After nearly 20 years at Fremantle – serving as Head of Social Media and later as SVP of AI and Automation – Tom Hoffman departed the production company to launch his own venture, Gemba, a consultancy at the intersection of technology and entertainment.
In this interview, Hoffman tells us what his predictions are for the future of the industry, and how broadcasters can keep up with the likes of Twitch and YouTube.
Can you tell us a bit about your career so far?
I came from television. My first real job was operating a studio camera for a news show in upstate New York. I then moved to Los Angeles and developed YouTube and Social Media at Fremantle. At that time, YouTube was a fascinating incubator to develop content at a low cost and deliver it to television.
Once my tenure at Fremantle came up in 2024, I created Gemba. I can code, and wanted to run towards the thing that scares everyone else: Generative AI. I was Head of AI and Automation at Fremantle for the last year and a half. It was fascinating to see how social media fit into the television industry, and then eventually how AI fits into that.
What challenges does the entertainment industry face?
The secret to YouTube is ultimately the interpretation of data. I think the entertainment industry is struggling with TikTok and newer platforms, because there is not an open API that is easy to use. TikTok is very different for each user. There are questions about what people are actually watching and how they feel about it.
TikTok’s AI is really revolutionary. It uses collision-less embedding to cluster behaviours together and deliver content to people with similar interests. Depending on what the user does, the algorithm will alter what content it feeds them.
With Gemba, I wanted to build an engine using AI that can watch TikTok videos and learn about them. The engine can watch much more than I ever could as a human. The AI looks at videos and their comments and analyses the emotions of the content and the audience.
How can Gen AI transform the way we interpret data and create content in the entertainment industry?
I think we’re transitioning into a post-data era – that data and structures of it are more important than ever but we don’t necessarily have to see it. And I think that’s the sort of magic that AI brings. With Gemba, I started making dashboard with data, but I realised I should be developing something that could just skip to the end, because that’s the promise of Gen AI: the death of complexity. I considered what people would actually want, and how they could interact with content in a way they will keep getting what they want. Then, I created an idea engine from it.
Rather than providing dashboards with a list of trends – dances, cooking, etc – I skipped to the end. I removed that layer which could send you in the wrong direction if you interpret or analyse the data incorrectly.
Instead, it gives a list of television show ideas based on what is working on TikTok. I put it into an interface that is like Netflix because it is familiar – I call it ‘Notflix’. Instead of a dashboard simply showing the trends, it shows a page to scroll through of AI-generated show ideas.
I am also planning on working with brands, so I could then make a ‘Notflix’ for brands, based on their brand identity. For example, a car brand could create ongoing lists of content ideas that would be relevant to their brand.
What part of the entertainment industry will be most disrupted by AI?
I think of AI, even more so than social media, like the birth of the printing press. AI is not necessarily going to re-create a popular show with a big production budget. But some daytime television shows, and classic TV studio worlds, could be democratised by AI going forward.
There’s a lot of generation and work that goes into it. There’s a death of complexity, but not a death of utility or human input. We’re just working differently now with AI.
AI cannot read your mind, but what it provides can be refined until it fits an idea in your mind. I think the AI hype correlates with the amount of effort it takes to generate the product the user wanted to begin with. But the process takes a lot of time, possibly hours, days, or even months. I think by 2028, the Oscar Award for Best Picture could be shot by an iPhone and stylised by AI.
Will AI shrink the size of film and TV production teams?
Most of the ideas generated by AI are really terrible. But eventually, you get a diamond in the rough. It’s the same scenario – it’s a different type of work.
Even digging through thousands of television show ideas is work. The data mining will be done for us by AI, but the work will come down to the interpretation of that mining. I think we’re a long way from pushing a button that could give us a new hit film. At this point, there are a lot of modalities that still have to be accounted for.
We’re seeing a lot of amazing AI models and engines around specific focuses, but they don’t necessarily talk to one another. They may be able to generate voices, but not be able to generate a voice with the emotion of a human – that is complex. Ultimately, the “army” of television and video will shrink. But to get the quality that really works, I would argue there is almost as much work as before, to get a final product that is actually consumable.
How do you see content platforms evolving in the era of AI-generated video?
Broadcasting brands are extremely valuable, because they represent a certain level of consistency for grouping content – in the way that cable TV always did. I think in this newer YouTube universe, the rise of categorical brands has a life that we have yet to see. It thrived on cable, and I think it will transition nicely into platforms like YouTube.
As these AI video generation engines start to take off and become accessible to the public, there will really only be one place where those videos are put: YouTube. I think YouTube are concerned about being flooded with garbage.
With Facebook, their video platform had the potential to be as good as YouTube, but it was neglected, and it’s flooded with clickbait content. The algorithm has become so obvious to favour video creators. YouTube has escaped that, and has done an incredible job with using AI for recommendations.
Can traditional broadcasters learn anything from content platforms in using AI?
Going into the future, YouTube could learn something from Twitch. In 2025, YouTube announced they want to bring some of the biggest streamers from Twitch to YouTube’s live-streaming platform. Live-streaming has taken a backseat, and that’s where a lot more experimental things are happening. So it’s not surprising that YouTube would look at established, TV-like platforms in the social space to take information from. I think the broadcasting industry could take inspiration from platforms like Twitch, because of that live element.
I think with streamers, there is a dopamine hit with interaction that is reminiscent of what came from the Golden Age of television – you are part of that broadcasting experience. There are even mechanical details like text-to-speech, powered by AI, that allow viewers to type comments that are spoken out loud to the streamer. There is an interaction there that is more human, but also more traditional television in ways that were not able to be captured before. YouTube is working on bridging that gap, because no one in the digital space has really bridged that gap before.
