Listen on Spotify with Jason VanDevere
The short version
- I have been doing automation in some form for twenty five years, from a Hotmail misspell in 2001 to Infusionsoft funnels to HubSpot to fully autonomous agents
- Two levels of tools right now: OpenAI and Claude for almost anyone, OpenClaw and Hermes if you want dynamic learning and full control
- I treat AI like employees, and the difference is humans will leave you and AI won't
- My agent Luna says up front that she is my AI assistant, because I do not think we should hide AI
- Grok is the sleeper, it has access to the X API natively and it prepped me for this show
- The point of the time back is impact and presence, not stuffing more work into the week
Twenty five years of this, starting with a misspelled Hotmail
When Jason asked why AI, I had to go back further than most people expect. I am 44. Back in like 2001, so like twenty five years ago, I started dabbling with automations. My first dabble was owning the misspell of Hotmail. Back then Hotmail was the premier email address, and this was the time when people just typed, so if you meant to send to a friend at hotmail.com and you typed hotmial instead, that email came to me.
I intercepted those emails, corrected the address, forwarded them along and spoofed them so the sender never had a clue anything went wrong. Your message got delivered exactly the way you wanted it delivered. In the middle of that I scraped all the data out of it and we became a data warehouse. Jason called it kinda freaky. It is. But it was the first real automation I ever built.
Around the same time I was helping take an ad agency into an automated environment. They were billing off estimates and not tracking actuals, so I started tracking how many times the press ran and how much ink we used. Fast forward to about 2014 and I got into the Infusionsoft craze. We built Jeff Walker's very first marketing automation funnel. Then we moved into HubSpot, became their rookie agency of the year and a platinum partner, and onboarded Michael Hyatt and his company.
The HubSpot work is where the logic started to look like what AI does now. If you read three blogs about focus for Michael Hyatt and his Full Focus journal, we went and did the research to figure out whether you had bought a journal. If you had, great, keep reading. If you had not, you entered a sales cycle to get you the journal you were clearly interested in. That was hard lined. Now I can say make the best next decision for this person, and it can kind of think on its own.
I delivered what you wanted to deliver and in the middle of that I scraped all the data out of it.
What agents are already doing in my business
For one client we deployed over a 100 voice agents inside physical therapy clinics across The US and UK, replacing the front desk phone person with an AI agent instead of a human. That grew about $2,000,000 in ARR out of that company. That is the big, visible version. The version that matters more day to day is smaller and quieter.
- An agent that keeps my inbox at zero, which is where I like to live even though plenty of people are sitting at inbox 32,000
- A dashboard of revenue, market cap and the KPIs I care about in our community, kept updated by AI
- An agent that handles level one and level two support tickets, with anything extensive or elaborate reviewed manually
- An agent that does podcast outreach, which is the only reason Jason and I ever spoke
Here is the exercise I gave Jason for anyone starting from scratch. Make a list of everything you are doing on a daily basis, the routine and the mundane. Then make a separate list of the things you should be doing but are not. You do not like them, you do not have the time, you do not see high value in them. Then start looking at both lists and say AI can step in in almost every single one of these instances. We all have those things.
Two levels of tools, and how to know which one you are
At a novice level there are two major pieces of software people use, OpenAI which is ChatGPT, and Claude. OpenAI has Codex. Claude has Cowork or Claude Code. Both have connectors into a lot of things across your computer and your apps, a Google connector if you use Gmail, connectors to your calendars. AI has now become goal based, so you can say clean my inbox and keep it clean for me, answer a few questions about what to flag and how to handle starred items, and it goes and does it on a schedule or in loops.
A step deeper you have OpenClaw and Hermes, and they are built to run fully autonomous without your interaction. Same goal based focus, but Claude and OpenAI are pretty static in their learning right now, they do what you tell them based on your guidance and constraints. Hermes and OpenClaw are dynamic and constantly in a learning loop to improve based on your goals. The tradeoff is that you have to manage some things on your own, do some installs, and you will probably end up in a command line interface at some point.
I told Jason it is a bit like PCs. Some of us build our own gaming PC and assemble every piece. Some of us go to Best Buy and buy the best gaming PC we can buy. So you have to know yourself. Are you a tinkerer, do you like the technical stuff, do you want to hit walls and bust through them for more power and more opportunity? Or do you want simple and easy?
For most entrepreneurs the answer sits in the middle. Use Claude and ChatGPT yourself, then hire an AI implementation specialist to build the custom agent and honestly manage that team member. So when nothing is working you go back to a human and ask how many we reached out to and what the feedback looks like, instead of staring at black and white text on a screen that does not make sense to the average person. We had online business managers, we had virtual assistants, and now most people need an integration specialist to triage that middle piece.
