Brian, the minute you started talking about this on stage last week, I realized it was the productization of something I mused about on the MightyHive blog in 2018. As AI is able to interpret the subjective meaning of content (and the subjective experience of readers, viewers, and listeners), the execution of "context" in digital advertising takes on entirely new dimensions.
Maybe I'll substack about this. Haven't had the time yet!
AI is completely reshaping our relationship with information and so the way we generate, store, structure and present it. As a consequence, the existing web, which is basically an information retrieval platform, is going to become as obsolete as Blockbuster and Tower Records. What you've outlined focuses on contextual and structural elements like websites, pages and ad slots that I'm guessing will largely disappear over the next couple of years. I think the capabilities outlined are going to be essential, but how and to what they are applied is going to quickly and fundamentally be different from what we currently do.
I think we're heading to a world in which everyone has agents that are interacting with other agents (i.e. models interacting with models). Individual consumers are going to have personal agents with which they have consistently deepening and dependent relationships and these personal agents are going to interact on behalf of consumers with sources of model intelligence created by providers of products and services to collaboratively satisfy consumer requests. I assume the consumer models will also increasingly share intelligence with the general models in meaningful ways so that the entire ecosystem gets better and better at serving the needs of individuals.
Near term (where we are now) I think relationships with models will be episodic, purpose driven and mediated via superficial contexts, like chats and similar direct modes in which consumers interact directly with a shared model primarily to retrieve information and resources. Near to mid term I expect we'll be developing increasingly persistent relationships with the models similar to the relationships we develop with people and that responses will be increasingly tuned to us.
Longer term (and I'm thinking within the next couple of years), more and more of our relationship with AI will be with a constellation of personal agent models that we develop long term, trusted, always-on relationships with and which interface with the model universe on our behalf. I think we'll quickly develop personal model "staffs" that come to know us better than we know ourselves and which collaborate with model staffs representing other individuals and product and service providers to achieve whatever it is we're interested in accomplishing. At the same time, what we receive will increasingly be generated specifically for us. Initially we get text and pictures generated for us, but that will quickly include music and videos and products and services which will be generated on demand and just for us.
The web fostered an explosion of content and content interactions by all but eliminating the effort required to create and share into relationships, enabling us to scale our direct interactions and touch audiences numbering in the millions. I think AI will bring a similar explosion, but at the level of content, product and service creation as it all but eliminates the effort required to generate and orchestrate things and experiences.
Not sure what the implications will be for society, it could leave us self-sufficient to the point that we no longer need to interact with others, but I'm guessing our deep desire for human connection will lead us to deploy our new found abundance in ways that support community and foster relationship with people rather than isolation.
Love this. I think the current 'operating system' will need to change - the current computing paradigm is built around the idea of apps that are completely distinct in different windows, but what you're suggesting implies a totally different canvas for how we go about our day. Very fascinating to see how this looks, and who figures it out!
I hadn't really thought about the consequences beyond impacts to websites on the open web; considering the broader impact gets very interesting very quickly. Intelligent agents imply not just that boundaries of distinct windows fall away, but apps and compute contexts altogether as agents independently navigate and negotiate with data and processing to develop the constellation of resources needed to accomplish tasks. I imagine these "staffs" will end up having agents embedded in multiple devices which coordinate with a master digital "ego" to keep each other informed and organized. It implies that we'll eventually develop digital psyches, presumably with psychologies that mirror our own. Of course then we'll have to figure out how to maintain successful relationships with them.
Yeah, going to be fascinating to see how it unfolds and I fully expect it to happen a lot faster than any of us expect.
Great article. Do you have an opinion on cheaper LLMs like DeepSeek R1 and if it would lower the costs (for optimizing ad targeting)? The output quality could be sufficient to optimize ads.
Brian, the minute you started talking about this on stage last week, I realized it was the productization of something I mused about on the MightyHive blog in 2018. As AI is able to interpret the subjective meaning of content (and the subjective experience of readers, viewers, and listeners), the execution of "context" in digital advertising takes on entirely new dimensions.
