What AI Agents in Marketing Actually Look Like

**Welcome to AI Insiders**
(0:05) Hey everyone, I'm David Berkowitz and welcome to another edition of AI Insiders by AI Marketers Guild, part of architecture media. The architecture media angle is a little more relevant today than it is even if architecture is always top of mind in our world because we are just days after March's biggest ever event, Marchitecture Live 3.

**Marchitecture Live 3 Highlights**
(0:37) I got to say, yeah, full on I'm biased. Yeah, I'm wearing the brand. The community is a part of architecture, but I haven't been to many events that. I have a very short attention span when it comes to any of these kinds of events. Fortunately, this event was panel free, so a lot a lot of solo talks

**Marchitecture Team's Special Effort**
(1:05) and fireside chats, but the team behind it, Amelia Tren, Jesse Meyer, everyone involved working at March just did something special. I'd been on the planning calls for months. Hadn't done actually too much of the planning myself. So I could take 0.00101% credit for it. Just a phenomenal team effort there. And so I highly recommend

**Introducing Tom Riordan**
(1:38) trying to get to March events in person in the future should you ever have a chance. Today we're inviting one of our speakers from the event, Tom Riordan, who's been a longtime friend of Mark. I've even gotten to have this other role with Tom because Tom leads a lot of the AI curriculum for another phenomenal group, U of Digital, and I mentioned them

**U of Digital's AI Curriculum**
(2:09) in the past and we'll put links into their stuff, but they do some of the best training in the ad industry about the ad industry and related topics. And so they'll talk to brands, they'll talk to agencies, adtech companies themselves, and cover a lot of ground and the curriculum. I get to go and build on some of the

**Tom's Expertise at Word Power**
(2:33) things they're doing. I've been a guest lecture a couple times, including last month, and just the thoughtfulness of the curriculum development is so strong, thanks to some of the folks Tom, who also runs his own firm, Word Power. So Tom, I am fortunate to have gotten to learn from you in many ways, especially over the past couple of years and hear

**Tom Riordan's Introduction**
(2:58) some of your thoughts on where AI played a role, didn't, should have, could have, would have, especially in some of the market live conversations. But Tom, if you want to introduce yourself a little bit better, we're very excited to have you here. >> Sure. Thank you, David. Hey there, AI Marketers Guild crew. My name is Tom Riordan. I run a business called Word

**Word Power's Mission**
(3:18) Power that helps businesses go from zero to one on their AI journey. Although I'm increasingly finding some businesses are ready to go from one to two on their AI journey as well. David mentioned, I do a lot of work helping marketers understand AI, helping people understand key concepts and the fundamentals and then most excitingly for me is use cases.

**Hands-on AI Training**
(3:39) So, one signature piece of our sessions is that we're always showing demos or doing live work sessions, inviting people to participate in exercises, roll up their hands, and get them roll up their sleeves and get their hands dirty. So, my business has been around about three years. The first year we did a lot of consulting type work where we just

**Addressing AI Adoption Gaps**
(3:57) helping businesses think about how to use this new GenAI type stuff. And in my second year, it really became obvious to me that folks were being left behind at the average org. You'd have leadership teams thinking really hard about how they need to use AI. Maybe you'd have a couple power users who were building out special projects starting to see some return, but the

**Word Power's Mission: Lifting All Boats**
(4:15) majority of people, 80% plus of people at a given organization, weren't really getting their hands dirty, weren't really feeling supported. In fact, a lot of them are reading articles about potential job displacement and reskilling concerns they need to be mindful of and these individuals just haven't had resources. So, the mission of Word Power has been

**AI Transformation and Support**
(4:34) focused really the last year and a half on being the rising tide that lifts all boats at an organization, helping people feel they're being brought along with the AI journey, being invested in, enabled and supported as part of a broader AI transformation effort at a business. >> Awesome. Well, well, yeah, that training and clarity is so

**The U of Digital AI Alliance**
(4:53) sorely needed. And also it was exciting to see you bring up the AI alliance that UFD is launching, so could you just mention what that's about? >> I'm probably not the best ambassador on that. We should get someone from UFD to talk about that a little bit more. I did have the pleasure of crashing the AI alliance dinner which was some folks

**AI Alliance: Enablement for the Industry**
(5:18) from ANA and some folks from Anthropologic which is one of the sponsors. So, it's about enablement. It's about bringing something to the industry. But, yeah, your link there is probably going to be better than any elevator pitch, I think. >> Awesome. Well, had to at least give you a shot given your close

**March Event Themes**
(5:36) affiliation there and it was great news to hear live at March. So, what were some of the themes that you were picking up at during the event? What stood out to you? >> Yeah, for sure. You couldn't walk around for 20 minutes without either hearing AI spoken about directly in one of the sessions or hearing it come up in the networking portion

**AI Adoption and Networking**
(6:01) which was really notable I thought at marketer. All the marketing, the networking breaks were very social, folks were not trying to hold off and do phone calls. I met a lot of folks I'd never met before. I've heard similar experiences from others and just one of the constant topics was the realness of AI adoption within orgs

