Francesco Marconi & Scott Austin: 2025 Year in Review, What Actually Changed in AI and Media

This is Newsroom RuleBots, the podcast where we explore the intersection of artificial intelligence and the news industry. I'm Nikita Roy, data scientist, media entrepreneur, and one of the many founders currently building their ventures at the Harvard Innovation Labs. On the Newsroom RuleBots, I'm excited to bring you insightful conversations with industry experts about how AI is impacting the way we do journalism. Today's episode is a little different. It's a roundtable conversation where we're doing an end-of-year recap on AI and media in 2025. I'm joined by Francesco Marconi, the CEO of Applied Excel, and Scott Austin, the head of business development at Symbolic.ai. The three of us have spent the past couple years working together on our upcoming book, The Science of First, coming out in 2026. So we've been tracking what's happening in AI and media up close, and in this conversation, we step back and ask what actually shifted in 2025, what trends are being overlooked, and what we think is coming next in 2026. Francesco, Scott, welcome to Newsroom Robots. Thank you. Thank you, Nikita. It's great to be here with this brain trust. Yeah, and back again with you, Francesco, I think we've been the three of us have been working over the past year on the book, The Science of First, together. And it's been a great journey putting all of those ideas, everything that we have together, seeing what the future of news is going. But I wanted us to come together as this, one of the, some of the people that I think about where the future of AI and news is headed towards, and sort of recap what the year 2025 has been like for AI and news, and going into 2026, what are the key things that should be top of mind. I feel like 2023 was the year of the generative AI shock in the industry. We were just grappling with what this means, the capability of the machine to be able to instantly generate text, and then 2024 was a lot of tentative experimentation that happened. But 2025 I felt like is will be recorded as part of in history as a sort of operational reckoning moment for the industry because our business models have started to get challenged. The way in which technology companies are using news data is changing, the way in which workflows are evolving, everything is changing. And so I think with someone who's been keeping up with AI, I feel like everything is moving so quickly as well. And so I wanted to recap what has 2025 been like? What has been top of mind for you, Francesco? Being talking about the, you know, the future of news and the future of AI, but we should be talking about the present, particularly in 2025, I feel there was a major inflection point in the adoption of these tools and technologies in many newsrooms and actually becoming a reality. So from what you were describing 2023, seeing the shock of this disruptive technology to 2025 coming to the realization that actually this is going to be a foundational system and it's going to be a completely different way of thinking about information. The same challenges, they'll, you know, Nikita, you've been talking about for the last several years of adoption and sort of concerns about accuracy are still there. But I think we've gone through an inflection point in terms of, you know, applications of AI that are not just recommendation, native AI applications from, you know, answering questions about the news to data analysis. I really think 2026 perhaps it's going to be less exciting, but it's going to be a period of, you know, full deployment or at least I hope it will be, you know, we'll see a lot of adoption. I think one of the key things you're talking about as well over there that we've had to deal with is the whole shift in the user experience behavior right from people just going to Google to Google itself now introducing an AI mode and AI overview is where you're seeing AI giving you answers and that's the next shift something that we have created our entire business models out of. A user behavior that I think people growing up in the 2000s have all been accustomed to that entire thing is shifting now. I was just going to say that the dichotomy is really, really interesting right now in newsrooms because on one hand, as you're saying, that's a reckoning and you are trying to figure out how to adapt your your business models in this new immediate answer world where it feels like your content is getting pilfered at all times. And you have to guard it and understand that your business model may not be reliant on on the traditional Google anymore at the same time you can't be a journalist anymore without embracing AI and using AI throughout your entire workflow. And so you're grappling with both of those that you really need to embrace AI in a big way but at the same time, worry about the perils of how you're going to adapt from a business standpoint. And that dichotomy is just really, really interesting. I think right now how companies are dealing with it. But yes, I think going forward we're going to be seeing far more AI co pilots. It is going to be a companion in the newsroom and you know, we're going to have to deal with that reckoning. Scott, don't you think, you know, part of the value of news and of journalists is to bring information to you as a as a consumer. And so in a way, there's these two kind of consumption modes. One is, as you're saying, you ask questions, you get an answer and, you know, all of these chat interfaces are doing a pretty good job at that. But the other one is identifying what is important, assigning sort of an editorial eristic on top of that and then pushing it to you. And so I think that's a part that of news that is often overlooked. I actually call it the fallacy of chat GPT meaning I've seen many, many experiments of, okay, you can now converse with the news coverage of, you know, a news organization. I don't think that's the natural consumption mode for news. Of course, you can, you know, you can go and ask what's happening in your space or in your area. But the other part of information meets where you are that's that's not, you know, been discussed that that much at least from my perspective. Yeah, I don't know that just from talking with a lot of publishing executives, you know, that they are still behind in understanding the ramifications for the idea that people can get information instantly and how it's interpreted. And where it's coming from the sourcing. I mean, there's worry about it, but I think it's happening so fast that we're still playing catch up in understanding what the consequences can be in consumers getting information from all different places. And seemingly instantly we can talk about later, but there's, you know, the whole misinformation and disinformation, but there's also the misinterpretation and that really worries me in spitting out information that may not be exactly right. It's not that they're 100% obvious factual incorrect information, but there's nuance and there's semantics and there's all sorts of different ways that information can be interpreted differently. So I'd love to talk about that. Yeah, I'm excited to talk about fact checking as well. Another area that hopefully we get to talk to is that I'm seeing the emergence of new types of AI companies that are sort of, I don't have a proper terminology for it, but it's essentially proactive AI. And so an example of that is company called parallel, which was launched by the former CEO of Twitter. There's others like Cavali and and so on and essentially using AI to be able to monitor different domains. And so that's, I think the chat question and answer interfaces are well understood or at least better understood. I'm not sure about when you see systems that can proactively look for information on your behalf, like I do a lot of the reporting. What is that going to how is that going to change, particularly the news gathering side, not necessarily the