Nasser: 

Welcome to Shift Happens. I’m Nasser Sahlool, and today we have a treat. We actually have a returning guest with me. Today is our first guest from our first episode, Dan Temby, who leads technology and analytics at DAC. 

Dan: 

Hello, Nasser. 

Nasser: 

Hello Danny. Welcome back. 

Dan: 

Danny, like my father. You see, I was called Danny as a kid, and then it went away. And then you brought it back. 

Nasser: 

Well, I feel like when you reach a certain point of your age and experience, you need to reconnect with your youth. 

Dan: 

You know, I come across Danny’s from time to time in business. And I’m always surprised they didn’t make the change to Dan. But maybe they know something I don’t. 

Nasser: 

Well, most of them are women I think as well. 

Dan: 

Danielle. 

Nasser: 

That’s right. 

Dan: 

Yeah, yeah, yeah. That’s right. 

Nasser: 

So, Dan, it’s been a few weeks since we were last together on this show, and you’ve been a busy boy. 

Dan:  

Yeah man. That’s so much going on. What an exciting time it is right now. And in a material way, not like the hype cycle of AI and all the promises that we got after ChatGPT launched. And years of, like, toiling away, trying to, like, get something material happening. And then just in the last six months or so, stuff has really popped off,  and there’s a lot of really exciting material use cases coming out, and we’re in the middle of a lot of those. Yeah. 

Nasser: 

So what fundamentally has changed in the last few months? 

Dan: 

I mean, if you want to if you boil it down, you know, there’s a lot of business… The fear is subsided. The uncertainty is backed off a little bit. There’s expectations of material value. But if you boil it down, I think there was big improvements like step changes in models that came out that solved some fundamental problems to do with capability and context and thinking and rationalization and stuff, that all of a sudden, the things that used to frustrate us as builders using AI tools, all of a sudden it felt like lanes opened up on the highway, and we had all this capability and everyone like leant into that hard, and it’s moved us along really quickly in a few months, I think. 

Nasser: 

Funny you should talk about lanes opening up on the highway, because for those who don’t know, Dan and I have worked together for a long time. We live really close to each other and we actually carpool. No, it’s true, we carpool. And on the days that Dan is driving, lanes do not, in fact open up on the highway. They seem to block up, and we get to work late. 

Dan: 

And what happens when you drive Nasser? 

Nasser: 

Everything goes excellent. 

Dan: 

Great. Any other self-congratulatory anecdotes you want to throw out? 

Nasser: 

Not right now,  but we might come back to that. 

Dan: 

Okay there’s probably a bunch, yeah. 

Nasser: 

So you said a few things earlier around what’s changed in the landscape. And if I can think of a word that kind of sums up that I hear a lot in the last few months in the space, it’s agentic. So for people who don’t, who are kind of too shy to ask, I’m going to ask on their behalf. Because of course I know what this means, right? What does agenetic mean? 

Dan: 

Well, I’m glad you asked, because a lot of people I find are using it… It’s being thrown around a little too liberally, right? You know, DAC, we’re an agency. What does that mean, to have agency over something? We have agency over our clients’ media spends and their budgets and their marketing activities. We have been given the authority and the freedom to go and make those decisions and do that work on their behalf. That’s the definition. 

So when we talk about AI having agency or being agentic, it’s exactly that. It’s giving an automation or a language model interface or an endpoint, the authority or the ability to take action on your behalf. So, in my opinion, and this is what I think and I run into this a lot, going to ChatGPT and asking it something that you may have asked Google previously, is not really agentic. Yet that gets called agentic search a lot. I think that’s something that’s coming where there’ll be action taken after that search, after that discovery moment. 

Fundamentally, what we’re seeing now is the rise of harnesses and wrappers that sit over the top of language models, that are equipped to take action on behalf of the user and do that work. And it’s that work that’s being done that’s really getting everyone excited, because suddenly we’re seeing outputs that are usable and workflows that are streamlined. I don’t have to start here and cut and paste and move something over here, or go here and ask a question and then take that knowledge and do something else with it elsewhere. 

Nasser: 

So, so can you get a little more practical with that? Because you talk about, you know, harnesses and wrappers that get placed around models in order to take action. So walk me through what a typical AI interaction would have been like in the before, and what an interaction would be today with agentic capability. 