**AI in Marketing: Internal vs. External**
(6:22) if businesses are starting to see return themselves, ROI productivity enhancements, and then of course, in terms of marketing products and the actual way that we're advertising, that's the other piece too. So, two pieces: internal AI transformation for marketing type organizations and then how our campaigns are actually changing as a

**Challenging AI Agent Hype**
(6:44) result of AI. In AR's keynote when he was kicking things off he said something. What did he say about agents? He said, "None of us are really using agents right now. None of the people in this room are really using agents right now." And I thought that set the tone in an interesting way. I think he was challenging the hype around agents and

**Agent Use: Reality vs. Perception**
(7:06) saying we hear all this talk about agents but who's actually seeing return on their day-to-day and he was pushing the audience that there's a little bit of BS in terms of how much we in marketing are actually using agents and I thought that was interesting because I actually had I was talking to somebody else after at the event and they said, "I can't believe he

**Defining AI Agents and Their Use**
(7:23) said that. I'm getting all this return from agents. Can you believe that he's dogging us and saying we're not using them enough?" And I think there's a lot of fog there to clear, what exactly are agents? What does it even mean to be using them? Did you use it once? Do you use it every day? What sort of level of return do you need to see to be

**Stirring the Pot on Agent Usage**
(7:41) a productive agent user? And that came out at the beginning of the conference and already stirred the pot a little bit. So, I think that's maybe an interesting one to start on. >> Well, well, yeah, and we should definitely dive into this and so I'm just going to ask you yourself, are you actively using agents for anything

**Personal Agent Use**
(8:01) you're doing? So I am. I think I'm constantly using agents. Got to reorient myself though because the AI chat experience is becoming more agent-like. So for example, for today's we have an AI accelerator for U of Digital this afternoon and it's the agents session. So I'll be demoing agents to people and showing them agents that we use and we are constantly

**Demoing Claude Co-work**
(8:25) using a new agentic tool in that because we're just going to show people what's new. Today we're going to do Claude Co-work. Anyone on the call used Claude Co-work? >> Played with it, but. >> Lots. >> Nice. So, on one hand, you can do things in Claude Co-work that are just normal AI chat. And it looks and feels a lot

**Claude Co-work's Agentic Capabilities**
(8:44) normal AI chat, but if you spend a minute with it, you realize there's a lot more firepower under the hood. And Claude Co-work is different than regular AI chat because it doesn't just do what you tell it to do. It comes up with a plan and it deploys sub agents, little AIs to do things as part of executing that plan. And if it makes a mistake, it realizes it and it might course correct.

**Beyond Traditional AI Chat**
(9:06) If it needs to use a tool or go get information, it's going to do that. If it needs to write and create a document for you and leave it on your computer, it's going to do that. And these are all things that traditional AI chat really doesn't do. So Claude, the folks in Anthropic, are definitely calling Claude Co-work agentic but I bet there's many people

**Understanding Agentic AI**
(9:25) who use it or have seen it and don't really understand the paradigm shift, how that's different than traditional chat. So on that first point of what even is agentic, I think there's definitely still a lot of gray area. Yeah, one of the examples that I would bring up to folks is deep research. So

**Deep Research and Base 44 Agents**
(9:48) as agentic and how and for a specific task it meets a lot of the criteria, but it's also typically a one-and-done process. The agents that I'm using most consistently right now is so I'm a big fan of Base 44 and I've got a bunch of properties. I'm playing with this all. I actually was just checking

**AI-Generated Content for SEO**
(10:15) today to see, is it wait, am I on the monthly or annual price? And it's oh, I'm on the annual, but I just keep going to the next plan up and I keep burning through more of their credits because I love it. And one of the things I have running on a couple other properties are AI-generated blog content that then feeds into the site map and yeah, that can then go and

**AI-Driven SEO Strategy**
(10:41) and feed into Google Search Console and it's designed heavily for AI engines. It's a part of the site I almost don't want. It's on brand. I have some parameters. I'll spot check it, but it is not important for humans. I'm trying to train AI to look at my site and find the relevant content there and have that cross-link and all of that. And it's

**Base 44 and Vibe Coding Tools**
(11:06) the stuff that is created on Base 44 and it's also I put my own resources page just there. So, anything I reference, a lot of these are on there and you can get to them directly. Base 44, lovable. They're a little bit more, I call them the "what you see what you get" wizzywig version of these vibe coding tools. And so yeah, I use agents

**Super Agents and Weekly Audits**
(11:32) in some ways deep research or use them on other things but that's one of those things yeah it is part of Wix that I just set forget, it happens regular week. And there are some things, Base 44 for instance, has a new tool called Super Agents and I've set one up that I'll talk to it about my portfolio on there and it will and it can

**Automating Google Search Console Tasks**
(11:57) directly now connect and act to Google Search Console and do some this work for me where for instance it runs now a weekly audit report across Google Search Console. I don't have to do that. It can make recommendations and then I think there's that. I sent a link to the slides from Jeremiah Ayang and we can discuss that keynote as well but Jeremiah

**Agents Making Decisions**
(12:24) talked a lot about agents and web 4.0. And this idea of going from those big tasks that agents can do on their own to doing on a recurring basis to then making decisions for me. Guess what? I've been working in SEO for 22 years working at SEO agencies and dealing with all this. I know so little about the day-to-day.

