Podcast: Play in new window | Download
AI used to be about prompts, now it’s about delegation. Christopher Penn breaks down what agentic AI actually is, why it represents a leap forward for small businesses, and how tools like Claude Cowork are turning strategy into execution. If you’ve been curious about moving beyond copy-and-paste AI and into systems that actually get work done, this conversation is your starting point.
Stop Prompting and Start Delegating with Agentic AI
If you’ve been using AI to write captions, clean up emails, and draft blog outlines, good! You’re ahead of a lot of people. But you’re also still in phase one.
There’s a phase of AI that most small business marketers haven’t reached yet: agentic AI. It’s the part where you stop feeding the machine prompts and start handing it projects. And according to Christopher Penn, co-founder of Trust Insights and one of the sharper AI thinkers in the marketing space, it’s closer than you think.
From Copy-Paste Monk to Delegator-in-Chief
Christopher has a useful way of framing the AI evolution. He calls it done-by-you, done-with-you, done-for-you.
Done-by-you is what most people are doing right now. You type something into ChatGPT. You copy the output. You paste it somewhere. You’re the monk.
Done-with-you is when tools like GPTs, Zapier, or Make start automating parts of the process. You’re still involved, but some of it runs without you.
Done-for-you is agentic AI. You hand the tool a project plan, and it executes—start to finish—then comes back and says here’s your thing.
As a real example: Christopher’s CEO, Katie, needed a slide deck. She handed the task to Claude Cowork with their brand standards and an outline. Forty-five minutes later, she had a finished PowerPoint. Complete with logos, on-brand formatting, and bonus suggestions the AI added on its own.
“We both looked at it,” Christopher said, “and went, holy crap, this thing is really good.“
The Tools That Make This Possible (Right Now)
You don’t need a custom AI build or a developer on staff. If you’re already paying for any major AI platform, you likely have access to agentic tools today.
Claude Cowork is built into Claude’s desktop app (Mac and Windows) on any paid plan. OpenAI’s Codex does similar work in the ChatGPT ecosystem. Google’s Agentic tools (like Google Anti-Gravity) work the same way in the Gemini world. And if you want complete flexibility, free tools like Microsoft Visual Studio Code with an add-on like Klein or Kilo can connect to almost any AI platform.
One important note: Christopher is emphatic about this. If you’re using the free tier of any AI tool for your business, stop. Free versions typically opt you into data training by default. Pay for the plan. It’s worth it.
The Framework That Keeps AI from Going Off the Rails
You may have heard some Agentic AI horror stories: someone hands something to an AI agent without clear instructions and then wonder why the output is garbage or why the tool deleted all of their emails.
Christopher’s answer is the 5P Framework, developed by his colleague Katie Robbert at Trust Insights:
- Purpose– What are you trying to accomplish, and why?
- People– Who’s involved? Who is this for?
- Process– What are the specific steps to get there?
- Platform– What tools are you using?
- Performance– How will you know it worked?
That last one—performance—is the most important. Agentic AI behaves when it has clear, measurable success criteria. “This blog post must be exactly 800 words, use active voice only, and contain no adverbs.” That’s a success condition the AI can check against. Vague instructions lead to vague results.
Christopher uses this in his own coding work: “Success is 100% test coverage, 100% unit tests passing, 100% integration tests passing. You are not done until all three are met.” The agent keeps working until it hits those marks.
This same logic applies to any marketing task—landing page optimization, email strategy, competitive analysis, quarterly planning. Define what “done” looks like, and the agent will work toward it.
Your Data: More Exposed Than You Think
One of the more sobering moments in our conversation was about data privacy. A lot of business owners hesitate to connect AI tools to their calendar, email, or CRM. The concern is valid. But Christopher’s points out that your data is already with Big Tech. Your email is in Gmail or Outlook. Your docs are in Microsoft 365 or Google Drive. Your accounting is in QuickBooks Online.
“You’ve already forfeited your privacy,” he said. “It’s just a question of which big tech companies have that data.”
That doesn’t mean you hand over everything without thinking. But it does mean the hesitation shouldn’t be about whether to use these tools—it should be about which ones have reasonable privacy terms.
His quick tip: copy and paste any platform’s terms of service into your AI tool and ask it to score how safe your data is—one to ten, across whatever criteria matter to you. Let the AI do the legal parsing. That’s exactly the kind of task these tools are built for.