distribution, but the news gathering side. I also want to bring up the proactive AI. We often tend to think of just the chat answer model as a user shift where people are going to interact with it through chat GPT. I completely agree. I don't think that's the future of the UX of news, but one of the big experiments that's currently only available in chat GPT's pro version if you pay for the $200 is chat GPT pulse, which essentially is proactive AI layer on top of to give me the top stories that I need to follow about the day. And I have been using this for the past three months since it's been launched. And every day when I go into my chat GPT, it gives me this highlight, check out your pulse for the day. And it is the most personalized practical news information that I need from my day. And why is that, first of all, because chat GPT has, it has all these conversations about me. It kind of knows this sort of location where I live as well, because I ask it about restaurants near me. So it's telling me that the street car is down and this is happening. And I'm based currently in Toronto. And what's even more concerning is that most Canadian news publishers, all the major ones are suing open AI. So there are data does not show up on chat GPT pulse. It's telling publicly available information, all kinds of new sources and blogs and information from outside of Canada, new sources from outside of Canada, and like Reuters and AP that they have partnership deals with to surface that. And it's been helping me keep track of even things that are happening in AI and journalism that I might have missed from other sources, because it knows exactly what I need to know for my day. It knows exactly what I need to know about. I'm traveling. It knows that I'm going to be traveling to Norway. And it's like, hey, you might find face delays at the border because of these new rules that have come out. I would never have gotten that. And that's coming to the next level of distribution where it's proactive. It's personalized. And now it's a system that knows everything about me more than any newsroom has. If you're a news organization, how do you compete with that? It was always the idea that it's hard to compete with Google because they do have all of your data. Now, when you're talking about these AI companies, it's even Google's one of them. It's even more because you're talking about conversations now that you're having and your most secrets and questions and your day to day really. Once you get ahold of that, a news organization is going to be very difficult unless they have to pay for this data. But, you know, you're talking about a whole new era of personalization and also velocity at which they can give you anything instantly that's dedicated to you. And news organizations, you know, it takes three hours to create a newsletter that's just general, right? And what you're talking about is milliseconds. That's completely personal. That's just what you want. At the same time, you do need the source material, meaning correct. You need the news to be generated. And, you know, of course, it can source from blogs and government websites, but it relies heavily on the supply chain of news organizations. And so, I agree, in a way, it's Google news on steroids or even the Google alerts on steroids, but I do think there's part of that process that although pulse is a great product that it's not able to capture. So, and that gets to the core of journalism and let's assume that your reporting a story where all the information is available online somehow and he then obscure databases. And I'm not even talking about actually talking to people interviewing, which I can do to some extent, but not to the extent of a journalist. But that part of the workflow, I think it's currently untouched by the big AI companies. And, you know, if I were running a huge news organization, the first thing that I would do is to try to understand the allocation of resources in kind of the production and distribution. So, you know, I'm going to go to the local end, which is being completely commoditized, disrupted by these systems and try to understand, you know, how much of overall resources is being deployed to data gathering. And then what I would do, likely what you'll see is that most news organizations spend a lot of money on the packaging and then the distribution and newsletters and curation and so on. These companies need to be data gathering companies in a very specific way, which is even if the information is available in blogs or government websites, there's so much work that needs to be done to properly organize structure contextualize that information. And that becomes something that is unique to news like the AI companies are not necessarily interested in that because their business model is about scale, it's about as many agnostic use cases as possible. So, I don't think they'll, you know, they'll become specialists in covering, I don't know, elections or campaign financing, but I really think that's that's going to be the biggest opportunity for news, which is to becoming the sources of curated data. That's why I say like this has to be a first principles reckoning for the news industry. And I think very often, as you just said, it's so much on the, it's such a production heavy workflow. And now any TikTok creator can sit and spin up almost a well-class studio in their bedrooms and have access to all of these content level softwares and how AI is enabling production itself. It's also you can have a scriptwriter, an export scriptwriter to be there for you. But the biggest strain that AI is putting on just all of these different workflows is its economic value, the unit of producing content starts going towards zero. And I think there's a diverse, I think there's, it's not just a yes or no, like we can only do one thing. I think there's going to be more value for certain niche content, human authored content where people can create community, but then I completely agree with you, we move our values upstream. And where what is the value of the news organization lie, and it is in that information gathering, it's about making sense of that information that we are at the end of the day, the only industry that I say is a real time structured knowledge network of verified information. Nobody else is, that's the four principles of journalism, that's what we're in. Yeah, and you know, on then note, if you think about it journalism or news organizations sell two things, they sell solving the problem of scarcity of information, which to your point now it's being commoditized maybe on niche domains or like trade publications that still protected. And the other thing that they sell is a filter of the world, right, so you, you subscribe to the New York Times or the most return or the Washington Post, because you want to see the world process through, you know, an editorial system that is a particular point of view. Now that filter can be hyper personalized filter as you described on your chat GPT pulse experience. So I think the reimagination of scarcity and filtering, I think it's going to be a key to be able to survive and thrive in the in this new AI age. So scarcity has to do with how do you find the most unique contextualizes sources of information that an AI can touch that could be offline, that could be, you know, private data, and then the filtering, how do you sell these the editorial rubrics, the point of view in the world of perspective, not to humans, but to other machines to use those are my two bets for how to build an AI first news organization. Yeah, I was just going to say those are great points that I think that's really exciting for the future of news organizations if we can do it right and we can figure out the business models is that idea that real journalism matters and interpretation matters. And it's not going to be about what I was when I was in the newsroom for many years, it seemed to be come about commoditized information who could