Artificial Intelligence Podcast

Takeaways

• Start small with automation: Focus on automating small, repetitive tasks rather than trying to automate entire complex processes at once. This approach is more manageable and can lead to incremental improvements.

• Prioritize based on time and frequency: When deciding what to automate, consider tasks that take significant time and are performed frequently. This helps maximize the return on investment for automation efforts.

• Use AI as a tool, not a replacement: AI and automation tools should be seen as aids to augment human work, not completely replace it. They’re particularly useful for tasks where perfect accuracy isn’t critical, allowing for human oversight and refinement.

• Assess processes objectively: Use methods like shadowing or self-recording to objectively evaluate current processes. This can reveal inefficiencies and opportunities for automation that might not be apparent when simply describing the process.

• Balance automation with business needs: Be aware of potential pushback against automation and AI-generated content from platforms like Google, Amazon, and YouTube. Stay informed about evolving guidelines and ensure that automation efforts align with business requirements and industry standards.

Video

Transcript

Johnathan:

From musician to automation with today’s special guest, Blake.

[AD READ]

Now, your passion is a passion near and dear to my heart, which is automation. A lot of people aren’t even really sure what automation means. So let’s start at the beginning. What exactly is automation, and how does that apply to a business?

Blake:

From my perspective, it’s all about trying to figure out what are the levers that you’re pulling on a day-to-day basis? What is that monotonous work that you find yourself doing again and again manually? And trying to pick apart how can I take each one of those actions, and is there a way to write a script to do it for me or to take myself out of the picture so I don’t have to be a middleman? It’s really all about just trying to streamline one basic, everyday task and free yourself up from having to do it continuously.

So a great example there is, imagine someone that is trying to contact people that went to an event or something else like that. You might have people that fill out a Google form or something like that with their information. Then at the event itself, you need to verify if they showed up or not. And then afterwards, you need to add all of them to your CRM system, like HubSpot or Salesforce. And then you need to send them all an email thanking them for coming to the event and doing some sort of drip campaign to follow up with emails. Every single one of those steps, like after they put the email in, could probably be automated along the way so that you can get those types of emails and communications set up to immediately get sent to people.

Johnathan:

Now, the biggest mistake most people make when they hire their first VA is giving them broad goals without the mechanics or the steps. They’ll say, grow my social media. What does that mean? Which social media channel? What’s your metric? What’s your ROI? You don’t know, and what’s your process? How many posts do you want per week? If you don’t give them your recipe, then they don’t know if they’ve done a good or bad job and you’ll never be satisfied. This is why everyone hires and fires their first VA. It’s such a common mistake. What’s the really big common mistake people make the first time they attempt automation that causes them to give up? Their hand gets burned, they go, oh, this isn’t for me.

Blake:

I think a lot of people end up trying to go too big initially. I really suggest figuring out what the entire process is and how to break it into chunks. That’s okay if the whole thing isn’t automated. If you can automate just a single step of that task along the way, you’re going to be in a much better spot.

Like even ourselves over at Shipyard, we run a newsletter called “All Hands On Data,” and our team submits articles via a Google form that we automatically generate into markdown that we can copy and paste to send out in a newsletter. I would love if we didn’t have to copy and paste that markdown and it just automatically sent the newsletter every single day based on those contents. But it wasn’t worth the squeeze; it would have taken a lot of engineering effort. So we just focus on the small little bits that would ultimately improve our efforts overall. So I would say breaking it down into those small steps and being okay not getting the whole thing done.

Johnathan:

A lot of people think that there’s a magic button that can solve all of their problems. It’s the same way they approach AI. “Oh, this AI will now do everything for me.” And this automation, which is just an earlier iteration of the same idea, solving the same problems, which is, “Oh, this automation will solve all my problems. Now the automation can run my business. The automation is the computer VA.”

Where, let me put it this way, how do people determine where they should start? Which problem they should try to automate first when they’re looking at their business? A lot of people, when they approach AI, they want to fix the coolest thing. They want to learn how to make AI videos. Yeah, everyone likes AI video, which is useless for business. The same way that people love AI images, which are not really useful. They’re cool, and I love AI images, but stock photos already existed. The difference in ROI or result between stock photos and AI images is very small, but it’s cool and exciting. So a lot of people chase the cool and exciting. Where should they start instead? What’s the right way to approach your business when you’re saying, which task should I automate? What’s your process?