If you’re dealing with genuinely sensitive data, there are fully local AI models (like Alibaba’s Qwen) that can run entirely on a high-spec MacBook with no internet connection required.
What Your Role Actually Becomes
Here’s the bigger picture. As agentic tools get better—and they’re getting better fast, partly because AI companies are using AI to build smarter AI—your job shifts.
You stop being the person who does the work and start being the person who designs the work. The project manager. The delegator. The one who asks the right questions and sets the right success conditions.
“What things have you deferred as a business owner that you’re just not making progress on?” Christopher asked. “How would you hand that off to a machine?”
His own example: he has a newsletter with nearly 300,000 subscribers that wasn’t earning what it should. He gave Claude Cowork a 5P plan and told it to evaluate the monetization strategy. It came back with 82 specific things he wasn’t doing. He’s been working through the list ever since.
Where to Start This Week
Christopher’s recommendation is simple: use the voice memos app on your phone. Talk through your current business situation—what’s working, what isn’t, what you’re stuck on. Those apps transcribe automatically. Drop that transcript into an agentic tool with a 5P framework and ask: What are the three to five things I could do Monday morning to start making more money?
By end of day, you could have a plan.
That’s the promise of agentic AI—not magic, but a serious upgrade in how you delegate and what you’re able to get done. The marketers and business owners who learn to hand off work clearly are going to have a significant edge over those who are still copying and pasting.
Transcript from Christopher Penn’s Episode
Rich: My next guest is an internationally renowned keynote speaker who energizes audiences with his expertise on artificial intelligence and marketing. A recognized thought leader, he demystifies complex AI concepts, providing practical strategies that empower marketers to achieve breakthrough results.
His journey in AI began in 2013, a full decade before tools like ChatGPT brought the technology into the mainstream, giving him a deep understanding of its potential and practical applications.
As co-founder and chief data scientist of Trust Insights, he leads a team of experts dedicated to helping businesses unlock the full potential of their data using cutting edge AI solutions. His proven track record of success includes building AI powered systems that have delivered double digit ROI for prominent brands you may have heard of, Cisco Systems, T-Mobile, Citrix Systems, GoDaddy, AAA, McDonald’s, Twitter, and many others.
He’s a 2025 IBM champion in IBM data and AI, and co-host of the award-winning Marketing Over Coffee marketing podcast. He’s also a published author in the Journal of Applied Marketing Analytics, and his work is also reflected in over two dozen marketing books, including the bestselling, Almost Timeless: 48 Foundation Principles of Generative AI, which I am enjoying right now.
He didn’t even mention in his bio his great newsletter, which I was just complimenting him on, Almost Timely. But today we are going to be exploring the world of agentic AI with Christopher Penn. Chris, welcome to the podcast.
Chris: Thank you for having me.
Rich: Alright, so you’ve been playing around with AI longer than most of us have even been thinking about it. What was it that you found so interesting when most of us thought of it as sci-fi?
Chris: So my AI journey started in 2013 when I was working at a public relations firm. And we had to try to figure out how do you measure the impact of PR, given that a lot of the time there’s no click stream. Like from here to here to conversion. And the PR folks I worked with, very nice people, but mathematics was not their strong point.
So what it came up with was we had to start looking outside of regular marketing and saying, how do other disciplines do this? And one of the disciplines that did that was a discipline called bioinformatics, where they have to look at a bunch of different measures in the human body when you administer a medicine, a treatment, and say what’s the statistical relationship with all these other things like your heart rate and your blood pressure and your blood sugar levels, et cetera, versus before and after when this thing happened, controlling for other stuff?
Well, those bioinformatics concepts were something you could apply to public relations. And so that was how I got started in AI by saying, can we use machine learning? Can we use bioinformatics technology to measure public relations? Which is kind of like using a jet airplane to go to the grocery store. And it worked. We were able to say like, this is the uplift accounting for everything else that was going on. This is the uplift that you got from this particular piece of press coverage.
Rich: Wow. I know a lot of PR people who still are unsure how to measure their impact, so they would probably want to check that out.
Now, I asked you to come on this podcast and I said, what do you want to talk about? Because I enjoy your newsletter and I’m reading your book, and I’ve just enjoyed seeing you present at the different conferences we’ve both been at. You suggested agentic AI. Why do you think that’s so critical to businesses right now?