**Agentic Actions: Deep Research**
(12:51) I'm not an SEO practitioner. Hell yeah. I trust B44 making some decisions on what meta tags to use on my site. I don't want to make those decisions. >> Sure. Yep. Yeah. I think you hit on some great points there. These are agentic things you're doing. These are. And why are they agentic? So deep research specifically,

**The Agentic Process**
(13:12) you align on a goal with deep research. If folks haven't used deep research, it's in every AI tool now in your little chat button. You select deep research. You align on a goal and a plan. The AI goes to complete that plan. It reasons and course corrects as it collects more information and works along the way and then produces the final output for you. So it's the

**Agentic Planning and Reasoning**
(13:31) planning and reasoning ability. The fact that you don't need to be there to guide it throughout the entire process. That is a total game changer. You're already you're already in that territory. With what you were describing, you then started to talk about scheduling things. >> Mhm. >> And so that's another piece.

**Agentic Continuum and Co-workers**
(13:50) Generally speaking, when I'm talking about agents, I always use the adjective. I say agentic. It's on this continuum, but the most agent-things are things that you don't have to tell them to go. They just work alongside of you, and they sort of feel a co-worker. So, they take some of that initial stuff, maybe the reasoning

**Scheduling AI Tasks**
(14:06) and the planning, the ability to land on a goal, and they also keep working on it over time. So, you mentioned scheduling stuff, that's the simplest way to bring your AI to life, and Claude has added some stuff recently. I actually haven't used it. I think ChatGPT has something too where you can just schedule things and a prompt or a certain killer

**Gemini's Proactive Reminders**
(14:24) prompts runs at certain times of the day that starts to feel agentic. It starts to feel a co-worker is part of your team who's contributing to tasks over time. Well, well, and even Gemini today, I'm starting a new role, this contract role alongside everything I'm doing here with AMG and I was asking Gemini about all these benefits questions

**AI-Driven Reminders**
(14:48) and it's asking me, "Do you want me to set up a reminder for when the benefits kick in for you to log into this portal and do these kinds of things?" I'm, "Hell yeah." Those little things go a very long way. And in that case, is it connecting to your calendar or something to set those reminders? Have you set up a connection?

**Gemini's Google Integration**
(15:05) >> I don't think it's doing that directly, but I do have the some of the names of all their subproducts. It's Google personal information. What do they call the >> Yeah. >> personal connection >> where it's what? Yeah. Where it's pulling in all your Google information

**Agentic Flag: Using Tools**
(15:25) across different. >> Yeah. So now I've just Google can run wild across >> Sure. All that stuff. And >> Well I ask because is that that the ability to use tools and not just read information from the outside world but actually write and create documents and manipulate and change things. That's another one of your big flags for

**AI Tools: Manipulating and Changing Information**
(15:43) Agentic. >> Oh man, that's Yeah. When it actually feels it's helping you with your work, >> Actually doing stuff. >> Yeah. >> And many of your AI tools can do this in normal chat now. If you connect MCPs or connectors or apps, they're going to get all sorts of

**Connecting AI Tools**
(16:00) different names. >> But there's different ways you can connect and some of them just take a few clicks of the mouse. You can connect your AI tools to these external connections where they can actually do things and not just read information. Soon as you start experiencing that, again, it's another key step towards agent. Well,

**Proactive AI Connection Queries**
(16:17) and one of the just odd tips I'll throw out is if you're not sure and if this conversation right now is sparking some ideas for someone here, go ask. It doesn't matter if you're using ChatGPT, Claude, Gemini, Perplexity, whatever it is, just ask it. "Hey, can you connect to Google Calendar? Can you connect to Dropbox? Can you do this?" And you might be surprised

**Unexpected AI Connections**
(16:41) even things Google Search Console I didn't expect Base 44 to have read write access to that. And this can change my work in a significant way for some my own consultant. So just even if it doesn't suggest it itself, can you connect to Stripe so it can update some of your payment things if you're charging for

**AI for Google Search Console Issues**
(17:06) webinars or ebooks or then on your website just ask it. >> What would you do with Google Search Console? >> Well, now I get updates every week of, oh, this there's indexing issues. Do this, do that, and now I'd rather just have AI scan Google Search Console for these errors. If there's anything that it can fix on its

**Automating SEO Fixes**
(17:30) own, fix it. And then if there's anything I need to know about, oh, I need to change X Y and Z in GoDaddy because I don't have a canonical URL whatever, any of this stuff or I need to go and change on some other property then just tell me I'm 10 years old, here are the three steps to do it. But increasingly yeah as it can do these things itself and