get that same story out first or maybe twisted in a different way and there's a hundred of the same stories. I'm excited about a future where we can talk about bringing news out that unearthing news for the first time and really interpreting the news in a different way based on human experience, you know, the reality is a lot of people don't really realize this, but journalism is hard. There's a lot of nuance and words in semantics, there's negotiation in wording because words of meaning and meeting meaning can be interpreted differently, there's controlling of information out of sources, there's the interpretation of data that requires the human experience. And so I think AI can help us with a lot, but it has to be a symbiotic relationship, which I know book really talks a lot about the idea that it's a copilot that it really has to work together for newsrooms really to emerge a campy one or the other. Ian Scott, in one of the passages of the book, we have, you know, a segment that describes what you were just talking about, which is the hidden process of news, when we were all talking about this, I felt that that's often underappreciated and not really understood and, you know, you being a journalist in an editor for many years at a top news organization. And I think that appreciation is you really understanding and care about it, and I just loved how you, you know, elaborated on what's going on before, you know, the Wall Street Journal publishes a story. There's so many things, right, so it's not just getting a tip, it's you were talking about, you negotiate with your editor, you engage with the sources, you try to get new sources. And so I'd love to, for you to tell us a little more about that kind of perspective, because I think it's by having this understanding of the hidden processes of news, we can really understand what AI will not be able to touch. So I'd love to just hear your thoughts. It's always great to hear the human perspective of news. Yeah, I think in the early days of the book, you and I debated a little bit, because I felt like it was maybe a little bit tilted towards the automation of a lot of the processes, which I think again is very important when we talk about, hooking up to data sources and interpreting almost like what meteorology does now. Human can't do as fast as a computer can, obviously. And so there's going to be so many great ways that we can apply AI and just sophisticated technology to news. But yeah, when I think back to a lot of the investigations we did at the Wall Street Journal, that might began with a tip, one of our reporters at the time, Rolf Winkler works at the Wall Street Journal, got a tip about a company called Outcome Health in Chicago several years ago, and it was the darling of Chicago startup scene in Chicago. It was suddenly had a multi-billion dollar valuation. It had the Priskert family, the famous Priskert family, was an investor in the company. And he got a tip that there was something to miss inside the company. This is not a publicly available data, but this is information that somebody has that works closely at the company and something was was a miss with the sales or the sales tactics and the accounting. It was one of these things that took months, but he was pulling one string after the next talking to a new source and getting some sort of internal information. And even when we felt like after 50 sources that we had real evidence that there was actually fraud going on and since then the CEO was put in prison. The idea that any sort of AI system could help with this. Now, it could be a companion if we had a lot of data and we needed to synthesize this or detect whether something was real or not, maybe we could use technology for it. And once we had started piecing together the story, then you're talking with the lawyer that's representing them and the editors that are involved with it, it becomes 20 people involved instead of just you and that reporter. And there's no technology, there's no AI that could help us move faster because every word that we put on that page mattered and each word had meaning. And so I really guard against sometimes when I read or hear about how AI is going to change the newsroom. I want to talk about also the process that goes into journalism, the Chris Whipple interview of Trump's chief of staff Susie Wiles the other day right that came out in vanity fair. He spent what was it? It was something like 11 interviews with her. She never does any public interviews and that became the story she unearthed quotes and anecdotes that became the news. So there was no way to really predict that. Yeah, those 11 interviews are kind of a reflection of the things that AI doesn't have. So, and I think the most important thing is that AI does not have the human persistence of doing the 11 interviews or to investigate something further. It can do a lot of the research, but on information that already that is available. And so, yeah, and I don't want to sound like a cliche of like the automation versus augmentation. I've started thinking more about the speed versus persistence as it relates to the relationship with between AI and human journalists. This debate that you are talking about is something that I see everywhere when we start to talk about the future of the newsroom. Often times it does seem like it's looking like this completely automated future and that's also not at all what I support or I hope because that's the value of journalism as you said is. That was asking the right questions, keeping on finding and then having that human instinct and judgment. But what all of this is pointing towards is we also have to reimagine what our workflows might possibly look like in this era. When the computer came and before the and the typewriter like it's now that everybody was able to be able to connect and send something and they could type you do not require a type is to be able to use that you suddenly democratize that power. And I think that's something a very similar type of thing that's happening with this technology where you're now democratizing a very powerful technology, which means you have to reimagine what those workflows look like. Before you had computers and it was just a typewriter looked very different to now when you're having AI as this technology that can do a lot of things that some things were not even possible to even imagine that it could do so what does that workflow now look like for people. I think that you think there's a future just you know expanding on the workflow from the typewriter to the computer to online search and AI and so on. So like you become more and more efficient and sort of the nuance of being human becomes even more important for this workflow and we're talking about the things that AI cannot do in terms of the persistence and the investigation. It's finding kind of internal tips that are not publicly available, but do you think there's a future where everything becomes digitized right now we are having this conversation it's being digitized when you text someone you have those signals when you when your car drives by a sensor in your license plate is being read so just extrapolating that into the future. There's sort of this massive availability of data even in that future I think humans will will be critical and so the workflow is not just about how you work and systematize the research process I think it's going to be completely completely different going to be more like a control tower at airport where you have these highly specialized professionals. There are very well compensated because they are doing a critical thing they are controlling the flow of information in that case it's like very specific to you know to the airplanes taking off and and landing. But that's how I imagine the future of workflow that doesn't exist like I don't know if any organization even sort of the intelligence agencies have that type of capability. But for instance in Scott you live a lot more perspective you