Chris: Agentic AI is kind of a time skip, and here’s what I mean. If you look at the evolution of internet access, Africa in particular kind of jumped the line. What they did was they did not build out a large, wired infrastructure, they didn’t have the budget to do it. And so they kind of skipped right to 4G and 5G going straight to wireless. And so if you go to many different African communities, particularly out in more rural areas, everybody is just using these. They’re not plugging into ethernet jacks and stuff. They leapfrogged.
Agentic AI is one of those opportunities in the evolution of this technology where if you think about the evolution, November 2022 is when ChatGPT came out. And it was dumb as a bag of hammers, but people could understand what it did. January 2023 is when GPT 4.0 came out. And it was smart, it was still kind of dumb as a bag of hammers, and for a couple years people’s use of AI was, I’m going to talk to ChatGPT and I’m going to ask this, that and the other thing. And we’ve seen plenty of our friends and cohorts and stuff say, just ChatGPT it and we get these amazing results.
Over time, the technology started to evolve. And I use product market fit as a framework for this. So you have done by you, done with you, done for you. Our mutual friend, Brooke Sellas, talks about this a lot in terms of customer care. Done by you is you being the copy/paste monkey, copy and pasting things in and out of ChatGPT. Done with you is when you start getting things like GPTs, and Gems, N8N, Make, Zapier, where the system is starting to do stuff with you, like parts of it are automated, parts of it are not. You’re still doing a lot of copy/paste.
But agentic has done for you where you give the tool something like a project plan, you hand it off and it just goes and does its thing and comes back to you and says, “Hey, here’s the thing that you asked for.” And for a lot of business owners, you can almost get some foundational learning about done by you, done with you, but then you can move straight into done for you. How do I use an agentic system like Claude Cowork to do this thing for me?
Yesterday, my CEO Katie Robbert, was like, “I got to do another slide deck for another thing.” She handed it to Claude Cowork and said, “Here’s our brand standard, here’s my outline. Make me the slide deck.” And 45 minutes later, Claude Cowork says, here’s your slide deck. And we both looked at it and went, holy crap, this thing is really good. Brand standards are spot on. The deck is coherent. It actually added stuff because the opus model is very smart. It said, hey, you could have used some more transitions and stuff. Here’s some more examples so I put them in for you. And we’re like, okay, cool. But that’s done for you. And that’s where agentic is today.
So if you think about this, anything that is repetitive, anything that is automated, anything that is boring and is templated, as a business leader, you should be saying, “How can I take this template, boring, repetitive task, and just have the machine do the entirely?” Oh, I got to do end of month reconciliation in QuickBooks, how do I tie QuickBooks into Claude Cowork to say, you do it.
Rich: And it goes ahead and does it. So Claude Cowork, so this is an example of agentic AI. How does Claude Cowork differ from the Claude that I’m getting at $20 a month?
Chris: So Cowork is, there’s a concept of what’s called a ‘harness’, which is the surrounding parts around a model. Think of the model like a car engine. The car engine is very useful, very powerful. Your car doesn’t go anywhere without it, but nobody goes down the road sitting on an engine block. You sit in a car. A harness is basically the rest of the car around the engine. So when you use Claude the regular one, that’s a specific kind of harness, that is a specific kind of car. You do a lot of driving.
Claude Cowork is a different kind of harness. It’s based on their coding tools. Actually, it was inspired by their coding tool. And what it does is it’s more like a self-driving car, where you give it the destination and it makes a lot of the decisions and kind of just prompts itself to try and accomplish those goals.
And so that’s the difference is regular Claude is a car you drive. Claude Coworker is a self-driving car.
Rich: And what are some of the other tools out there? If people aren’t using Claude, what are they able to upgrade to, or what are some of the common platforms you’re seeing out there that you think are good, especially for small to medium sized businesses that maybe don’t have all of the AI expertise?
Chris: So if you’re paying on any paid plan, Claude’s paid plan, the Pro Plan, you can get some usage out of Claude Cowork. It’s built into the desktop app and it’s available for Mac and Windows.
If you’re using ChatGPT and you’re paying… I’ll put this out here. If you’re using the free version, you get none of this. A free version of anything. You get all this, and you should never be using a free version with your business anyway, because you generally can’t opt out of giving your data to the company to train on it. So please, if you are using any AI for your business, be using a paying version, hands down, in the open AI ecosystem.
If you are using ChatGPT, OpenAI has a desktop application called Codex, which is technically a coding tool. However, you can use it in a very similar way as Claude Cowork. You can give it non-coding desks. You can say, here’s what I want to build for a marketing strategy, go off and help me do this thing. And it will go off and do that thing.