How the podcast agent actually got built
I told Hermes I want to get booked on podcasts. It came back with a plan. It did the research, told me who I am, the type of shows I should be on, the type of content I am prepared to speak on, and then it told me what access it needed. I helped it get that access. I was its assistant, very much like when you hire a new employee and they ask how to get into Slack. You have to help them get into Slack.
I still had to know what a quality pitch looks like. I do not think you can hire and train an employee if you do not know what that employee is supposed to do. So I said we need to go read the transcripts of their existing shows, ask me every question you can to get my background, go listen to every other interview I have ever done, and then write personalized emails to every single person you outreach to. Most software writes a standard pitch that pitches me to you but never pitches what we can create together.
Then there is the learning loop. When we get booked, go back and figure out why. What type of show was it, who were their guests. One improvement it proposed to me this morning makes total sense. If we are confirmed on a show, look at every other guest who has been on it, find every other podcast those guests appeared on, and reach out to those too. Jason had Chandler Bold on recently. Chandler and I have different businesses but we talk about some of the same topics, so a show that had Chandler probably should know I exist. Maybe they want me, maybe they do not, but it is a better chance than not.
It also hits walls like a person does. Some of our emails were bouncing back as invalid, so it came to me and asked for an API key so it could verify addresses before sending. That is an employee saying I hit a problem, can we solve it together. We solved it and got fewer bounces.
I don't think you can hire an employee and train an employee if you don't know what that employee is supposed to do. And I treat AI like employees.
Do not hide the AI
Jason gets two to four to five podcast pitches a day and his eyes are trained to spot an AI pitch. He named the pattern without naming the software, all the same style, a bunch of fluff about how Tom is a great storyteller who will captivate your audience with insightful insights, and you finish it still not knowing who Tom is.
My pitch does not try to pass. It says this is Luna, I am Gary's AI assistant. I believe wholeheartedly we should not try to hide AI, we should be forward with it. Luna has a personality, a voice, and she can make phone calls. I can tell Luna to go make dinner reservations for me tonight and she will call and make them. I have spent time with her and worked with her the way you would with a team member.
That is the whole frame. I treat AI like I would a human and I spend time training and teaching and working with it the same way. The difference is humans will leave you and AI won't. You train someone, they get amazing, then they take the next opportunity or start their own company and you begin the entire process again. Every hour I invest in AI compounds, it only gets smarter. I am going to have the smartest podcast outreach that exists in the world in the next six months.
I treat AI like I would a human. And I spend time training and teaching and working with AI the same way I would a human. The difference is humans will leave you and AI won't.
The sleeper tool nobody is using
If you focus on Claude and OpenAI's models you are ahead of the curve for most of what you need. The one sleeper most people skip is Grok. It has access to the X API natively, so it gets data faster than most humans can. Go get a $40 membership and use it.
My prompt before this interview was simple. I am getting ready to be a guest on Goal Crazy podcast by Jason, give me a quick prep sheet. It told me how many users he has and how many countries he is in, that he does real estate in Akron, that he is in his young thirties and started his business about eight years ago. It gave me the questions he was probably going to ask and some interview tips. All things I would have had to dig up on my own.
I do the same thing offline. My wife is a local content creator here in Austin and we were driving to JewBoy Burgers so she could interview the owner and shoot a reel for Instagram. We have a Tesla, so on the ride over I hit the grok button and asked about Mo. It gave me a ten minute conversation, so when we walked in we knew Mo was an ad man, knew he was from West Texas, knew why he started the place and how he named it.
The next step is stacking that. Every morning my Hermes agent sends me every podcast I am booked on with a full report based on Grok's data, so I do not have to go prompt anything. Beyond that there is a tool for everything. Hey Jin if you want an avatar, and Julie McCoy has an avatar with a YouTube channel at about 250,000 subscribers, built on scraping news, writing scripts, designing the avatar, 11 Labs for voice and putting all the motion in. Remotion does video editing. There are models in Claude that take a long form video like this podcast and clip what is likely to go viral.
Where to actually start if this feels like too much
Jason said out loud what a lot of people feel, that it is a deep world and the volume of opportunity is what stops you. People go into analysis paralysis and do nothing. My first suggestion is to get involved in a community of people who understand business and are working on practical implementations of AI. Not pie in the sky dreams about three d printing a robot to fold your underwear. That is cool, but how are we going to use that in business?
The second suggestion is a shift in how you prompt. Instead of telling AI what you need, ask it what you do not know. Look at your calendar, look at the things you do not like doing, describe the real tension. I do not like checking email every day but I want to give really good proactive responses to people who write me, how do I solve that? Then read what comes back, ingest it, think about it, pray on it or meditate on it, whatever that is for you. Solve little problems at a time and build trust that it can do a good job for you.
Instead of telling it what you need, ask it what you don't know.