Blake:

One of the best ways that you can try and prioritize things that are right for automation is figuring out what is taking the most time in the business to do, and then how often it is being done. There’s actually a great XKCD comic about this that basically shows you how much time you should spend automating something based on the amount of time it typically takes to do the task and how frequently you’re doing it. But because a lot of engineers, they often spend more time trying to automate the process than what it would actually take to realistically do it. And it’s easy to fall into that trap as well, which is the kind of the opposite problem. But I’d say that’s one way to really be able to prioritize things.

I think especially when you talk about automating more things with AI, it’s really looking at, is the end result something that you can deal with imperfection on? Because a lot of people that are working with AI know that hallucination is a big deal in this space, and you have to be really careful about not outsourcing some sort of activity to the AI when you want that output to look a very specific way, when it needs to be a hundred percent accurate every single time.

But for things like being able to write in your brand voice, a product description, or a hero-like header or call to action or something else like that, those are easy things that it’s okay if it doesn’t get it right, because half your battle with those types of things is A/B testing them, making sure that they’re super effective, and verifying what’s going to resonate with people the most often. But I would say that if you’re really trying to automate stuff with AI, figure out what you spend a lot of time doing, how frequently you’re doing it, and then do you need the output to look the same way every single time? Because if so, AI’s probably not going to be the right option right now. But if you’re okay being able to look at and evaluate the output and that helps you save time, then I would say that’s a perfect opportunity to start using AI as part of that process.

Johnathan:

It’s pretty interesting. My approach is, I always tell people, what’s a task that you do every week for at least one hour a day that doesn’t take a hundred percent of your attention? The kind of task where you can listen to music with words in it, or you can have a television running and you can still do the task. ‘Cause we all have those things. It’s like doing the laundry. It’s a mindless task that has to get done. That’s where I always tell people to start. Then you can take that time and use that free time learning your next skill and next optimization.

Unfortunately, a lot of people, just like with automation, they’re so obsessed with time management. They spend all of their time, they’ll spend 40 hours to save five minutes, and it starts to become that battle of diminishing returns. And then you go, I don’t want to take advice from you because you’re too optimized. You spend all of your time optimizing; you don’t ever actually get anything done. And it’s exactly that balance, which causes people to… It’s the same person who goes, then goes, “Oh, I don’t want to go to the gym. I don’t want to get too jacked.” It’s like, you’re fine, you’ve never been to the gym. But it’s that same mindset of “I don’t want to be too optimized or spend too much time,” and people will pull away from it.

Or a lot of people, when they hear automation, they think it’s really technical. “Oh, I have to be a programmer to automate.” What level of knowledge does someone need to start dabbling in automation? Do they need to know how to program?

Blake:

I don’t think they need to nowadays. One of the coolest things to me about the recent evolution of AI is how it can write and execute code for you. Great examples there: I’m technical, I know my way around code, but I did take a month of my life to just not actually write code and only use AI to end up building anything for me.

And so, certain things like I attended online conferences where I wanted to be able to figure out who the other attendees were. I was able to download the source code and upload it directly to ChatGPT and ask it, “Hey, from this page that shows the attendees, can you grab all of their names and titles for me?” And it would go through a process of generating the code on the backend and spitting out a file and saying, “Hey, does this look right?” The first time, the answer was usually, “No, it looks terrible. You didn’t pull the right information. Can you pull this instead? Here’s an example of someone’s name. Here’s an example of someone’s title.” And it goes back, it writes more code, and then it spits out another file.

And it becomes an iterative process along the way to where if you are super technical, you can dig under the hood, take the code, execute it on some other platform if you want to. But if you’re just trying to parse data on the web, or if you are just trying to write a simple script to convert a file from one format to the other, or make a background transparent or something else like that, I’ve used it for all sorts of utility tool-type things that don’t actually require me to code. They just require me to have some background knowledge of what I’m trying to accomplish and a good enough vocabulary to explain what the end result is that I want to finally get to that end result.

Johnathan:

A lot of people get really scared when they hear words like “coding.” Anytime I say it’s on GitHub, people freak out because GitHub is the scariest website in the world. And there’s this fear that “I’ll try to write a piece of code with ChatGPT and I’ll break my computer.” A lot of people have that fear of programming. It’s that hard fear of the same thing like clicking a link 20 years ago that will crash your computer. “Oh, you activated a virus.” Is that really possible? Do people need that fear? And how do we alleviate that fear?