**AI Integration with Business Tools**
(17:55) update settings on some of your back end. And I'm seeing all these things that I'm not even using much, connecting to HubSpot and Salesforce and all of these things. It is. And then if you get if something changes in one record in this Google sheet, I personally think some companies Zapier are going to be less relevant

**MCPs vs. APIs Debate**
(18:20) this is all just being done through these other MCPs. That's a good point. >> Well, well, I actually do want to pick up on a point Daniel made because I'm curious, Tom, and Dev, if you want to chime in, too. But, he's saying in the chat MCPs are already dying because we're moving to APIs. Do you have an opinion on

**AI Standards in Advertising**
(18:40) that? >> Yeah, I do. So to tie back to market too, I think Terry made a point when he was talking about his AI predictions, he was saying we will be aligning on standards as an advertising industry. >> So MCP, being the AI standard for how AI should talk to each other. Then there's all these ad

**The Need for AI Standards**
(19:02) oriented standards, ADC CP and what is IAB ARF or something, the AMP one that they just announced. So it's, do we need these standards for how AIs can talk to each other? It certainly seems a good idea. And I've definitely done trainings where I've explained to people what MCPs are and how they might use them and people's eyes light up a

**Developer Perspective on MCPs**
(19:22) little bit and say, "Hey, that's a great idea." >> But when you talk to technical people or developers about these things, they actually think they're silly. At least some that I've spoken to feel a sweat. AI tools, LLMs are really really good at making sense of any sort of text that you throw at them. The idea that you need to include

**Why MCPs Can Be Flawed**
(19:40) very specific fields and parameters to make a successful hookup between two AIs is a little flawed in the first place because they're absolutely amazing at reading blobs of text formatted a bajillion different ways. So it's, well, so what are you getting then if you adhere to these standards? Well one thing you get is a bunch of more tokens than you

**Token Usage and MCPs**
(20:02) actually need. So the language of the AI talk is they're sending tokens back and forth and if you are using MCP for example, there might be extraneous information in an MCP call that you don't necessarily need for your tool to work. And if you're going to do this at tremendous scale or you're building an application, that might actually be more expensive

**Softening Stance on MCPs**
(20:20) than if you just do some direct API connection or custom coding or some different style of handshake between your AIs. So the more I've talked to developers about MCPs and other standards, the more I'm softening a little bit on them. I think Carrie might not have been right about this that they'll consolidate. I think it's possible that you end up with just

**AI's Ability to Figure Out Details**
(20:36) a mess of these and it doesn't really matter because as long as we're sending the pertinent information for a transaction to get done, the AI can figure out the details. >> Don't tools have less token usage than MCP? >> Well, tools is sort of an abstract definition. Tools is anytime an AI is going to use something to do

**MCP Framework Limitations**
(20:55) something and MCP is one language that you can use to execute the handshake in order to do a tool. >> Fair point, right? But because MCP is a framework, there's things that must be included every time and they don't always need to be included. They're just not always relevant to every handshake between the two AIs. So

**Skills Replacing MCPs**
(21:12) >> Yeah, I guess to add to this too, we've sort of replaced MCPs with skills, and skills have this function in most models where you give it a description and then the model knows when to call the skills. So you can load a lot of skills into a context because it doesn't really stuff it the same way that an MCP would. So when you use an MCP, it adds all of these instructions to the

**Skills for Context Window Management**
(21:34) prompts that you're sending, which really kills context window management. But with skills, it's almost you give it a short description of what the skill can be used for, and the model will call the skill, which might be sub agents and scripts and all kinds of things. And sometimes that includes MCP servers, but it's almost if you had labels in your toolbox that told you,

**Universal Concept of AI Skills**
(21:53) you need a screwdriver when you have flathead screws. And if you don't have flathead screws, you don't need a screwdriver. Yeah, I love that. And I think skills, let me know if you agree, Daniel, is starting to become a more of a universal concept that AI practitioners I feel all need to understand how to layer skills in AI conversation, which skills your AI

**Skills: A Cleaner Concept**
(22:12) has access to. MCPs is more a detail. I would say that once represented the flow of information but now just calling it skills I think is cleaner. To your point, they're slightly different concepts anyways, but MCP is a little bit of sausage making, I would say, for most practitioners. >> And and this to me sounds a

**Moving Beyond Acronyms**
(22:35) big net positive where we're where the more we're able to move away from the acronyms and jargon. And the more it's, okay, what are we trying to do? Oh, we're trying to update HubSpot when new information comes in and right and just automate this and trying to optimize. Yeah. Ad yield or something, all this stuff. Okay. Yeah.