work with publishers every day you know just the fact that there's not a centralized hub for internal data that's you know that's a big a big challenge there's the human adaptation but there's also like the infrastructure the technology and so on so it's not really going to be a slow process that's right I you know the New Yorker just. Finally digitized their entire archive and several publications have done this and you know they talk about how for years and this is true and so many different big giant news organizations as you have a librarian and you ask a question and they go back through the archives it's almost like what we used to do with micro feature you know these days you're going back to the archives trying to find it and now when you start to build the architecture. This important architecture within your newsroom where everything is digitized and at your fingertips and you have not only your own content but you have the world's data your fingertips and now you're able to bring in that information that's when it gets really really interesting but we're not doing that today as a as a news industry and we can fall behind pretty quickly if we don't build that infrastructure. No I was just going to comment on on that Scott which is I worked in the same use room as you did and you know there's this the default state is you know we can build this internally and I was running the team designated to build some of these tools and by default you think like that but the reality is that there's institutional inertia for either human inertia or the systems or not. The systems are not in place or that is not in place that it's almost faster for and more reliable for these organizations to collaborate with technology companies and I know you have some experience and exposure to to that where it's not just about whether or not something can be built. We're telling us about some of the fact checking tools that you're developing it's not just the ability to develop it is having the kind of the bandwidth within a company to build it and I think it's the you know we both are building startups I came to the realization that the function of a startup at least in the information industry is not to innovate is to build technology and solutions faster than a company. It's faster than a company can can build and I'd love to hear your your perspective on on that and of course Nikita you're talking to journalists all day are they reluctant to use you know third party tools do they want to see them developed internally what are what are your thoughts on that. It's very interesting I was just doing a talk the other day and I did a quick survey to see people been using generative AI tools and it was around 38% had never used an AI tool before. Another stat that had shocked me it was group of fully career journalist and around 50% of that group had never said that they never used an AI tool and they were actually also very against AI they do not want to use AI so you have a big like this big divide happening on the ground I would say where newsroom executives have to deal with the AI because it is impacting their bottom line and they have already if they had ignored it for the past couple years 2025 was the year they couldn't ignore it because they started to see drop in traffic with Google AI overviews. But at the same time I would say the AI literacy and adoption is not scaling at that same level as at least experimentation you have a big divide between product teams and I would say still journalists in using AI or being exposed to AI and it comes rightly so because of a lot of the ways in which the technology company is the impact on environment which is really a big concern what's happening to your data and just it feels like it's AI is just being pushed onto you. Rather than I think like them co-developing it or building it and being a part of this it's something that it's just being pushed again and I feel like the industry is a bit bit tired with so much of change in new change coming forth. And do you think that lack of adoption the 38% or so is it for moral reasons that the issues with AI between AI companies and publishers but also do you sense that perhaps they are thinking if I use an AI tool to help me write is that is that not up to standards am I cheating? Do you sense that there could be an issue like the kind of the human blocker as opposed to the curiosity? That's exactly what I have noticed in my conversations with early career younger journalists because when they were back in journalism school and chat GPT came out that was the language that was being used right if you use chat GPT you would be penalized for it. So you come out worried about how to use you're not really knowing how to use and I think reducing AI responsibly is also a skill that needs to be taught. That's the first thing and the second thing is a lot of newsrooms still lack an AI policy those discussions are happening but they don't have a roadmap in terms of what am I allowed to do. I don't want to get fired or have my name out there is one of those examples of AI slop. I think that's such a great point having the guard rails to begin with but having a leader in the newsroom not just a publishing executive or somebody on the business side right it's got to be working with the editor in chief. And the news organizations that I've seen that are really transforming where if you took that poll probably in the newsroom. If not all of those journals would be using AI in some fashion it's integrated it's starting to become integrated they talk about the benefits are developing their own tools. I think you do see it in some smaller local news organizations others where maybe an editor has a distaste for it and so it doesn't flow down to those reporters and editors but I would say the reality is that we are shifting fast. And you're going to be hired and you're going to keep your job if you can work in tandem with AI and do not be scared of it. I've seen my conversations because I do at symbolic AI where I work we have a platform that helps with AI systems across the entire content workflow. You mentioned that question Francesco are they trying to develop their own or are they okay with using third party systems. I would say my conversations have shifted drastically in the past year where I think a lot of news organizations want to feel like they own and control everything. The reality is they don't have the sophistication the tech teams to develop a lot of this they might develop some tools some incremental tools but to really make a transformation in their newsroom you're going to need a major investment and it's most likely going to be a third party system that you can then develop on top of or work with versus trying to create everything by yourself. So right now there's some really interesting incremental tools and so newsrooms are adopting it slowly or you know slowly I would say it's fast in the news industry but it's probably too slow. But I would also argue that there's room for a huge transformation so we're seeing some of our customers who the CEO or the editor in chief have really bet on AI throughout the entire company. So they're putting a lot of investment in too. Yeah and even if you know they do have the resources and the expertise and in fact there's a lot of companies doing really interesting work like the New York Times, the Washington Post in AI. I think the adoption is is is also human factor and I couldn't agree more with you of like in it's to come from the editor in chief the CEO whoever to give clarity and they actually reminds me of this quote from the CEO of Axel Springer Matthias Duffner and he said something along the lines of you know you don't have to explain why you're using AI you only have to explain if you're not using AI. And you know I think that at the time it was kind of you know a bit thought provoking but I think it's you know it's the right way to go. I don't know even you're a reporter are you still like