If you’re using the Gemini ecosystem, there’s a lot of options. But the one that’s probably the most sensible is a tool called Google Anti-Gravity. Again, it’s a coding tool. You don’t have to code, you never have to write a single line of code. You just give it complex tasks, and it will fire up its own agents internally to try and figure this out.
And so almost every major manufacturer, so if you use Alibaba Qwen-Code will do that. And there’s UIs for it. I think the only major one that doesn’t have a coding environment dedicated to is DeepSeek, but even there if you install the completely free Microsoft Visual Studio Code you can add an add-on like Cline or Kilo code that can connect to pretty much anyone’s AI and again, use it in as an agent.
Again, you don’t have to code, you never have to write any code with these tools. They’re just language models. You can use them for anything a language model does, but they’re very, very capable.
Rich: So before I dig into some of the things that you just said, I want to go back for a second. Because you were talking about how Claude Cowork helped Katie develop this slide deck. What was the final product like?
And I ask this because my experience using LLMs is they’ll often say, “Oh, would you like this in a PowerPoint presentation?” And I say, yes. And then it says, “Oh, I can’t actually do that for you.” Was it a PowerPoint or a keynote presentation that then she could work with within the program itself?
Chris: Yes. Yep. It was a genuine PowerPoint slide deck with our brand standards, with our logos, et cetera. All that stuff, it was a regular deck. And I actually just finished recording a webinar for our academy I did a previous version of this yesterday for a different company. I said, okay, here’s my existing slide deck that was for this particular industry, generalize it and also kind of fix up my slides. Because they were kind of janky. And Cowork is like, great. I’m going to get started on this. I have this list of 20 questions for you to answer. I answered all the questions, and it often went 45 minutes later and said your slide deck’s done. I look at it and go, that is definitely better than what I did yesterday.
Rich: All right, so for those of us who grew up on Skynet and nervous about giving AI too much autonomy, how can we approach agentic AI? What are the guardrails for keeping AI from going off the rails?
Chris: Andrej Karpathy, one of the co-founders of OpenAI, said in 2023, “The hottest programming language is English.” In 2026, the hottest programming language is project management skills. How would you delegate, how would you build a project plan for whatever you’re trying to do? Forget about AI, how would you build a good project plan to do something like revise this slide deck, or build me a marketing strategy, or help me increase my sales by 10%? What kind of project plan would you put together?
Katie calls this the 5P framework. The trust in size 5P for purpose, people, process, platform, performance. Which by the way is the perfect project plan. That’s a lot of P’s in one sentence.
Purpose, what are you trying to do and why? People, who’s doing it and who are you doing it to? Process, how do you do the thing? Platform, what do you do in the thin. And performance, how do you know you did the thing?
And if you think about it, this applies to everything. If you are making breakfast, purpose – you’re hungry, you want breakfast. People – who’s doing it? You are. Who’s who? You doing it to you because you’re making . Process – what are you making? How are you going to make it? I’m going to make an omelet. Great. How do you make an omelet? Mix the eggs, et cetera, et cetera. You know, put the stuff in the pan. Platform – use a non-stick pan unless you hate your life. And performance – did you end up with something edible?
If you build out really solid project plans, you give it to an agentic system, and it will execute the plan. The two things that are most important are the purpose and the performance. If you say this is what success looks like, this is how we define success, then as the agent runs autonomously, can keep referring back to its plan and go, have I met the success conditions? Yes, no. Yes, no.
For example, when I’m writing code, I will say success is 100% test coverage, 100% unit test passing, 100% integration test passing. That is success. You are not done until all three of those measures are met. And so when it’s working, it’s spinning over here in the little window. You will see in the chat. In 98% test coverage passing that says, but the success criteria is 100%. I’m going to go back and keep trying again. AI goes off the rails when it does not have clear objective, very deterministic outcomes.
This blog post must be 800 words, may not use passive voice, and it must contain no adverbs. If you give it success criteria like that, it will work until it figures out. Here’s the thing, you want it to be exactly 800 words. You want to have no adverbs. You want to have active voice only. Here’s your blog post.
What this forces you to do as the operators, the project planner, the project manager, is think very clearly, what is success? And if you don’t know that, AI is 100% going to go off the rails on you. And it more often than not is just going to create kind of garbage. But it can misbehave, right?