Dan: 

Sure. And again, quite simply, and this might be remedial for some listeners. And if it is, I apologize. But I’d give you a personal experience of writing code in the ChatGPT window. It could always write code and functions for me since the very beginning, and some of them not to complicated, but it could do the work. I would ask the question, get the code I required, copy it, move over to an IDE, say I was doing a project on a microcontroller like an Arduino, where I wanted to write a package to make it do a certain thing with a certain module. I would have to then take that code, put it into my Arduino IDE, and then compile it down, load it onto the device and see if it worked. 

In the now, you can go into an IDE, which is an Integrated Development Environment. That’s what that term stands for. It’s where coders write code. And I will go in there and I’ll just simply say what I’m trying to achieve. And then the agent says I got you, I know what you need, picks up my question, interprets that to the right request to the language model, generates the code, and then it will go ahead and compile the code and write the code to the device for me all in one cycle. 

So you can see how at scale, that’s a personal hobby project of mine. But at scale, if you’re writing hundreds and hundreds of functions, or tens of thousands of lines of code across many, many code bases or many, many projects, having that actually execute the work, compile it, and publish it all in one sequence can be very powerful. 

Nasser: 

So if I understand this correctly, it’s its ability to connect with other systems and take action on your behalf. 

Dan: 

Certainly, yeah. Within a controlled environment with the right permissions and the right authority. And it’s that setting, how do you set those permissions? How do you establish that authority? How do you set up the trust and security that I think is the big problem people are trying to solve is we take this agentic capability and we push it further downstream into the hands of consumers, which I think is coming right? For sure it is. 

You know, OpenAI rolled back their instant checkout feature earlier this year. They were kind of first to the game and had kind of low adoption, and there was some real clunkiness, but those were some of the troubles that they were having. How do you actually take a user’s credentials and act on their behalf to actually make a purchase all the way through? But it’s coming for sure. 

Nasser: 

So that’s that concept. And we hear a lot more about this these days around agentic commerce, where you’re able within an LLM or your interface to have the conversation and for it to execute commercially on your behalf. 

Dan: 

Right. It’s probably good to talk about the evolution of that to put that right in context. This agentic capability, the use case is early on with the coding. I’m in an IDE, I explain what I want, I get code written, it compiles, it works, I check it in and I ship a feature. Great. The most recent evolution, I think Claude Code and CoWork have really pushed this, and now Codex has just come out with their launch, that’s OpenAI’s coding agent system called Codex. They pushed it just last week, I believe, which is bringing that kind of agentic: “Hey, let me do some real work for you” capability to people on their desktop computers who may not be coders. 

So you build a PowerPoint presentation or an Excel spreadsheet, or you want to analyze some data, or write a document, or do a series of events in sequence that ends in some meaningful output. That might involve using your computer to conduct some of that work, browsing a website, catching information for you. It’s the same idea of this harness or a wrapper over the top of an LLM, but it’s working on behalf of a maybe less technical or non-technical user to perform some other administrative or knowledge work task. And I think that’s very exciting. 

And now that is pushing the consumer, the B2C world, where every consumer is just relying on an agent to do meaningful work for them, is what you’re talking about where, hey, and it can be done now in ChatGPT if you set it up correctly. Book me a table, it’ll use the OpenTable integration and do it. Hey, order me a Starbucks. It’ll order you a Starbucks, but you have to go in and purposefully connect those tools. First, set up your accounts and do all of that. I think the real evolution of that, of this space is where it starts to do that for you seamlessly. 

Hey, buy this thing for me, find me the best pair of running shoes and buy them. And it could be from a retailer you’ve never dealt with before, but the agent will have information and authority from you, and it will have your payment details and everything stitched up to act on your behalf. And it will go find the best retailer and make that transaction. But now that opens the question is what is that retailer doing to prepare for that? 

Nasser: 

Right. So what should those retailers and looking beyond retailers, what should those marketers be doing to prepare for that environment? Because it’s not quite here yet is it? 

Dan: 

No it’s not, it’s not. And it’s interesting, you know, and I don’t know if you used like the Atlas browser or the Gemini overlay where you actually get the AI to use your computer. I did it just a few weeks ago. We built a system, and I was trying to get it to add a bunch of users through the interface for me. I just copied and pasted the list, and I watched it play around with this thing. It was very awkward and a little frustrating, if I’m honest. 

