, this notion of we have now all these vibe coding tools, we have all these AI agents. So, one of the questions that I want to bring up here, could you in principle vibe code an energy management system? I think that's a really good question that all our customers and we should be asking right so what are the how are we seeing this from the different angles
Benedetto: yeah I think so I would separate maybe two things and one is like okay vibe coding and the other is like AI assisted coding with tools like cursor and claude code and all this and I think they both have like very different but relevant implications and so to answer the first question which is can someone vibe code an energy management system. I think you can to some extent, but I don't think that when someone is paying for a system like Ento is paying like for the initial development that that you're doing, which okay, you can get to from I don't know zero to 0.9 quite quick in a weekend or whatever. Uh but then you're basically paying for maintenance and for having like a group of specialists or very smart people thinking really hard about these problems every day. And so you really don't want to do that. And so like I I was also like listening to some podcasts like from like people from Anthropic for example and they were like okay like we are Anthropic but we you know we are not coding I don't know Slack from scratch like we we are the we have access to the top models but we still pay a service for like for messaging because we don't want to invest oursel like the time to develop something that someone else is already like doing maintenance. They've been thought about. They've been thinking about this for a very long time and there is really no point in like losing track of your the mission of your own company to do something that is on the side. And then when it comes with yeah about like AI assisted coding, I believe it's just allowing us to service our customer so much better and so much faster to the extent they like they can't even believe right the speed to which we develop the new stuff. Maybe Malte, you've been like working at one of these large customers lately and you can like comment on how that's been going.
Malte: Yeah. So we I think we see this in two ways. This whole cycle of development that is just being short. One example of this is when we do physical workshops with the clients and then we have a so I dedicate a full day typically with with Mads or one of the other colleagues Maria has also done some customer workshops full day workshops but there we let the team know that hey I'm at this client today when I have if there are certain things please prioritize them so we put them in the in our chat and then they just get developed and by the end of the day even rather complex features can be pushed or during the day. So that like intraday development has been super fun. I think the and valuable for the clients of course because it's the things that they need and I think the extreme of that was at some point I was in a meeting with I believe Henrik and and Roberto Kiosa and he live coded something. He was he all of a sudden just went silent and I thought he was just taking notes but then he refreshed the page and then the feedback from the client in this product was live and he could of course only do that because he was showing it local but still this development speed has just gone nuts and that is basically tying into your first point when saying should the clients then do this themselves or do we want do you want to have an external company deliver a product that is then just exponentially developing faster and faster. And I'm in the same boat here. I don't see our energy management teams or core user doing development on software. One of the key values we bring in is that they don't have to manage the metering infrastructure anymore. So uh pretty recent client we got he said well now I can actually do energy management because we are connecting to the main meters from the utility before he he was actually servicing the gateways that would pick up the signals from the meters and he always had problems doing that. So he spent 80% of his time collecting metering data and just by automating that he was a able to actually do something. it would be a step back for him if he should manage any kind of software application and run that. I think it it's just not his competence. So, so, so fully agree. We need in the future, in my opinion, at least in the immediate future, is that we're just having software services that are the line between services and software is probably getting smaller because you create software to fit a certain need that would almost be a service before, but but not in a way that you should vibe code anything from scratch.
Henrik: Yeah. Okay. I think it's also it's so one of the things is just talking about when we and many others can create features faster or iterate on things much faster. It also has the danger of producing a lot of slop, right? AI slop. So just the bad code feature slop.
Malte: Yeah. Too many features. Yeah. And bad code that'll just haunt you later, right?