crafting your story in a gourmet way in like the old school manual way that that feels like when you have electricity you're still using a candle or something. That's saying like you know give a prompt to chat you can have it right the story they can do that. But things like you know spell check or alignment with grammar I just feel like there's things that are based level I'm not thinking about I'm not even talking about some of the other tools that are out there in terms of like fact checking and things like that. It just feels a human moral question that needs to be kind of understood. There are many tools that newsrooms are at least like even prototyping for themselves to change those workflows right creating like a GPT to act as a body as a body that can read through and give you suggestions maybe help you through different perspectives as you're doing it. The Philadelphia Inquirer which we just had on the podcast was talking about now because of the investment that they got from the LENFEST AI collaborative they had their first AI engineer part of their team who was then sitting and able to build things for them and then they have this new archive news archive that allows their journalists to use it to help them research. So you're seeing different ways in which workflows are starting to change but that all comes with that adoption. But I think this other point comes is very important. I agree that leaders have to be talking about AI but sometimes the way in which they lead like lead from the front and talk about AI also tends to because I am also very much on the ground listening and talking to journalists who are worried about what's happening. I see both sides of it I feel like and they are worried about what happens to their jobs and we as an industry are not talking enough about that and that's what causes a lot of fear. And I think that's why we need to talk about reimagining workflows thinking about the human value all of those things that we were talking about because then we are not just saying your job is to sit and AI is going to write all of the articles for the future and your job is just seeing what the AI is written is right or not nobody goes into journalism school thinking that they are going to be doing that and that's not what it is. And I think that's why we have to talk about what is the future workflow, how do we re-imagine that and then people are able to see them fit themselves there. And do you think the concern of journalists as it relates to their work is that their work is changing and they will have to do things that they then sign up for and perhaps they are not as engaged with doing certain types of tasks. The concern of job loss like are those pretty even or it's more the concern of economics intermediation of AI in newsrooms. I think it's mixed but definitely a lot more on the job loss. If you come to somebody and one of the most striking ones I remember from me was when I had gone to a journalism school and I was talking about how a lot of the newsrooms are having this headline writing bots and helping with all of those kinds of things and the student had asked me, we just spent like learning how to write headlines and working on that and so what are we going to be doing and that was an important question. We always talk about AI is the new intern. Well, what happens to the early career journalists who learn on the job to embed and learn this domain expertise. Well, they are now coming into a world where we're seeing AI does as good as a job. It's quick. It's there to have created all of these bots that have been trained on the or entire newsrooms voice so it can help you and write your headlines. Where do we see as the training grounds also for the new younger journalists. Yeah, and then it happens also with engineering where cursor, cloud code, all of these tools. It's beyond the intern. The way that I think about is I totally agree like a big concern is sort of the how do you get in the door. I think it has to do with being super charged with the knowledge of these tools like used to your say you're working at a news organization. You're you're hiring someone initially you want a great storyteller that knows how to write then evolved some you know in the 2010s or so in actually you need to be multi media. You need to know how to edit video and audio and take photos and so on and then eventually you know you also need to be savvy about data and there was maybe in the last five years. And I think the requirements are all of the above plus you need to know how to use AI tools. I think a lot of journalism schools at least the ones that I have exposure to are you know investing in kind of AI courses Nikita you're you have a front row seat into that. I think there's still a disconnect with some of these journalism schools that you know are AI forward and then you're going to a more traditional newsroom and there's kind of a shock Scott you're developing a lot of tools. There are thought to be journalist first what type of feedback are you getting from news organizations yeah I think when when I talk about you could use AI to help with your writing or with your research or your editing and you talking to those very general ways. I think people put them not only the the best detector but the skepticism comes out and they think about hallucinations and all this worry the reality is you know what I talk about is actually use cases and how you can use take that we use the word drudgery a lot you can take some of the drudgery out of journalism so you can focus on the real writing and the real reporting when you're putting together a newsletter no nobody likes to aggregate their own content that's just the reality when you're at the end of a story. You may not have the time to completely go through every single piece of your story it would be unfathomable right now I think for a news organization not to have some sort of like spell checker and style checker and all these things right those. And I think AI can help with a lot of those those pain points I think that one of the biggest features now symbolic that has really really resonated was this auditing fact checking feature because and that was really important to me if you're going to use AI I want to know if my story is checking out and even if I don't use AI can I use a robotic power to help with cross referencing my story with source material. And the wider web outside that and so that's been really really I think a blifting for people that are either using AI or not and they're able to use generative AI tools to backstop that fact check and go through every fact in that story and cross reference it find whether they're just a discrepancy. And it's really working you know we're using it to where it's very particular even you know things that you might have written that says you know Wednesday and said a Tuesday those are obvious but even claims claims that have a little bit of nuance attached to it. It will flag it and it's your job as a human to interpret whether well that's actually okay I'm okay with interpreting it this way and so it becomes a companion it doesn't become a replacement for that editing. And Derspiegel I'd featured them in a case study the German newsroom last year actually had built something like this for their newsroom internally. And that time it was a pilot and I was speaking to them recently and it's they have they put out so much of content right not just the stories but social media posts and all these kinds of things and they can always run it through this fact checker. And it's not just fact checking with other sources but also their own archive and data and now what you're essentially having is also building up the trust them out of corrections and it. It's able to detect a lot of things and I think that's one thing AI can help with a lot of the fact checking one thing that I often do is put things through before publishing put it through chat GPT's pro model and tell it to fact check if I've