And that’s because user interfaces are designed for dumb humans and eyeballs and brains and the way we perceive the world.  But an AI agent is nothing like us in that regard. It is a structured data transactional machine. So the navigating user interfaces is the worst way to prepare for this stuff, even though for broad adoption, all of these vendors and providers are trying their best to make their systems navigate the web as it exists today, which is user interface heavy. Right? The emergence of headless systems. What does that mean? 

It means your website, with everything it can do, every piece of information that it can offer, every transaction, it can perform every purchase, it can make every download, every whatever all available through like a command line or an endpoint or really ugly, almost like a terminal view that you would never use as a human. But an AI agent is going to love that because it’s structured and it’s meaningful and it’s reliable and it’s the same shape every time. It doesn’t have to interpret the graphics and wonder where to click to do a thing. It will ask for which endpoint I need to perform this function and it will do it. 

So what that means, I believe, is this emergence of like the headless internet or the UI less internet. You’ll hear about things called Web MCP. That’s web-based model context protocol. It’s an emerging standard for AI systems to talk to each other, understand what they’re talking about, and not have to work it out every time, a Structured Information Schema. There’s another one emerging called UCP, Universal Commerce Protocol. And that’s more for transactional endpoints where I want to make a purchase. 

So what I do as an AI agent, I will expect to receive these kind of authentication signals, these kind of validation signals and so on. And then there’s emergence of things called CLI Command Line Interfaces, which anyone who’s used Claude code will know a command line interface. Again, that’s back to this terminal thing. It’s software that will run on a computer or run somewhere as a command line utility and not a visual interface. And that is empowering everything we know about computing and everything we know about marketing and consumerism and the internet, to be an agent native, it’s preparing it to be operated by robotic agents that do not need the user interfaces that we need. So it’s a really interesting time. And this abstraction between the front end and the back end, in a way that is really focused on where we think the stuff is going. 

Nasser: 

So it raises an interesting question in my mind, where an agent is going to be operating on your behalf, making decisions on your behalf, making comparisons, coming back with recommendations, and making suggestions to complete a purchase on your behalf from a given brand or what have you. These brands, many of whom have spent enormous amounts of money building brand, building creative, building capabilities whereby they are able to separate themselves from their competitors. And they’re doing it based on emotion. Right?  

A lot of that emotion is visually based. And how is a machine that when we originally thought of AI, we were thinking down the lines of, it’s going to think like us and make decisions like us, but in truth, it’s not. It’s still a machine. It’s still thinking like a machine. How does it make these decisions against brand investments and brand concepts and creative on your behalf? 

Dan: 

Do we have time for a little side quest, abstract analogy here? 

Nasser: 

Sure. 

Dan: 

Okay. You ever seen the movie 50 First Dates? 

Nasser: 

I have. 

Dan: 

And who’s that, Adam Sandler and… 

Nasser: 

Drew Barrymore. I don’t like Drew Barrymore. 

Dan: 

I thought she was good. Yeah. In that movie. 

Nasser: 

In that movie yeah. But I’ve never been a never been a fan.  

Dan: 

Okay, well, I’ll let her know.  

Nasser: 

Sure. Thank you. 

Dan: 

At the end, the final scene with the she finally she wakes up in the boat. And she puts in the videotape, right? And she gets the information and. 

Nasser: 

Hold on. For those that were born after the year 2000, we used to have videos on these physical things called tapes. It was magnetic. You had to wind it when you returned it to Blockbuster. 

Dan: 

If you didn’t rewind it, you got charged a fee. Anyway, so Drew Barrymore puts the cassette in and they see a video which basically primes her on her whole life and who she is and why she shouldn’t freak out. And then she goes upstairs and her family is there and everything’s great. I think of that a lot when we’re working with agents, because engaging a language model or an agent surface is like a stateless experience, meaning it doesn’t know you make a transaction and everything it needs to do that transaction well has to go in that first call. 

And getting that right and making sure it’s got the right information so that it gives you the response you expect is really an emerging art form in this space, like navigating that context. There’s a big project that’s really sparked this agentic revolution recently called OpenClaw. Have you heard of that? 

Nasser: 

I have. 

Dan: 

Yeah, and it took off like crazy. I think it was the most starred GitHub project in the history of GitHub in one day or a week or some incredibly short time frame. And a lot of people get very excited about, and a lot of people spend a lot of money on it, too, because they set up these agents that were going off and working for them autonomously, but burning thousands of dollars in AI credits and giving them no real value. 