Henrik: So I think this is one of the main differences between vibe coding something and then developing something quickly with the use of AI but within a very structured platform right so that that's the only reason we can do that is because everything is set up and there are the right pieces in the right places and then on top of that of course you can do things quickly and in particular if it's creating some kind of UI or you know adding a new endpoint or whatever it might be. So I think this is probably speaks to the main value of existing SaaS software and solutions in general is you have this data foundation probably and not just the data but also the platform foundation of how do all the pieces fit together and on top of that it you can really create a lot of things maybe much faster, right? And so it opens up the question to how should we uh as a company also enable maybe creating more on the customer side. Uh and I think those are some of the product discussions we're having and obviously working with as well and there I just think it's important that we keep in mind what is actually valuable and what is just hype right and so—
Malte: Where do you see that going Henrik?
Henrik: What do you mean?
Malte: Well, like how are we enabling that?
Henrik: Yeah. So, I think an obvious learning that we're seeing is okay, we have an API within Ento. It's never been the first class citizen. It's always that we've catered in our product more to the end user that's, you know, clicking around, but we're seeing the API become more and more important, of course, because people want to integrate. And then you know on top of an API you can have an MCP that then agents can also use and we're starting to get more and more requests also for just interacting with from Claude or from somewhere else with those APIs and I think it makes a lot of sense like we all know our mission is to accelerate these energy optimization processes right and if we can see that our customers and our partners can just have a better and faster process because they can just integrate to an infrastructure instead of necessarily ly clicking around everywhere then of course it makes a lot of sense
Benedetto: but also maybe to the extent to which to which like you decide to implement this AI chat features within the product is also like a key point that needs like a lot of reflection because to some level it can it's like a spectrum right and then on on the one side I think you have like okay I would like to do some creative things that it's difficult to have like within a few clicks because my use case is very complex and I want to have like maybe I want to create some reports and then in within to create these reports I do not want something boiler plate but I want to be able to do this like kind of iteratively and creatively with this assistant right but then on the other end of the spectrum I think you have a product that is not usable enough and so it's like okay potentially we could service this need of yours which is pretty standard just with a few clicks and a few defined workflows which is really they really used a lot right in the in energy work. But then some products we see going like okay even though you could only do that with a few clicks we still implement the chat way in order because our product is not usable enough uh and then like people can't find out which three clicks they need to do and then we just we are lazy and we implement a side chat so like that that people can we'll just do things like in this maybe more complex way always asking about things while having something that is maybe more essential more clear, more minimal and avoids like the general experience of talking to a chat to do something standard which in general I don't like when I use other products right I want to be able like if I know that I need to do something and that thing is not created I don't want to talk to an agent I want to just do it so I think we need to think really well and that's also I think we've been taking this approach of the API right which is like okay specific things that you can interact with but not make the whole product like on chat which is probably not the best user interface for this type of work.
Henrik: Yeah, I think it's a really important point like our customers expect that when they want to get numbers they get the right numbers. They expect that when they get this type of analysis that it's done the same way as yesterday and the week before. Right? So that type of thing is really important to keep remembering. You cannot just put everything into a you know uh natural language and then have an agent decide how things are done and then taking the as you say creative or maybe more you know analyzing a situation like a human would can sometimes be done in this way. So we have our root cause analysis which is very much built into the to exactly where the customers are working on fixing an issue that is using these LLMs right to to generate something useful the next steps to check and I think that's one way to to think about it another way is if we need to generate reports that have some kind of text or description or commentary on what's going on then that's as you say a creative thing that can can be added in this way but the overall analysis should be the same next week right so so then you need the fixed workflow so I think really this in an enterprise setting this balance between fixed workflow and creative or you know nondeterministic agents are really the important part to to get right
Malte: Henrik for the listeners that may not have a PhD in machine learning the nondeterministic can you just folds that out a little bit and
Henrik: yeah sure. So, so it's so so when we say workflows where that is done the same way every time then those are deterministic based on some data comes in and if the same data comes in
Malte: yeah exactly the same the same results will come out again
Henrik: within that there can be some flexibility of course like a machine learning model is is can sometimes change the result but you should be able to reproduce it and then you know on the other side of that is nondeterministic that is you give something to an AI, you don't actually know what comes out on the other side and you just have to keep improving your prompt so that it becomes closer and closer or you should actually use it for what it's good for which is more of the creative stuff.