**The Simplicity of AI Connections**
(23:02) Because adding more acronyms is the last thing that marketers want. >> Yeah. Totally. And a year ago I was trying to coordinate an MCP server by setting it up locally on my desktop and it was semi-technical. Now, any AI platform in the bottom left corner, you click to your settings, you click connections, and there's going to be some version of

**Easier AI Setup**
(23:20) this. It's either can read your Google Drive or can maybe control your Asana project management board. There's different flavors, and they get richer depending on the connection, but it's becoming much much simpler to set to set this stuff up. >> Yeah, it's tremendous. >> Speaking of, Jeff Green mentioned during

**Trade Desk MCP Server**
(23:40) the Trade Desk speech that there is a Trade Desk MCP server. >> That was a cool little leak that had not come out before. >> So, can you explain what this then means for Trade Desk customers and others? >> Yeah, for sure. And I've actually done this myself in Manis. Do you want me to show people what this

**Manis: Programmatic Control**
(24:02) looks on my? Maybe I'll try and pull this up as I'm talking if I can. >> Yeah, go for it. We love >> So, the idea is using a UI can be cumbersome. Using the Trade Desk, while it is supposedly a very trader friendly piece of technology, there's just all sorts of stuff you might want to do that could be better if done programmatically and not having

**AI Chat Control for Trade Desk**
(24:25) to rely on your mouse clicking around. And one way to do that could be telling your AI to do those things and allowing the AI to do that stuff programmatically. So the idea with the Trade Desk MCP is that there are people who are not logging into the Trade Desk. They are >> You're not sharing. >> I'm about sorry I've said enough as I'm

**Manis as an Agent Tool**
(24:49) talking. There are people who are not logging into the Trade Desk but rather are communicating with it programmatically via an AI chat tool. And this is something you could set up yourself for your Facebook ads account. So this tool I'm going to hop in is called Manis (MAN us). Really cool agent tool, got bought by Meta, I think 6 months or so ago. And if

**Manis Demo Setup**
(25:13) I start a fresh conversation it would just look a normal chat. And then once I kick things off, it opens this computer over here, which I'll show you in a minute. But this is >> Yeah, >> I'm so sorry to cut you off. What when does one use Manis as an agent tool versus just going and spinning up native

**When to Use Manis**
(25:29) tools on self? Native agents >> using a developer when you say native agents. >> >> Yeah. Or just setting, yeah. Setting up your own. Is the idea that as a, if we're not a developer, set up and running an agent and having it just persist and always be there similar to a cron job. Is just, I was just

**Manis vs. ChatGPT**
(25:56) trying to get a sense from you, clearly have a lot of experience here, is Manis a substitute for anything or this is >> you got to do something this to get agents going. >> Manis definitely was unique for a year ago. It was very different experience than ChatGPT. You talk to a ChatGPT or Claude, it gives you a response. You talk

**Manis's Unique Task Execution**
(26:17) to it, it gives you a response. Yeah. It plans, pulls up a computer, uses the computer to execute tasks on your behalf, uses tools, and then gives you a response. So, it felt night and day when it was released a year ago. Now, we were just talking about, tools Claude might do some of those things on the fly as you're chatting with them

**Manis for Complex Tasks**
(26:36) depending on what mode you're using and how you've set things up. So, it's a little less black and white of the answer I'll give you, but I use Manis when I need to control a computer, manipulate lots of documents, complete a complex multi-step task, something that might take 20 minutes or a half hour rather than something that's more conversational.

**Manis Meta Ads Audit Demo**
(26:55) So, the ask here was I said, "Hey, I got this old Facebook Meta Ads account, hasn't been optimized forever. Can you do an audit of it?" And I have a prompting framework here. That's why it's broken into all these different letters that we teach. And it says, "Yeah, for sure. I'll do this audit for you. I had already connected my Meta Ads account. So you see this connected

**Connecting Apps to Manis**
(27:14) apps button. So you see this Meta Ads Manager right here. I had logged in. You can log into whatever sort of apps you want to connect to your Manis account. And again in every AI platform now you have this connections or apps or whatever they call it, but you always click on the bottom left and see what you can hook your AI up to. So I hook mine up to my Meta Ads Manager.

**Manis's Programmatic Meta Connection**
(27:32) Said, "Hey, okay, sounds good." It pulls up a computer and then sometimes Manis will navigate the web in a web browser. In this case, it had a programmatic connection to Meta and so it just started running code and making API calls to Meta to get information. So, it's not very interesting to look at, but this is a coder calling Meta over and over again,

**Manis Generates Audit Report**
(27:52) assembling this information. Eventually, they put together, starts to look a structured report. Start to type out text and then creates this final document. Let's look at maybe this. Yeah, this looks good. Cool. So, it makes this document that is a complete audit of my account. So, it puts together there's two separate ad accounts within my account. There's

**Audit Report Details and Recommendations**
(28:14) eight historical campaigns. The data is all really old. The advertisers were a band I used to play in at a restaurant that I used to do work with. And it lists out the objectives for all my campaigns. And then makes best practices recommendations. What does it observe? It makes fun of me for being super outdated. Analysis

**Agentic Task Completion**
(28:37) of the legacy account structure reveals several areas where practices deviate significantly from Meta Ads best practices and it suggests all these things I can do to modernize my campaign. So agentic, because I gave it a high-level task. It went and completed the task on my behalf. It probably made some mistakes, it probably pulled some information back that wasn't right or

**Agentic Output: Verification Challenges**
(28:54) looked funny or something that. It course corrects, put it all into one place. Review it, sign, seal, and deliver. So, a much more complete product than working with chat. It also is harder to verify the work. So, here I have six pages that it produced for me in a few minutes. Double-checking this is a little harder than if it was