The time back is not for more work
I do not think we were put on earth to work. I think we are stuck in a weird evolution cycle where we have to be the worker bees. We have a short amount of time and neither Jason nor I knows how much is left in our time bank. AI is going to give us our time back, and I think we are going to be able to stop trading time for money. Elon Musk has the same hours in the day we do and he is also one of the top 10 gamers in the world. I do not play competitively. So what has he figured out, and why is he building Grok and Tesla AI?
When Jason asked whether my own hours have dropped, the honest answer is no. I am working the same amount of hours, but I am making more impact inside them. I used to spend hours on routine mundane tasks and finish the day having worked hard without feeling effective. I avoided podcast outreach for the longest time because it is tedious and I did not like the junk that goes out. Now an agent does it and I get to show up. I am not letting AI do this conversation for me, I do not believe in that yet.
What changed is anxiety. Someone opened a support ticket while we were recording. It is already solved and they already wrote back to say thank you, and I never touched my keyboard. Last night I went to pop stroke with my wife and my son, we laughed and danced and played and ate, and I did not have to touch my phone because I knew we had a triage system for any fires. Normally I would have checked it multiple times.
Jason made the point that matters most here. If you do not set your own goal, you will fill the freed hours with more work. I would add that goals need containers. If he and I were sharing a hospital room in our final days, neither of us would say we wish we had worked every Saturday and Sunday. Jesse Itzler talks about the fact that by the time you reach 18 you have spent 90% of the time you will ever spend with your parents. Both of my parents have passed away from cancer, so I do not get that chance anymore. My son is 18 and lives here with me, so I do. That is what the time is for.
I'm still working the same amount of hours. But when I'm not working, I feel more, I feel less anxiety.
Why I show up as a giraffe
Jason noticed he had to look pretty hard to find a video with my actual face in it. On social you see a giraffe. There is a reason. I am going to age and I do not want to have to work my whole life, but I believe my knowledge and my intellectual property and my thoughts will be valuable for many, many years to come.
If you get used to learning and engaging and interacting with my character, I never have to make some low quality AI generated version of me that either ages or never does. I have also lost friends. Jesse Lee was a big influence on me, a creator who passed away with cancer early in life, and seeing a video of her today would throw me for a big loop because I know she is no longer on earth. But if I saw a giraffe teaching her principles I would think, this is Jesse the giraffe, that is amazing. So I built a brand around a character instead of my face. My character can live forever, very much like Mickey Mouse.
The moments worth clipping
Short cuts from this episode, if you want the highlights before the full thing.
Questions people ask about this
Which AI tools should a small business owner learn first?
Focus on Claude and OpenAI's models, which is ChatGPT, and you will be ahead of the curve for most of what you need. OpenAI has Codex and Claude has Cowork or Claude Code, and both connect into apps like Gmail and your calendar. Past those, the sleeper is Grok, which has access to the X API natively and is worth the $40 membership for research and prep work.
What is the difference between ChatGPT or Claude and tools like Hermes or OpenClaw?
Claude and OpenAI are pretty static in their learning right now, they do what you tell them based on your guidance, constraints and controls. Hermes and OpenClaw are built to run fully autonomous without your interaction and are dynamic, constantly in a learning loop to improve based on your goals. The tradeoff is that you manage more yourself, do some installs, and will probably end up in a command line interface.
How do I figure out what to hand off to an AI agent first?
Make a list of everything you do on a daily basis, the routine and the mundane. Then make a separate list of the things you should be doing but are not, because you do not like them, do not have the time, or do not see high value in them. Then look at both lists and ask where AI can step in, because it can in almost every single one of those instances.
Should I tell people when an AI agent is contacting them on my behalf?
Yes. I believe wholeheartedly we should not try to hide AI, we should be forward with it. My outreach opens by saying this is Luna, Gary's AI assistant. Jason gets two to four to five podcast pitches a day and can spot generic AI pitches instantly, so the thing that makes it work is not concealment, it is that the message is genuinely personalized and pitches what we could create together.
Has using AI agents actually reduced the hours you work?
No, I am still working the same amount of hours. What changed is impact and peace. I used to spend hours on routine mundane tasks and end the day feeling like I worked hard without being effective. Now a support ticket gets solved while I am on a podcast without me touching my keyboard, and I can go play mini golf with my family without checking my phone.
I feel overwhelmed by AI. What is a manageable first step?
Get involved in a community of people who understand business and are working on practical implementations of AI, not pie in the sky projects. Then change how you prompt. Instead of telling AI what you need, ask it what you do not know, describe a real problem you dislike solving, read what comes back, and solve little problems at a time until you trust it.
Watch or listen to the whole conversation
The full episode is on Goal Crazy, hosted by Jason VanDevere.
Written by
Gary Henderson
Founder of Gary Club
Keep reading