Blake:

It is really interesting you say that because I’m in more technical circles and I haven’t heard that as often, the fear. But I feel like a lot of the AI tools right now, they’re so big on safety measures that it’s really not possible to have anything that could break your computer or cause issues. Basically, everything they run on their side is sandboxed, which means that it’s only living in the cloud on their servers while things are running there and nothing is actually ever going to live on your computer.

In fact, if you ask it to do anything that could be potentially malicious, it’s not going to write that code for you. It’s not going to execute it. And so any sort of downloadable links or anything that you might get out of the activity are just, they’re pretty much guaranteed to be safe at this point because there’s been people on a red pill team trying to make sure that this stuff can’t be used for bad purposes.

And so I think the code aspect can be daunting initially, but I would also say that a lot of these tools have tried to make it more approachable by literally hiding what’s going on under the hood and just saying that it’s processing until it can finally give you some sort of linked output that accomplishes whatever goal you have.

Easing the fear really just comes out of trying it more and seeing that it maybe isn’t that scary and that you’re not having to run code on your machine. You’re literally just coming up with product requirements and verifying the output afterward, which is no different than if you handed off the task to a developer at your own company. You’re not expected to dig under the hood. You’re expected to evaluate, “Did this solve my problem like I wanted it to be solved?”

Johnathan:

We’ve talked a little bit broadly. Let’s get specific. What are the most common tasks people should automate? Maybe the top five or ten that probably everyone, it’s one of these, that’s where they should start.

Blake:

I think it’s really difficult to say because it depends a lot on your context and what you’re doing on a day-to-day basis. So I have a lot more automation opportunities that cater themselves towards sales and marketing. So the whole thing I mentioned about being able to find attendees that went to a virtual conference so that I could reach out to them or being able to have automated outreach for people that sign up for events or that sign up on your website to download a white paper or something else like that.

I think, if you’re in the marketing space, the automation of writing copy for the website can be very helpful. Although I would not advise just taking exactly what you’re given and putting it on the website because that stuff is starting to get shut down by Google because they don’t want there to just be a bunch of AI-generated content out there.

I also think there are opportunities to…

Johnathan:

Maybe?

Blake:

No, I’m not sure. It really just depends on the type of situation that people find themselves in on a day-to-day basis. And as an engineer, I could talk through a lot of things I’ve done, but they’re not fully automated. I’m almost treating it as outsourcing that I don’t want to do the work. I don’t want to take a while, and I might only be a one-half activity. But if I can save an hour of me having to do the work by just outsourcing to some sort of AI tool and asking it to accomplish a very specific task for me that is just converting a file, transforming data, figuring out how to write some sort of SQL query or something else like that to get data out of a system, then that’s what I do now.

It’s not necessarily automation, but it is things that ensure that I’m not having to spend a bunch of time on my side to do very basic activities.

Johnathan:

A lot of people don’t really self-assess, especially new entrepreneurs or smaller businesses. They don’t track how much time they spend on different tasks. They don’t track which task generates the best return on investment, and they don’t track which tasks are the most efficient. So they don’t really have that data to come from. They don’t track their process. They don’t write down how do you decide what you’re going to post on Facebook? What’s your process? And they go, “I don’t know, I just post what I feel like.” How could anyone learn from that? But that’s how a lot of people describe their processes.

So even before the step of automating, how do people, or how do you teach people, or how should people assess what they’re doing to see where the opportunities are? And how can they start to create a process to describe their own, what they’re doing, so they know the recipe they’re trying to create when they look at automation or tools to speed up those processes?

Blake:

What I found in a lot of businesses is that shadowing is a great exercise. If you are able to watch what someone else is doing and go through their process and you’re just a silent observer, there’s a lot of things that you’ll see that are just slightly weird, or you’re like, “Why did you do it this way?” But you don’t want to make someone self-conscious in the moment; you want to just see if you can find any weirdness about how they’re going about something. Like it takes a bunch of clicks, they’re having to switch back and forth between a lot of windows and things like that. Usually, those are indicators that something could be fully automated.

But I would say for yourself, the similar tool is just making a video recording using a tool like Loom to record yourself going through a process and then watch it back later. Because you might realize that there’s something weird about what you do when you’re watching it as if it was a third party.