**New Validation Techniques for Agent Work**
(29:12) turn-based and I could look at every single piece. So, we need new validation techniques as we do agent work and we're constantly recommending people insert themselves at the right moments in time to make sure this stuff is headed in the right direction. Any questions on this? So when you're running a live campaign, it's also going to

**Live Campaign Optimization**
(29:32) make recommendations on how to improve it. I guess >> It could. When you're running a live campaign with Meta, for example, it's already optimizing itself. They've been using machine learning to tighten up the performance of Meta Ad campaigns for a long time on the micro level. Constantly looking for new users and things that. Then

**Agent for Strategy Shifts**
(29:49) there's moments in time where you might want to do a strategy shift. You might want to think hard about your reporting to date and decide if you want to update your strategy moving forward. That to me is more when you would bring out an agent and talk through it and ask it to audit your account, do reporting, things that. >> What about your landing pages? Can it

**Manis for Landing Page Analysis**
(30:10) analyze your landing pages and videos? >> Can it watch your videos? I'm not certain if Manis can watch your videos or not, but that seems likely to be something that they are trying to add. Yeah, go check out my navigate to my landing page and audit it. Yeah, that's a perfect one for it. And it'll literally pull up a computer, navigate to your

**Manis for Competitive Research**
(30:31) website to start clicking around, doing stuff, taking screenshots >> if you want. So, we do another demo. It's competitive research. Go to the Meta Ads Manager, which is a place you can get public Meta ads that are being ran, and it takes screenshots of them and does competitive research for brands in your competitive set. And then maybe analyze if your ads

**AI Assistance in Analysis**
(30:50) and your landing page are appropriate or >> Yeah, for sure. But as you're saying that, there's probably a bunch of people on this call who are really good at that sort of analysis. So I'd encourage you to get help from AI with that sort of thing. But lean in yourself. Providing lots of context about really what you're trying to do with a website, your business

**Tailored Website Recommendations**
(31:09) objectives, all that sort of stuff is what's going to be the difference between getting generic website improvement recommendations and actually getting something that's powerful for your business. Looks a question from Earl. >> Yeah, just looking at defining AI users adoption levels. >> I think there probably are four

**Four Levels of AI Adoption**
(31:27) actually. You missed the category zero though. There's people who don't actually use AI. They might be people who talk about AI. But there are lots of people who don't use that. Even at a market conference, if you really pressure test people are using AI, you might find that many have messed with it, but very few are getting returned regularly. I'd say there's

**Low AI Adoption in General Workforce**
(31:45) a bucket of zero users. There's a huge amount of people who've messed around. Some of them have had some successes, some failures. When people say, "Hey, to use AI," they say, "Yeah, I have ChatGPT." And maybe there's a couple things that they get returned from. I think we've already covered two-thirds of general knowledge workers at that point. And my

**Conservative View on AI Adoption**
(32:08) opinion is probably I'm a little more conservative than many on how I think AI adoption is rolled out. I think there's tons of opportunity for folks to ramp up. So I'd say about two-thirds of people are really not getting much return from AI. Then there's maybe of that remaining third a subset of true power users, many of them are developers.

**Power Users and Knowledge Workers**
(32:31) Maybe the vast majority of them are extremely technical developer type stuff, using it for software engineering, vibe coding, things that. And then there's this spot of 15% of knowledge workers who are using the tools that the vibe coders are using but to do knowledge work and that's people are doing things Co-work, Manis, agentic Claude. So I see

**Aim for Top Tier AI Usage**
(32:52) this is there's really four buckets. I would encourage folks to try and get in that bucket I was just describing as fast as possible. Maybe many folks on this call are already in that bucket. But some things that get you in that bucket are definitely actual return on your AI usage. So, when you hear about, yeah, go ahead. >> Well, well, this is one of the

**Defining AI ROI Beyond Time Savings**
(33:12) first things I want to ask you before we went down the agent rabbit hole here is when you're talking about returns, I'm still seeing so much about it's return on time. And some of these productivity metrics. But that to me still strikes me as a field in its adolescence at best because how do you go to the CFO, the CMO, the CEO and

**Hard vs. Soft ROI for AI**
(33:39) say I got this return on time? I saved I save four hours per week using AI and it's okay but what is that really doing for us? >> Yeah. Yeah. So what what are you seeing with what returns even mean and where is it worth even trying to get a hard ROI metric versus where do our softer metrics okay? What's

**Time Savings: A Starting Point**
(34:05) your point of view? >> I do think time savings is a great thing to target and that is when I see people's eyes light up. It's, oh, that used to take us so long now I can do this so much faster. That's great. But to your point that's only going to be exciting for so long. Your customers are not going to be very excited that you saved time doing their

**Turning Efficiency into Value**
(34:22) work. Neither are your managers. They want to see more value. So the second thing you do, once you are actually seeing concrete return and you're seeing productivity gains and efficiency, you need to turn that into value. And the simplest way to do that is just produce more useful stuff. Do more of what you were doing. Maybe do it