If you use a tool like Claude Code on your computer and you don’t know what you’re doing, you just say, “Hey, here’s access to my hard drive, unlimited”, you’re going to have a real bad time. So purpose, people, process, platform, performance is to make sure that you are dialed in, particularly on performance, on success, and the tools will behave themselves.
Rich: Alright. And I’m sure when you’re working with the agentic AI, is it a similar experience to the way that we’re currently using LLMs. Like are you just having conversations and uploading documents and going back and forth with it, or is the experience different in your opinion?
Chris: The experience is much different because you’re handing things off. So if you think about an agentic AI, it sounds so fancy, it sounds mysterious. It’s no different than a travel agent, right? Or a tax agent. Or a real estate agent. You give someone a task, they go off and do it, and they come back and say, “Hey, I sold your house”, or “I filed your taxes”, or “I booked you on this cruise with this thing that you said you wanted to do.” That’s what an agent is, it just goes off and does things.
So when I’m using agentic AI, for example, in some cases, yes, it’s interactive. Like I have my Claude Code wired to connect to my Asana, my calendar, my email, my HubSpot, and my Jira task board. And so every day the first thing I do when I come in the morning is I open up my Claude Code and say, “Give me my daily briefing”, and the agents go off and do their thing, gather all the information, say ‘here’s your briefing’. That’s an example of a more interactive one. But I don’t have to sit there and prompt it because I’ve built the agents within the system to go and do that thing.
Other times, like I’m in the midst right now of doing a data processing thing. I gave it, “here’s the project plan with the five Ps. Go.” And in the beginning it will ask some clarifying questions, and then it just goes and it will ping a little notification when it’s done and say, “Hey, I’m done. Here’s your thing.”
Rich: So as I’m thinking about this, it sounds wonderful. But there’s this still part of me that’s like, okay, I’m giving it access to my calendar. I’m giving it access to my emails. These giant tech corporations that make so much money off of data gathering are now in my grill. That’s where I start to get some concerns, and I’m sure some other people listening may have some of those same concerns.
What’s the progression like from being really careful about the data you’re sharing with these tools, to where you do feel like you can trust them, at least with specific tasks?
Chris: I’ll point out this, a big tech company already has all your data. Where do you think your calendar lives? Where do you think your email lives? Where do you think your word documents and stuff live in Microsoft 365 and stuff? You already don’t have control of this information. Your software is in QuickBooks, which is online, right? Your accounting’s there, your legal documents are in DocuSign. You already have forfeited your privacy to big tech companies. It’s just a question of which big tech companies have that data.
When it comes to AI, one of the first most obvious things you should do, and you can use AI to do this, is review the terms of service and say, “before I click pay, what are the terms of service on this?” And if you’re big enough, you might even be able to say like, you know what, I want a separate contract.
Almost every service that’s paid has what I would call commercially reasonable privacy, which means that your data is mostly safe, most of the time. The exception across the board, without exception, is lawful requests from the government. The companies will just hand over your data. So if there are things that you would prefer never be exposed, there’s what’s called local AI. These are AI models that are smaller. They’re dumber, they require a lot more handholding, but you can run them completely on your computer if your computer’s nice enough, and unplug the internet and use them.
So an example of that would be Alibaba’s Qwen3 next models. If you’ve got a really nice MacBook, like you paid top dollar for the nicest MacBook you could get, you can run that entire model, self-contained, still give it access to all your data, but your data never leaves to go anywhere else because the model’s running on your computer.
Rich: And I’ve seen you talk about this before, but what you’re saying is it could still be agentic AI, but you’re just running it locally, everything stays locally. This obviously changes once you need it to go outside of the boundaries of your laptop and get some information from the Cloud, say. But other than that, you’re pretty much locked down.
Chris: Exactly, you’re completely locked down. And even there I wrote, for example, a piece of code that uses DuckDuckGo as its search engine. DuckDuckGo is famous for its privacy, they don’t track what you search for. So I connected my local AI to DuckDuckGo, and so now it can go and safely search the web. I have it connected to a VPN, to tour, and a few other privacy safeguards, so that if I’m doing something that certain government officials might be very interested in, it would be very difficult, not impossible, but very difficult to trace that back.
Like, I want to know so and so’s recipe for strawberry pie, and that’s a classified recipe. The agents can go figure out how to do that with safe tools.