We use a tool like Hotjar to record how people use our application, and it is one of the most revealing things in terms of figuring out where people click, where they’re getting lost. And I always equate it to my team that it’s like watching your favorite football match or something else like that, and your team’s not doing what you want them to do. You’re like, “No, what are you doing?” You start to get pent-up anger and excitement at the same time of, “You’re so close to being able to do this better, but you’re just missing these one or two details.” And I think that only comes from being an observer, both of yourself and of your peers around you, to figure out opportunities and ways that you might do things differently that could potentially be streamlined. ‘Cause you’re right, people aren’t very good at describing what their processes are, but video does not lie.

Johnathan:

That’s why I’m such a big fan of video as well. When I switched from writing down instructions to sending Loom videos and recordings of me doing something, and when I’m trying to learn someone’s process, I always want to watch the video of the process because there’s something people forget to describe. A great example of this is when you’re trying to record a process, which some people call a macro, right? It’s a series of steps to order. And I say, how do you make a peanut butter sandwich? And everyone always says, put down the slice of bread and put the peanut butter. I go, “Where’d the bread come from?” You didn’t, yeah. “Oh, you opened the bag. How do you unseal the bag? Is it a twist tie? Is it a pull-off? Where’d the bag come from? Was it in the refrigerator? Why didn’t you open the refrigerator door?” When you start thinking like that, you realize we naturally skip over the steps that are easy.

When I first learned to drive a car, there were 13 things I had to do before I turned the key. I had to check all the mirrors. I had to check the seat adjustment. Now, come on, I don’t do any of those things. It’s natural. I don’t remember them because they’re just part of my process, but they’ve been forgotten. When we make a video, we can’t skip over that step, so then I see it. This is why when I’m learning from someone else, I always prefer to watch a video than to learn from a written guide, because writing instruction manuals, it’s hard. That’s why there are people, that’s their entire job, who are experts because they remember every step. Most people don’t.

So that’s very good advice. I think recording a video of yourself, that’s where everyone gets good at anything is to see what your process is rather than try and describe it or even write down what you think the process is. Record yourself doing it and compare the two things. I use a new tool called StepSee, and what it does is record every time I click my mouse, it takes a screenshot. That change when I, when that’s on, I completely change my behavior because I have to go back and delete all the extraneous screenshots. I have streamlined my process when I’m doing that. It really changes how you do things ‘cause if I don’t pay attention, there’s 200 screenshots and I need to get rid of 90% of them.

And when you watch yourself, you start to see how inefficient what you do is, especially because I’m recording a video and teaching the process and recording the screenshots. So much teaching when I just need to be showing this is the demonstration. Just demonstrate. It really forces you to become efficient whenever you’re watching yourself or tracking yourself in any way. It’s so revealing how inefficient you are or how strange some of the things you do are, because sometimes you have a process. My kids are great at this. They’ll always say, “Why do you do that? Why do you do that? Why do you do that?” There are so many things we do and we don’t know why we do them ‘cause we just saw someone else do it and we don’t know why they do it. This is a really great place to start.

Where do you think automation is going in the future? Do you think it’s all going to become AI-driven, or are we going to pendulum swing towards automation and pendulum back? Because I see certain platforms are trying to push back against automation. Now on Amazon, if you have AI-written content, you have to tell them as of last week. YouTube, you have to tell them if there’s any AI content that could confuse someone, which is a very…

Blake:

…gray area. It’s very broad.

Johnathan:

It’s something I struggle with because I use AI-generated pictures of myself in my thumbnails, and could someone believe that I’m really a centaur? Maybe, I don’t know. That’s the problem. They shouldn’t, right? They should know it’s fake because I’m holding a, you know what I mean? It’s an anime drawing, but it is of me, and I always wonder about that, where the line’s going to be drawn. Do you think that we’ll actually solve the problem? Or are we going to go through a period where automation and AI get all merged together and people swing back away from them before we find the middle road, the right path?

Blake:

I’m not quite sure. And I do think it’s interesting specifically from the automation side of things because, to your point about using macro recorders and things like that back in the day—like we’re talking maybe nine years ago or so—I used to use tools like, I think it was called like gbit macro recorder, where I could basically have it emulate all of my mouse clicks, everything that I typed, and I could see all the space in between. And I could reduce that as much as possible to try and take the human activity of me clicking around on all these pages and make sure that it could be automated on my computer without me having to sit there and run it. And I use that quite a bit. And you’re right that it’s very revealing when you’re doing way too many tasks and things like that.