Malte: Okay. So in a real life setting and what you're saying is that for instance in energy engineer you have you've you've bought a a very expensive new lighting system for your for your logistics warehouse and you've entered a performance contract with the supplier. So here you're doing a IPMVP, you're doing a a performance-based verification of the savings and there you want to have a result that is robust and you can basically recreate in in any scenario recreate with the variables that you've agreed upon in the contract of of this M&V. Correct.
Henrik: Yeah, exactly. And I think that is something you always want the calculations to be this at least the calculations should be the same and then the results will depend on the data right and then maybe the analogy there or the the additional thing that you could then add with an LLM which is also something we've been working on or you in particular Benedetto is then based on those results what are the things that we would like to highlight or comment or flag as notes
Benedetto: That is a perfect a perfect task for an AI to do because it's much more like taking many different data sources, maybe looking up some building information and then giving that comment, right? And that's something that, you know, an energy engineer would would sit and do and they probably would as well afterwards, but they just may not have to fill out 80% of the comments that can be done automatically. So it's the interpretation part of the so you have the results the statistical results that you can replicate and then afterwards you may create a full report which is an interpretation of that and if you gave that to you know 20 different energy engineers you would also give get 20 different results because there's human interpretation but there also machine interpretation that can differ.
And then that's also maybe where the experience of the user also comes in, right? So I think it's like okay, we have this AI tools is get they're getting the models are getting more and more capable. They're getting to PhD level or whatever. But I believe like whenever we use them, we should always make sure that we would be able to get to the same conclusion by ourselves and without using the tool. Like it's okay if I have so many data points that it's not humanly possible to evaluate them by hand. And so I have a machine do it and then I can interpret and validate the result. But if I start to like if I'm doing a completely different job and then at some point I get to Ento and I and I ask Ento like okay what is the top things that I should look at in the portfolio but then I get a result that I cannot understand and I would not have been able to get to the same conclusion by myself. I believe the result is useless because I cannot validate it. I cannot understand if it's actually good enough or not if I should act on it. So this maybe also goes into the direction I don't saying like will AI take all of our jobs like I don't think so like it's getting us it's enhancing our skills but we still need to like be able to do the things ourselves and so like there is more and more need of experienced professionals to then use the tools more critical thinking less manual work.
Henrik: Yeah.
Malte: Yeah. Great. Well, I think I mean I don't know if it's a surprise that our conclusion will be that you cannot vibe code management. At least it'll take a long time. Like I'm just seeing at some of the things we're working on every day on analyzing this and that data or managing the data that is coming in weird formats and changing resolutions and all those kinds of things. So there's years of work in just managing that. So, so that's just one practical part. And then some of the things that we've discussed here are of course important to to consider. Anything else on on that before we wrap up?
Benedetto: Maybe like a good analogy I was thinking of is like okay if you if the vibe coding now can help you do like 80 or 90% of an energy management system. It's still like you know the last 10% is the one that requires the most effort and the only one that matters because like getting to 80% is useless. If you then you can get your last 10% which is like the most difficult part and the one that matters and that's why like there are companies that specialize on it and across really any domain and I think we will see more and more of of this instead of just like anyone vibe coding their own tools to do stuff.
Malte: It's the same analogy as you gave before and right you have very large very good companies that are specialized in software development. They're not inventing all their tools themselves. we are using off-the-shelf tools. So, so the idea of of inventing and running the infrastructure of a product which you would could pay, you know, half a full-time employee salary or two full-time employee salaries for just doesn't make sense if you need five people to run it or one even, right? So, agree on those notes.
Henrik: Great. Well, yeah, we're uh I think we're at the end of the discussion here and I think we can just say we're looking forward to the to next week and the coming weeks. So stay tuned and see you there.
.jpg)
.jpg)