Rich: All right, so for those of us who did not get into business to wrangle with AI on a daily basis, how can we evaluate the privacy policies on these AI platforms? Because it does seem daunting to read through 87 pages of terms and conditions to find out if this is a morally ambiguous or just a morally corrupt company that we’re getting into bed with.
Chris: Easiest way is literally to copy and paste that policy into your favorite AI tool and give it a list of criteria that you want to know about. How safe is my data? When will my data be handed over? What’s good about this privacy policy? What’s vague? What could be weaponized against me?
Ask the questions that you have concerns about and have the model evaluated and say, score each of these questions one to 10, where 10 is I’m very safe and one is, hey, here’s my data everybody, and have it evaluated. These tools are extremely capable of language processing, so give them language tasks, and that is a perfect example.
Rich: All right. Now, all of this sounds incredibly exciting. And just the idea that, although I still love doing my own slide deck, so that’s probably not going away, but for other things, I can imagine that I’d want to get off my plate it sounds incredibly exciting.
But if we’re running a small business and we don’t have a dedicated IT person or a big tech budget, how approachable is this really today? Should we try our luck today, or should we wait for tools that are maybe easier for the mere mortal business owner or marketer to use?
Chris: I would say if you want to get into agentic AI, the easiest starting point for the non-technical person is Claude Cowork, which is part of the $20 a month Claude subscription. You install Claude Desktop, it’s available for Mac and windows. The Cowork window is there and then you install. There 11 built in plugins that you can turn on, such as marketing, sales, legal operations, data analysis, et cetera.
And these are just little add-ons that are free. They’re included with your subscription, and you could start using them and they’re very, very good. So you might say, okay, startup Claude Cowork, and say, “Here’s the homepage of my website. Here’s who my customer is. Claude Cowork, use your marketing plugin and tell me is my site any good? Am I appealing to my customers?” Give it a landing page, turn on the sales module and say, “Is my page going to sell to my customer, or what should be better to make this landing page stronger, to help me convert more?”
Think about the things that are real business problems and using one of these very straightforward agentic framework you’d give it a 5P plan. Purpose. I want my sales landing page to convert 20% higher people. Here’s who my ideal customer is that I’m trying to attract.
Process, you’re going to read through my landing page, you’re going to compare it to your known best practices. Maybe go search the web for best practices for my kind of company.
Platform, obviously it’s called Cowork. And then performance, give me three to five options to improve my landing page’s conversion rate by 20% using your sales tools. And it will go off and do that thing. It will come back with your analysis and then you, human, have to read it, think it through, and go, is this reasonable? Is this something I haven’t tried? And the more information you can provide it, the better it’ll perform.
One of the most powerful tools you have is right here in your hand. It is literally the voice memos app on your phone. You would sit down and say, alright, listen up, here’s the situation. Our sales were down last quarter, the market’s changing. I’m paying 82% tariffs on my thing, and I need to find maybe a different couple places to import from. I got to do this, I got to do this. Here’s all my problems.
And you take all these voice memos, apps, transcribe, they will produce a transcript. You put that transcript into Claude Cowork with your 5P plan so that it has all the background about your business, and you will get great specific results that are dedicated and tailored to you. That’s if I was a business owner.
If I was a solo business owner, my first stop is I would say, here’s the current state of my business. Here’s my goals. Here’s all as much information as I can throw at you as possible. Let’s evaluate the strategy of my business, and what are the three to five things that I could do Monday morning to start making more money?
Rich: Awesome. So as you see it, as these syngen tools become more capable and more accessible, how do you see the day-to-day role of a marketer or a business owner evolving in the next, I was going to say five years, but let’s be honest, six months?
Chris: One of the things to know about AI, and this is true across the board, is the reason this is accelerating so fast is because all the AI companies are using their AI tools to make smarter AI, right? So it is a feedback loop that is getting faster and faster. The Chinese company DeepSeek at the end of 2025 built an AI model that only does math and it’s self-learning. So as it solves problems it trains on it, and then writes new problems for itself to solve. It solves those, trains on it, et cetera. And so it is working on becoming sort of like the world’s smartest mathematician because whoever has the best math, has the best AI.
What this means for business owners and stuff is we have to take a step back from operations because as agents get more capable, they’ll be able to take on more and more complex tasks. There’s in fact even benchmarks like GDP, Val and RLI – Remote Labor Index – that measure the capabilities of AI tools to do very sophisticated tasks. Like, can you take this blueprint and do a 3D rendering of the house that this blueprint describes? Can you do this geospatial analysis? You know, very expensive projects.