got in my data right. And it's it's able to find things and nuances where maybe it's you should frame it slightly differently what kinds of ways might be slightly misleading if you're saying all of those kinds of different ways in which wins now when we think about the different capabilities. It's not just a plane model of language generation but you're also having these reasoning models that can go and spend 20 to 30 minutes looking at every single claim that is being made go and check the internet sources for that and come back and give you a summary. Yeah AI fact checking used to be kind of the flying car of journalism but as opposed to flying cars AI fact checking actually exists and it's working and that's one of those things that it used to be perceived as like super futuristic or science fictiony. And now it's like so you know so useful in even these cases that you've that you've described so I wonder what else that we previously thought was you know not realistic will be able to see in the near future. Well I think a lot of the agenda work that's that's happening you know I think when we talk about agents it's probably and it's definitely a hyped and inflated term that means potentially something different that it's reasoning on your behalf and it's acting like what you would do we're not there yet but there are a lot of prompts and a lot of things that you can do now that help with extracting data from a 200 you know page legal filing answering the same seven questions that you're asking all the time it's some of these wrote tasks that. We can now do with 100% accuracy in some cases I think once reporters and editors see this they just wake up to the fact that it's yes of course I'm going to use this but it's the trust factor at the moment and that's where I think all of us Nikita you spend a lot of time with newsrooms and obviously I do with with symbolic is getting them to use it and ensuring that they can trust what they're using and and that's so important. AI agents are actually very so very specific problem which is a model running out of memory so so you're doing an investigation or analyzing a thousand rows of data if you try to do it at once and use chat GPT or any kind of more user facing tools you will run out of that chat will run out of memory their context because of the context windows. And so the agentic process is actually a manifestation of a workflow that journalists have which is you have a big task you have to investigate this this case or these finances or something and you're breaking down this massive task into sequential reasoning steps and so it's about being able to communicate what those reasoning steps are so for instance you can have an assessment agent is this data relevant for the type of analysis that I'm doing you could have contextualization agent how is that this data point how does it relate to other data points around it you could have enrichment agent that brings in additional sources of data all the way to writing the full story so from the news gathering the processing the fact checking the contextualization the enrichment rather than thinking about it doing it once or having a human you know directing okay now do this now I'm writing this other prompt you can pre define those those steps in a way say evolution of kind of the OG type of news automation with natural language generation which was kind of you know these madly templates now it's not about the templates of writing it's the templates of thinking and you know journalists in the book we call these news theorems where as an experienced journalist you have this intuition about what types of questions to ask what particular sentence in a document means what does a press release on a Friday night is trying to hide so all of these innate kind of realistic intuitions can now be encodings into into agent that can help you do a lot of this work faster and very reliably and I'd say one of the best examples of agent behavior that's out there is cloud code and has a terrible name because it says code and so people think that's all it does but I use it all the time for like just textual things that's my default interface a lot of time because in AI agent at the end of the day is basically any software that you give it a goal that you wanted to do it can perceive the context it can decide on actions it can go and around execute all of those autonomously and with cloud code specifically I give it access to an entire folder of the research that I'm doing work that I'm doing and I can tell it to create multiple sub agents each agent doing a different task putting on a different aspect or a theme with which it's taking an agent going and now because it cloud code also connects to the web going and doing a big deep research on a particular topic and creating that and then finally an agent that goes and synthesizes what all the other sub agents did in their own reports and then producing things for me and then so now I like that was the big unlock for me in figuring out code using it and the way in which it has just changed my workflows I don't have a back and forth conversation with chat GPT I'm sitting and deploying agents literally across the work that I'm doing it can I can have like 10 different documents for 10 different purposes that I'm working on ready around at the same time all working on the same context and that's that true agenda behavior there do you find what the agent behavior do I mean is it you know I think of again I've looked at agents that I think you mentioned at Francesco that are that are like workflow prompts right we we set the prompt very deeply on this task and it performs that task and I think that's super help and then there's that next step which is it starts to reason it starts to really contextualize and go above and beyond what you first set it out to do where do you think we are Nikita right now since you've experimented with a lot of these agents house sophisticated are these agents are they fairly simplistic I think back to two or three decades ago that you know this website if this done that I think is what I have T T T T where you know you set a task and you kind of wire three different just comes together right that's the way I think of agents at the moment I know it's not that crude but I'm wondering how sophisticated they are currently Nikita it is so sophisticated and I think 2025 was deemed as the year of AI agents I think it was just the beginning of AI agent behavior I remember last year I think was March when I launched the genitive AI lab at CUNY I was teaching a class on AI agents and I said we still don't have a definition of what this is and I don't think we still have a true AI agent that doesn't hold anymore now in December 2025 just to show you like how quickly it's changed at that time like it was just theory they were coming up systems but I think a lot of people say that they are using AI agents in their newsroom and that's also because it's marketing by a lot of tech companies it's AI agent no it's basically just a wrapper around chat GPT it's it's basically your GPT that you're having that's not an AI agent a true AI agent is something that you give it basically a bunch of data on you tell it what your goal is what it needs to do it will understand which documents it needs to look into what it needs to do be able to reason through that and show you step by step what it did exactly what it did how it why you decided to choose that and that's the true brainstorming partner I feel like I need yeah I just feel like look at the reasoning I have the agents for any task okay goes down this route and then it figures out okay this I can't catch this data I can't analyze this and then it's actually communicating with you and to me personally it has helped me become a better systems thinker because you are seeing this like very logical way of breaking down problems you know written in in text even if you are using a cloud code yeah I think the transparency matters a lot