There was huge security risks, but it really opened everyone’s eyes within a certain community of the power of agentic capability. One of the files inside an OpenClaw environment is called Soul.md. It’s literally this agent’s soul, and it’s a markdown file which basically a text file, and in it is contained instructions and information on who this agent is to its operator. I’m getting to my point, I promise. 

And just like the video, I thought of the video, Drew Barrymore’s video in 50 First Dates. Every time that agent does any work, it has to read that soul file so it knows who it is and what its purpose is, why it’s doing, what it’s doing. So to get back to your question, the brand and visual identity of organizations who are trying to tap into different markets, they should be reflected in the values imparted by the users on their agents. 

So let’s take you and I for example, we’re very similar, we’re great friends. We’ve known each other. We have very different approaches to making certain purchases. I’m very much going to be driven more by facts and value and, you know, technical specifications and utility. And you might be looking more for user opinion or brand or alignment with your values. That would be reflected in the soul file, or the equivalent of our respective agents that are acting on our behalf. 

So when they go looking for information, they’re still going to need to find this through these structured endpoints that I talk about. But the stuff that you mentioned won’t be lost or forgotten. It will only be relevant when it’s relevant. Right? I believe that very strongly, I think because there is no way…you can’t have both. We can’t have this agentic headless world that is going to be well equipped for these AI bots to go and traverse the web and make action or take action on our behalf, and also all the beautiful visuals and stuff that go along with it. There’s a juxtaposition there, and I really think it’s about you training your agent to understand what means most to you, and then it will act accordingly. 

Nasser: 

My agent is my funk soul brother. 

Dan: 

Yeah, yeah, it could be. It should be, right. 

Nasser: 

That’s what it should be, it should be that. So last question for you. Given that context, what should marketers be doing right now to prepare? You mentioned things like Starbucks and OpenTable doing it, like what should people be doing right now? 

Dan:  

If you’ve got cycles in your Dev team and you should have because everyone’s moving quicker now, we’re getting more done, so find a way. But talk to your technical team. Talk to your development teams, talk to your agency partners about things like Web MCP, UCP, about any native applications you have and whether CLI routes are appropriate for that. Look into building a plugin. If you’re a retail brand, a B2C brand, look into building plugins that can go into these marketplaces of places like ChatGPT or Anthropic so that as this frontier of agentic consumption emerges, your brand can be there in the list of people saying, yeah, I have a loyalty card there, I’ll plug them in. And now hey presto, I’ve got agentic commerce happening for my coffee orders or whatever, like Starbucks have done. 

Nasser: 

And I know that as of right now, you and your team are actually consulting with everything from retailers to financial institutions to telecommunications companies and so on. 

Dan: 

Everyone. I’m having a conversation along these lines, which is why I’ve so much energy about this I think, like 3 or 4 times a week. We’re having a conversation like this with our clients, whether it’s about how AI practices can leverage new data, how they can tap into old systems, how they can generate brand new user experiences. It’s a very exciting time. And again, I walk this line daily and we talk about it all the time, between overwhelming enthusiasm for a space I’ve been in for so long, because all these ideas in my head can suddenly come out at warp speed, and then these moments of existential dread, it was like, oh, what does this mean for me? 

But right, you know, the more I talk to people, the more I think there’s a role for people like us and people like those listening who are really actively wanting to get their hands dirty, roll up their sleeves, and do this work because it’s not going to do itself, and it needs some guidance and some shape, and we’re just excited to be a part of it. 

Nasser: 

It’s alright Dan, I think we’ll keep you around for a little while. 

Dan: 

Thanks, buddy. I appreciate you as always. 

Nasser: 

It’s my pleasure. So here’s the shift. AI is moving from conversation to execution. For the last few years, the value was getting answers. Now the value is in getting things done. For marketers, that means preparing for a world where discovery, evaluation, and even transactions may be mediated by agents, not just users. The brands that win won’t just have great experiences for humans, they’ll have information and systems ready for machines acting on behalf of humans. 

Now make it happen. Follow Shift Happens, leave us a review and share this episode with your team. If you have any questions for the podcast, email us at shifthappens@dacgroup.com,  we’d love to hear from you. Dan, thanks for joining us again. 

Dan: 

Thanks, Nasser. Also, I look forward to you driving me home later. 

Nasser: 

Thank you. I will do it excellently. Thank you for joining us on Shift Happens, I’m Nasser Sahlool.