**Faster Turnaround and Customization**
(34:41) more regularly. If it's reporting, maybe you're the customer is hearing from you more often. And maybe you're just doing it faster. In my business where we have to create live work sessions. If someone wants a custom session this Friday, I can do it. But I couldn't do that two years ago, but now I have the right tools in place. I have the right

**Pricing Pressure and Customer Value**
(34:59) operating procedure in place. I could do this afternoon if somebody really needs a custom session is we have the right operating procedure in place to do that. So, the way we're trying to express it to our customers is faster turnaround, more customization, and then we're actually going to trying to slowly lower our pricing, too, which I think other folks

**Second Order Effects of AI**
(35:16) will see start to feel pricing pressure. It's, if I can do this for cheaper and faster, why wouldn't I? And our hope is that customers will buy more in total as a result of lowering our pricing. So, I think you're totally right. This first order effect that you feel isn't that exciting. It's all the second order effects you got to jump

**AI Enabling New Capabilities**
(35:32) into. I was going to say, don't you see improvements in where an ROI is doing things they couldn't do? People who are not doing things that now AI enables them to do. So, it's adding new capabilities. >> It is. >> Yeah, it's a great point. The product that many people who are producing with their newfound skills

**New Skills vs. Domain Expertise**
(35:54) isn't necessarily as good as the old product. If I can vibe code somebody a website, how much better is that than what they would have gotten from a developer who could also vibe code for example? So my newfound ability that creates efficiency for me doesn't make me the right domain expert necessarily to own that task over a longer

**New Opportunities for End Clients**
(36:17) time period. I was saying for the end client though, the end client thing then now has the ability to do a website where they wouldn't have paid someone. It's a new, the tool opens up new opportunity for them to do things that they would not have been able to do for themselves. >> Yeah, I agree with that. Yes. >> How to what's the extent to

**AI Adoption: Adding New Value**
(36:35) which you're seeing that happen as opposed to just the argument of you're going to save time. Maybe it's not an argument for you selling the services but from the perspective of business adoption of AI, you're adding new capabilities. >> More value. >> Yeah, I think I think it's a that's a great great focus point in my

**Delivering More Value with AI**
(36:58) business again where we create historically we've created a lot of trainings. People want Powerpoint presentations for us in the AI enabled era. You now get a lot more from us. You get one-pagers, you get glossaries, you get all more material. You get a custom micro site internally. So, yeah, I think you're right, Howard, more, I called it more stuff. But

**Unique Value and Continuous Learning**
(37:16) it's not just quantity, it's potentially more unique pieces of value that you're bringing to customers, too. So, I think you're right. Tom, I asked a question in the chat. For those of us who are in that 15% you were talking about, and I'm working hard to get to the top of that, what do you suggest? What is it? Your courses? Are there other places to go to

**Upskilling for Knowledge Workers**
(37:40) get this education? I'm teaching myself to build agents and learning a little bit of Python in the way, but I'm thinking I may not need that. But I am actively looking to get really really good at the knowledge work part of this. I'm not interested in being a coder. But I am absolutely interested. I think this is going to be the ability to take AI and translate

**Networking for AI Power Users**
(38:04) business processes into AI workflows is going to be incredibly valuable in the next 5 10 years. >> Yeah, well said. I think you're the bleeding edge if you want to get in that last group. So associating with other people in that group and sharing notes with them directly in real time as you're building stuff and figuring it out, I think is a good place

**Mastering Diverse AI Tools**
(38:26) to be. If you really want to be a power user, and pushing the limit at your org, you got to be using different tools and just wrapping your head around how one adds value versus another. And you might not be a power user in every single tool, but you're wrapping your head around the difference between a co-work GPT and understanding that.

**Operational AI Deployment Challenges**
(38:45) >> Yeah. >> There's also way more operational work to be done for deploying AI and getting value out of AI than I even realized at first. And by that I mean with my business, as we wanted to use AI more directly, thinking really hard about where we store information, what information we write down, obsessing over recording

**Centralizing Knowledge and Tools**
(39:08) every single phone call, picking one shared team's AI platform and using that and creating artifacts in that that we can share and pass back and forth with each other. That was actually a little more work than I thought. Or people have mentioned OpenClaude. OpenClaude is a really cool agentic tool. It's a live a teammate. It's

**OpenClaw and Claude Co-work Challenges**
(39:28) proactive. It can do stuff, but it's actually really hard to use. And thinking about how an org is going to use OpenClaude and Claude Co-work side by side is a problem that many people don't have a clean answer to. And trying to figure that adoption, figure out that next phase of adoption I think is going to keep you really sharp and give you a lot of useful perspective.