yeah release yeah and I think it's going to be altering the way that we think and I'm not talking about sort of the kind of superficial way of if because you rely on Google search or you can retain less information same thing for AI but I think people that get exposure to the reasoning kind of inherently get exposure to a new type of machine thinking and so I wonder if the way that we approach problems as a society will become more logical over time or if it will backfire and will be more emotional and kind of driven by by emotions rather than reason of course funny because the data scientists data journalist and you likes the logical side I probably like the more creative the the emotional end of things but I think I think the two of those together great compliments and I think that again that that transparency being able to see how these systems work is going to be so important in terms of ensuring the trust of people who are either using them or consuming them and I also want to bring up the next the other big thing that happened this year which I feel like it's still flying under the radar for newsrooms but it is hot and tech MCPs which is model context protocol it's basically something that had come out last year developed by Anthropic but now has been adopted as an industry standard that lets AI models connect securely to any external tools and data sources so this is they are calling it like the USBC for AI context which essentially means that I already have one where right within Claude I can or chat GPT connect to data sources if I have created an MCP server and now that is and why this is a big thing is because chat GPT just a couple months ago had come out with chat GPT apps that they are building that entire apps SDK is built on top of the model context protocol which is also pointing towards this new pathway where you can see agents to agents talking really there are a lot more payment gateways opening up via agents so I think when we talk about the business model of news and what happens to information I feel like that the business model aspect we are still yet to see that area is still evolving and developing and agents play a huge role in that in terms of how we can use it because one way the agentic browser is the other example right you have Claude code but we had open AI's Atlas come up we had perplexities, comment, this DIA which previously was ARC these are all agentic browsers and I can go to them and I'm literally having conversations with 10 different tabs at the same time mix of YouTube videos and everything and I can tell it to search and open up all of these tabs that I want to all while a conversation with it and so I'm imagining if I have a subscription to the times and the post and the journal and I want to be able to have my agent talk to all of this to help me summarize what exactly everyone saying about a particular topic and then contextualize it for me and the way in which I want to read it that again that's one possible future that I see that also excites me as a news consumer to be honest because I can now look at 10 different publications first of all I agree that the MCPs probably one of the most underrated kind of these interruptions or innovations that will happen to use at the same time what you'll see is that the first wave of MCPs will be kind of you know when newsrooms were trying to transition to mobile we're just trying to copy their website or their newspaper and so fetching your archives that's or your news tell me what's what happened today you know that's pretty commoditized but I think that's you know that's going to be the very first thing that publishers do but I don't think that's a winning game. The winning game for a publisher is to connect the MCP to proprietary sources of data. So for the Wall Street Journal could be markets could be their rankings for the New York Times could be all their data on product reviews and the use case you describe is interesting of like you have the network of publishers and you can compare but at the individual level I don't think it's about asking you know retrieving content it's about can you go upstream and interact with the information for instance we have you know as part of applied Excel we we are basically a news gathering or data gathering company that happens to use AI for every step of the process but we apply these to a different domain science and pharma and regulation and a lot of the work is around structuring contextualizing the data the way that you can use MCP to speak with data and produce I call newsletters articles is absolutely mind blowing in terms of for instance yesterday we were looking at all the new drugs that Pfizer was developing and it's basically it has the context of the data and that is in the MCP and we asked what is the most interesting thought provoking thing that Pfizer is doing without any context and it came up with something pretty pretty cool which was Pfizer is developing a Lyme disease vaccine and they even worked on that since 2002 when sentiment about vaccines was at its lowest in other ways we don't have sentiment data about vaccines but it's looking at the core data and then using some of the reasoning of the model to make those connections and so now imagine unlocking some of their reasoning into the wealth of data that publishers have that's something that investment firms, companies you know other companies will purchase that's a like that's a pretty pretty compelling business model to me as opposed to just circular you know distributing news articles I completely agree with you I think that's been the reflex of most people of like let's just create an entire archive into an MCP and put our content there going into 2026 I don't want us to think of ourselves as content factories and I don't think news value lies in content as such when you look at the first principles it's in the data it's in the way in which journalists is asking those questions and then you get that interview that's data at the end of the day those codes are all data that is all human value that's the work of human empathy that's able to get that conversation I feel like the same type of conversation that we're having will be a totally different conversation with somebody you don't have that relationship with right those are all parts of data and being able to encode that brings up so many more nuances signals that can be missed in that way and I think going into 2026 MCPs is something that newsrooms have to keep on their radar yeah I think if AI destroys the world we can blame MCPs when they first began talking with each other but yes I'm super excited about it I know when we talk about going a platform it's how do you connect to all these disparate sources of information and once you start connecting them it unlocks value that we've never seen before and all the cases that you're talking about is super super fascinating so I do hope to see more of this in 2026 whether MCP becomes the standard it seems like it's certainly leaning that way but the idea that these agents and bots and corporate sources of information can now all start talking and extracting data gets really really interesting we can keep talking for hours as we always do but I feel like I want to end with just maybe some predictions what's your prediction for 2026 yeah I would say generally that AI co-pilots you know will be the standard in newsrooms I think you're starting to see that more and more but I do think that we will see AI stretch across the beginning origination stage when you're just reporting taking you all the way to production it won't be just a you know a siloed process that AI will really be companion in the newsroom and that you're going to see a lot more news organizations use it in a smarter way and hopefully get to these places where we're talking about which is agentic work and really crafting a story not only using your human intuition to draw out a story that bots can't produce but also connecting to all sorts of pieces