Maybe I'll substack about this. Haven't had the time yet!
https://medium.com/@mightyhive/contextual-targetings-coming-renaissance-76f3afb0ae73
AI is completely reshaping our relationship with information and so the way we generate, store, structure and present it. As a consequence, the existing web, which is basically an information retrieval platform, is going to become as obsolete as Blockbuster and Tower Records. What you've outlined focuses on contextual and structural elements like websites, pages and ad slots that I'm guessing will largely disappear over the next couple of years. I think the capabilities outlined are going to be essential, but how and to what they are applied is going to quickly and fundamentally be different from what we currently do.
I totally buy this. But what comes next?!!?
I think we're heading to a world in which everyone has agents that are interacting with other agents (i.e. models interacting with models). Individual consumers are going to have personal agents with which they have consistently deepening and dependent relationships and these personal agents are going to interact on behalf of consumers with sources of model intelligence created by providers of products and services to collaboratively satisfy consumer requests. I assume the consumer models will also increasingly share intelligence with the general models in meaningful ways so that the entire ecosystem gets better and better at serving the needs of individuals.
Near term (where we are now) I think relationships with models will be episodic, purpose driven and mediated via superficial contexts, like chats and similar direct modes in which consumers interact directly with a shared model primarily to retrieve information and resources. Near to mid term I expect we'll be developing increasingly persistent relationships with the models similar to the relationships we develop with people and that responses will be increasingly tuned to us.
Longer term (and I'm thinking within the next couple of years), more and more of our relationship with AI will be with a constellation of personal agent models that we develop long term, trusted, always-on relationships with and which interface with the model universe on our behalf. I think we'll quickly develop personal model "staffs" that come to know us better than we know ourselves and which collaborate with model staffs representing other individuals and product and service providers to achieve whatever it is we're interested in accomplishing. At the same time, what we receive will increasingly be generated specifically for us. Initially we get text and pictures generated for us, but that will quickly include music and videos and products and services which will be generated on demand and just for us.
The web fostered an explosion of content and content interactions by all but eliminating the effort required to create and share into relationships, enabling us to scale our direct interactions and touch audiences numbering in the millions. I think AI will bring a similar explosion, but at the level of content, product and service creation as it all but eliminates the effort required to generate and orchestrate things and experiences.
Not sure what the implications will be for society, it could leave us self-sufficient to the point that we no longer need to interact with others, but I'm guessing our deep desire for human connection will lead us to deploy our new found abundance in ways that support community and foster relationship with people rather than isolation.
Love this. I think the current 'operating system' will need to change - the current computing paradigm is built around the idea of apps that are completely distinct in different windows, but what you're suggesting implies a totally different canvas for how we go about our day. Very fascinating to see how this looks, and who figures it out!
I hadn't really thought about the consequences beyond impacts to websites on the open web; considering the broader impact gets very interesting very quickly. Intelligent agents imply not just that boundaries of distinct windows fall away, but apps and compute contexts altogether as agents independently navigate and negotiate with data and processing to develop the constellation of resources needed to accomplish tasks. I imagine these "staffs" will end up having agents embedded in multiple devices which coordinate with a master digital "ego" to keep each other informed and organized. It implies that we'll eventually develop digital psyches, presumably with psychologies that mirror our own. Of course then we'll have to figure out how to maintain successful relationships with them.
Yeah, going to be fascinating to see how it unfolds and I fully expect it to happen a lot faster than any of us expect.
Amazing examples, already reality in some parts.
Great article. Do you have an opinion on cheaper LLMs like DeepSeek R1 and if it would lower the costs (for optimizing ad targeting)? The output quality could be sufficient to optimize ads.
My sense is that optimizing ads doesn't require a reasoning model, but there are probably use cases where the additional horsepower would be useful!
Thanks for sharing BOK, key Actionable takeaways for Media Agents are:
1. Leverage LLMs for contextual analysis
2. Build a simple Context Agent interface
3. Integrate with existing infrastructure
4. Pilot with a specific brand goal
5. Scale with client collaboration
6. Optimize for cost and efficiency
Love JSON examples, very clear