But that process is very manual still because I have to make sure that I click all things in the right order, in the right place, and that window is always going to pop up pixel perfect in this position on the screen because if it doesn’t, things are going to go wrong. And I think that level of manual effort is ripe for having AI come into the picture. I think some things like the Rabbit Large Action Model that they’ve talked about, where you’re able to record yourself doing a process and contextualize it so that it can click the buttons on your behalf, that’s probably where we’re going to be going to. Where I would love to have the ability to record what I’m doing on the screen and then have AI describe the context of the actions I was doing and then be able to perform it on my behalf. Just as much as I would love to be able to describe the process and have it figure out all the gaps and holes that I forgot to mention along the way because they’ve just become rote and I don’t know exactly where the balance is on things like that. But I think ultimately we’re going to have this desire to not have to specify every little last detail. What we really want is the ability to describe the context of how something gets done and the specificity of what we want it to look like at the end, but all the extra little details, we don’t want to have to sit there and record it. We don’t want to have to sit there and clean it up to make sure that it works correctly. We just want the work done.

I think the only way that we get there is by having more of these AI tools fill in the gap to make that easier. And I’ve even seen interesting things like this in the development space when it comes to testing an application, looking at all of the user events that people typically do and then trying to emulate, “Alright, normally someone clicks the ‘new document’ button and then they type things in and then they click save.” So that’s something we want to test: can someone open a document, type things, and then save it? Right now, QA people are having to manually write out those tests to make it possible, but that shouldn’t have to be the case if we know that 90% of the time a user does these three actions in a specific order. And so that’s where I think we’ll really start to see the balance when it comes to AI being able to contextualize the information of event-based data or click data, or even just watching users where they typically move. We’ll be able to streamline that process a whole lot more. But you’re still going to need someone to tell the AI, “Hey, this is exactly what I want to be automated. This is the end result, this is the input. Figure out how to get it from A to B.”

Johnathan:

Yeah, I think the biggest value of AI is that it can understand when you give it a more incorrect description of what you want. When I was first using computers with DOS in the 1980s, if you capitalized a letter in a folder and it was supposed to be lowercase, it would go, “We don’t have that folder.” And you couldn’t make any errors, even as far as capitalization and punctuation. And I still have the habit of, I don’t put spaces in file names because I know that used to cause your computer to crash, even though now it’s not a problem anymore. They’ve learned that. I have that habit inside me. The same way I manually hit save every three to four minutes when I’m writing a document. I’m constantly hitting save on Google documents, and Google’s like, “What are you doing?” I’m like, it’s a habit. Okay, you lose one term paper and you learn to be very paranoid about your saving habits because there used to be no automated save. Now that has an automation, which is a very useful automation.

I think that the beauty with AI is you can more and more describe a process and it knows the steps in between. You don’t have to program step 1, 2, 3, 4. It will go, “Wait, you missed step 3.5,” and it will see that because of context. And that’s really the value of it, that you can give, as AI gets smarter, your description can be more wrong, but AI still figures out what you mean. And that’s what I think is really exciting. So I love what you’re saying. I think this has been very interesting. I know a lot of people are interested in automation. A lot of people are trying to figure out where does automation end and AI begin. And I think that there’s a gray area where they, or a Venn diagram where they cross over quite a bit. So that’s why I’m passionate about this topic. It’s anything that can save me time gets me excited.

Where can people learn more about what you’re doing, connect with you online, and see more about your approach to automation, and maybe just follow you?

Blake:

Yeah, you can find me on LinkedIn. I write occasionally on Substack under Bytes Data Action. And then my company, Shipyard, we’re a data operations platform. You can find us at shipyardapp.com.

Johnathan:

Perfect. I’m going to put those links below the video and in the show notes. Thank you so much for being here, Blake. This was another amazing episode of the Artificial Intelligence Podcast.

Thanks for listening to today’s episode. Starting with AI can be scary. ChatGPT Profits is not only a bestseller but also the Missing Instruction Manual to make mastering ChatGPT a breeze. Bypass the hard stuff and get straight to success with ChatGPT Profits. As always, I would love for you to support the show by paying full price on Amazon, but you can get it absolutely free for a limited time at artificialintelligencepod.com/gift.

Thank you for listening to this week’s episode of the Artificial Intelligence Podcast. Make sure to subscribe so you never miss another episode. We’ll be back next Monday with more tips and tactics on how to leverage AI to escape that rat race. Head over to artificialintelligencepod.com now to see past episodes, leave a review, and check out all of our socials.