**Word Power's Course Structure**
(39:49) >> What about your courses? So, our courses that we give the businesses are plugged, I'll take my 50 bucks later. >> I appreciate it. Thank you. So, we usually show up and we do a few sessions worth of fundamentals because nearly every business in the world right now has lots of people who need to know how to prompt, need to know what tools are, need to know how to store

**Champion Programs and Use Cases**
(40:12) and share prompts and projects and all sorts of the basics. So, we usually do a couple of those. Many businesses, especially larger businesses, have champions programs where there's some amount of people who they really want to push to the next level and want to be driving use cases. So, we'll do parallel curriculum with them that's more focused on use cases

**Office Hours for Business Problems**
(40:29) and deploying the AI than knowing and learning the fundamentals. And then we do these office hours sessions where people bring their actual business problems and a small group will come to us. Three people from the engineering team or two people from product marketing and say, "Hey, we're trying to do this thing or we're trying to operationalize this."

**Practical Application Over Lectures**
(40:46) Help unblock us, help bring it to the next level. Yep. And that's typically what the more power users are more interested in. I don't necessarily have a 90-minute lecture that is going to help a power user get to the next level but rolling up our sleeves together and looking at their business problems, looking at the actual tools they're using. That's

**Theory vs. Application**
(41:01) usually when a lot of the goodness comes to life. >> Yeah, it's very clear here that theory and application are very different and where you need to be learning stuff is in the application part. >> Yes, I think so big time. I think that's totally right. >> Yeah. >> I had a I had a question about that.

**Context Engineering at Departmental Level**
(41:19) Putting it to use. There's folks here in the Boston area, the Kendall project, where they're trying to do some context engineering at a departmental level and every business process has a problem, a way of describing it, the people involved and they try and do that. Are you seeing a need in the

**Consistent Process Documentation**
(41:43) organizations you're talking to which is thanks for the AI literacy >> thanks for also helping us think through how to deploy it in general and yes we'd love your office hours but what we really would have is at least some if not all of the departments in the company have a consistent way of capturing processes or as context blocks so

**The Need for Clean Data and Documentation**
(42:10) that when we try and feed into whatever agent or chat interactive experience we're going to use, we all have a similar taxonomy. Is that a clear question? >> Yeah, I think it's a great point. Many organizations I work with are already deploying some strategy to clean up how they document and record things because they've had this realization

**Microsoft Copilot and Messy Data**
(42:33) already. We plugged in Microsoft Copilot to our shared SharePoint and realized everything was a mess. That is a very common shocking anecdote. Yeah. You'll hear that from lots of businesses. So many of them are already thinking to do something about it. Others that aren't, I'm ushering them along. I'm pushing them along. There's a lot of

**Best Practice: Centralizing Call Recordings**
(42:52) change in behavior. One simple thing is record all your phone calls and put them in one place so that people can harness them with AI and ask information about calls that you've had in the past. That is one best practice that many orgs are not doing. Maybe they're recording them, but are they putting them in one place? Are they making those accessible via AI? So

**Importance of SOP Documents**
(43:08) yeah, there's a set of those things, a set of best practices for getting your knowledge in order. And there's major change in behavior related to that. We advise businesses to keep SOP documents, standard operating procedure documents where you just write down how you do stuff. Most people don't do that. Most businesses don't have that. So

**Leadership Consulting on Knowledge Management**
(43:28) to answer your question, yes, many a lot of times that might already be in place and we're helping with that or when we're doing leadership consulting, we're guiding them on how to approach that. Just tell me quickly, what what do you what's your quick takeaways from the event and yeah, what would you want people to walk away with?

**AI in Advertising: Beyond Real-time Modification**
(43:51) >> The ad world has been using machine learning and AI to make campaigns better for a long time. >> Mhm. >> Even as people think about Ad CP and using MCPs for advertising, they're more about they're not about real time modifying the campaign. It's about speeding up the negotiation process, the strategy and planning

**Operational Changes in Marketing**
(44:09) process. When you hear about how agencies and publishers too are using AI, a lot of it was about efficiency and time savings and things that. So I think in the marketing world, thinking through the operational changes from AI is the stage we're on right now. Consumers are getting extremely targeted personalized ads powered by machine learning that have

**Inward Focus on AI Adoption**
(44:32) been for a long time. But how we do that work, how we set that up, how often we manage it, the intensity with which we do our check-ins and our analyses and our strategy recalibration, that is more what's about to change. >> Mhm. >> So maybe thinking a little bit inwardly was a theme that I got from the conference.

**Contacting Tom Riordan**
(44:51) >> Amazing. Well, Tom, this is awesome. What's the best way for folks to get in touch with you? >> People can drop me a line directly. I'm pretty accessible. I'm going to put my phone number in there, too. Shoot me a text. >> They may want to chat AI. This is what I do. I cover your AI. So, drop me a chat. All day. I'm

**Tom's Availability for AI Chat**
(45:11) sharing notes with people, best practices, things that. So, if I don't get back to you right away, don't >> don't hold it against me, but I'm available. >> We'll hold it against your agent for not answering the text. >> Yes. There you go. >> Awesome. Well, Tom, this may I'll

**Farewell and Thank You**
(45:23) I'll stay on for a few for anything else we want to discuss till the end of the hour, but thank you so much for joining us and hope you can come back sometime.