So we have to take a step back from doing the thing to say this tool, like Claude Cowork with its 11 agents, this tool basically gave me a team of 11 employees that are PhDs in these topics. How do I, as a business owner, what do I do with 11 new employees that are all PhDs in their respective areas? What things have you deferred as a business owner or as a senior leader that you’re like, ugh, this has been on my to-do list forever, and I’m just not making any progress. How would you hand that off to a machine? How would you delegate that?
The people who are going to be best at agentic AI are the people who are the best at delegating. So if you are skillful at delegating a task, you are going to be very skillful at working with agentic AI.
Rich: That makes me want to go and improve my delegation skills today. Alright, I’m feeling empowered. If I had to take a step on my agentic AI journey this week, where would you have me start?
Chris: Drop $20 bucks on Claude and install Claude Cowork. Install those 11 plugins and pick the one of those 11 plugins that your business is struggling the most with. Maybe it’s sales, maybe it’s operations, maybe it’s finance and expenses, whatever the thing is. And say, “Here’s the situation. Let’s evaluate my business in this particular area and identify…” do the four-part framework. So there’s a four-part framework. What’s good, what’s bad, what’s missing, what’s unnecessary? So whatever the task is, talks to the agent, give it that project plan. You’re going to do an audit. What’s good about my cashflow? What’s bad about my cashflow? What’s missing from my cashflow analysis? What’s unnecessary in my books? And have the tool do the evaluation. And then say, okay, now build me a strategy and a work plan to start making the remediations of the things that I say are issues.
That’s what you should do today. And you could have, by the end of the day, you could have that plan in hand. I did this recently with my own newsletter. I said, I know I could be earning more money from this newsletter. I’ve got almost 300,000 subscribers and I’m earning five digits instead of seven digits on it. And I handed it off. I said, do the evaluation. It’s like, okay, so here’s the 82 things that you’re just not doing that you really should be doing because there’s easy money to be made here.
And I’m like, well, shit. And then I read the plan and I’m like, okay, I’ve got the plan and now I’m starting to follow the plan. And even in cases where there’s something like, I don’t know if I want to do that. Not that it’s bad, but just like, that’s a lot of effort. “Claude, can you do this part for me?” And it’s like, yeah, I can give that crack because among other things, it can take control of your browser if you give it permission to, and you can turn it on and off, that permission, at any time.
So there are things like, you know what? I don’t feel like signing up for the service. Go sign up for it. I’ll just watch you do it. And running in another window and then we’ll just go off and do the thing. It will sometimes say, “I don’t have this information. Can you give it to me?” And it will do the thing.
But the thing I would say as a message to every business leader, is this is the excuse eliminator. This is the excuse eliminator. If you give it the problem, it comes up with the solutions. You agree with the solution because you just don’t let it run amok. And then you say, build me a work plan to fix this. You are out of excuses now because it literally did . It said, “here is the work plan to get you to your goals.” So if you do not get to your goals, it is now entirely your fault that you did not execute the plan.
Rich: Alright. Well, that puts a lot of work on my plate, or at least a lot of emotional work on my plate to actually get these things done.
Chris, this has been amazing. Where can we find out more about you and Trust Insights?
Chris: So the easiest place to start is trustinsights.ai. That is our company and you can find all the stuff there. And if you want to find me personally, christopherspenn.com. And those are the two places that will not be presumably taken over by hostile billionaires anytime soon. Although, if you are a hostile billionaire that’d like to give me billions of dollars, visit my contact form.
Rich: Exactly. He can be bought. Alright, thank you so much. I really appreciate today’s conversation.
Chris: Thanks for having me.
Show Notes:
Christopher Penn is the Co-Founder and Chief Data Scientist at Trust Insights, where he helps businesses use AI and data to make smarter decisions and drive measurable results. He’s an internationally recognized speaker and co-host of the Marketing Over Coffee podcast. Learn more at trustinsights.ai, visit christopherspenn.com, or explore his book, Almost Timeless: 48 Foundation Principles of Generative AI.
Rich Brooks is the President of flyte new media, a web design & digital marketing agency in Portland, Maine, and founder of the Agents of Change. He’s passionate about helping small businesses grow online and has put his nearly 30 years of experience into the book, The Lead Machine: The Small Business Guide to Digital Marketing.