of information to be able to display that to their readers as well so yeah I just think that if 2025 was there was a lot of experimentation I think this becomes a real operational reality in the newsrooms completely agree with you what I didn't inform my name and lab prediction was that AI is going to rewrite that architecture of the newsroom because this is the inflection point that all of our past logic is just not going to hold true because it's all going to be questioned your business model is going to be questioned the way in how you're delivering value with self is going to be questioned if your audience is not able to directly get that from you but it's getting summarized by a bot how are you driving that value to your audience what does that happen and I think that's one question but then that drives across everything into how workflows should start looking like I don't think this is going to happen overnight in 2026 but it is a shift that we will start to see where up until now it was about how do we use AI and plug it into yesterday's workflow and all we were doing with making yesterday's processes more efficient but with 2026 and the capabilities AI is getting better at a new technology where we start to reimagine what does that look like now what does a newsroom look like in this new era not just retrofitting into a print era architecture if AI is able to summarize across beats basically how do you organize a newsroom if you need both the technical capability and a product person how do you structure your teams and what are those new skill sets that you need to be hiring for and putting together a newsroom so it's a whole architecture shift that we begin we won't be able to accomplish that in 2026 because we have that institution inertia but that's where I think a lot of experiments and the work that I've been working with a lot of publishers and seeing those conversations and everything they're going to be forced to happen because tech companies are also going to evolve and it's not just tech companies it's also other startups that are evolving at the same time and so legacy institutions need to be able to reinvent themselves and I think that's going to be marking the big name of that I think it's interesting what you said about the architecture and changing from that print architecture I'm old enough to remember the idea that we're digital first now how it's silly to even say that or a mobile first but the truth really is that news organizations are still stuck in that old print architecture even if they don't print anymore where they obsess over every word and publish something high five and move on when the reality is the story is still a living entity and it's malleable and it needs to be personalized and there's more that you can do with it and so yeah I love what you said about the changing factors in the newsroom are you Francesca? Well my prediction builds on what both of you talked about which is kind of the intersection of the emergence of MCP servers and AI agents and the availability of data and pressure you know business pressure internally and from the tech companies and so I think we'll see you know what I call the emergence of these systems pre-new systems which if you think about it you know most of us think you know we are living in the present but when it comes to news we're actually living in the past if you think you know any recent breaking news and notification on your phone that event already happened you know whether it's a financial story and the market has already moved or if it's some sort of crisis the damage has already been done and so we aren't really reading the news even in breaking news you are actually reading a history book of the recent past and so if you think about all the things we've talked about the ability to process data faster than ever before real-time networks digitization of data availability of all of these sources we as we enter in this zero latency world where the speed of information or the speed of this reading information is collapsed to zero you know the gap between something happening and you knowing about it is going to be reduced even further and so there's going to be sort of a race to speed and so what what happens when you no longer are able to break the news and you are now seeing it what are the early signals that you can detect before it becomes news and in fact one interesting trend we didn't get the chance to talk about but I'll bring it up is how company news organizations like CNN CNBC are starting to use alternative sources of data like prediction markets is kind of a new lens you know on our new perspective or a new reinterpretation of bowling anyway my prediction is that you'll see the emergence of these pre-new systems and in fact that's the premise of our new book the science of first essentially our reflection on what does it mean to live in this world where anticipation is sort of the most important currency it's not about you know being fast it's about being first and so how do you think about information as utility and journalists are sort of the prime people that will benefit from sort of these shifts but they need to know how to use the tools anyway so I'm excited for 2026 a lot of great things in the works across the industry so we'll see maybe we can connect early in 2026 to see if some of our predictions have come to reality well I guess I was so general in mind it's probably hard to fact check my own purposefully but we have AI to fact check yes but I do love just Francesco we talk about journalism being this tedious cycle of react and repeat and that it's survival depends on helping society interpret what may happen next and that's true that's true used to be oh we got five to eight minutes to get this take out now you're talking about negative time right here yeah we are entering what we call the age of anticipation and in this age of anticipation you have all of this information in data available to you to understand sort of possible futures and outcomes and you know I think we'll be faced with two choices and it's either you are a witness to the kind of the aftermath of what happened or you can kind of be thinking in a new way about these frameworks for what may happen next so I think every single thing that we've talked about today are sort of some of the ingredients of the research that we have been doing which kind of combine not only the technology the human element the workflows and what the future may bring us so excited to dive deeper into this world of the science at first yeah I'm excited to talk more about all of this it's going to be putting all of that thoughts that we had now out there into the world as well and what the new era of journalism and news looks like I think it's it's again it's not about us automating the news or journalism but about finding an Olympian but the first principles reckoning of what the journalism industry is where our value lies in an AI mediated information ecosystem and what's possible now that just was impossible without this technology previously and so I'm so excited to get to get into all of this in future episodes but thank you Francesco and Scott this has been a fun deep dive of the now when I use the word deep dive I remember Google's that's probably where I've seen a lot of too many deep dives but it's been our human generated notebook and version of the world in 2025 the news industry in 2025 well always great to have a conversation with two brilliant minds like you and was fun to reflect on 2025 and think about what 2026 will bring us so thank you all right let's do it again in the new year thank you for having us that was Francesco Marconi the CEO of Applied Excel and Scott Austin the head of business development at symbolic.ai to stay up to date with the Newsroom Robots Podcast sign up for our newsletter at newsroomrobots.com this podcast is made possible thanks to the Harvard Innovation Labs Spark Grant I'm Nikita Roy and this is Newsroom Robots [Music]