Intro to AI for Business & Work with Matt Rouse (280)
In this solo episode, I share a practical beginner’s guide to using AI at work and in your business—plus a solid refresher if you’ve already dabbled with tools like ChatGPT. I break down, in plain language, how modern AI is trained, why tokens matter (and what they cost), and where the real money in AI comes from. Then we dive into the six essential skills you should build right now: prompting chatbots with context and constraints, generating and editing images, uploading and analyzing documents, using computer vision to solve real‑world and software problems, conducting deeper research, and creating lightweight “software on demand.” Along the way, I give concrete prompting frameworks, tips to reduce hallucinations, ideas for cross‑checking outputs, and hands‑on examples; from summarizing contracts to troubleshooting hardware with photos and iterating on AI‑built mini apps. I also highlight free and accessible tools you can start with today, including image generators, Microsoft Copilot, and Google’s NotebookLM for research, slide decks, summaries, and auto‑generated audio/video overviews. Whether you’re an employee sharpening AI literacy or a small business owner looking for leverage, this episode will help you build confidence, pick the right workflows, and get meaningful results, fast.
We wrap with practical do’s and don’ts, emphasizing context, constraints, audience, and outputs for better prompts; when to start a new chat to avoid lost context; how to cross-check outputs; and why computer vision is an underused superpower. If you want a no-nonsense roadmap to immediate AI wins—and a clearer picture of where the industry is headed—this episode is for you.
https://matthewrouse.com
If you like to hear about AI & marketing, and posts about chickens, connect with me on the LinkedIn: https://www.linkedin.com/in/mattmrouse/
or SubStack if that's your jam: https://mattrouse.substack.com/
Looking for a podcast guest? Author Matt Rouse
Hook Digital Marketing | Hook Digital Marketing Canada
Market your local business on autopilot: SMB Autopilot
Welcome back to Digital Marketing Masters. I'm your host, Matt Rouse. And today, we don't have a guest. I recently have done
several
both in person and online trainings
of AI for work and employment.
And essentially, this is a a beginner's
guide to using AI in your work or your business.
And,
you know, if you are familiar
with kind of using, you know, chatty for tea, maybe some generative tools,
this could be a good refresher for you. There's also a little bit of other information about how AIs are trained
and how those systems work and kind of where all the money that they think is gonna come from is actually gonna come from. So
we'll get started here in just one second.
For
an introduction
to AI for work and business, for beginners or as a refresher.
And find out ways you can use AI. With your host, Matt Rouse.
Episode through 80. On the Digital Marketing Masters podcast.
Episode through 80. Now on with the show.
Alright.
So we're back, and
I think the first thing that we wanna go over
is who I am.
And for those of you who've listened, you know, on and off maybe or all the time for the last eight years that we've been doing the show, you have a pretty probably a pretty good idea. But I'm the cofounder of Hook Digital Marketing. I'm the author of eight business books, two of them about AI.
Will AI take my job and will AI take my job too, which I also just updated, about six weeks ago. So you can get a copy, and it's up to date. And I'm the host of the Digital Marketing Masters podcast.
I'm a homesteader
in Nova Scotia in I live in a
county called Annapolis,
and,
I have a 10 year old human daughter.
My wife, Carrie,
we got two cats, a dog, and 46 chickens, and 48 eggs in incubator. So soon to be more chickens.
I think
one thing
that when it comes to things like
employment
and work and stuff like that, I I really like the last time I did this was in in Yarmouth, Nova Scotia for Nova Scotia Works, and that is a group
that helps people either find employment or work on their kind of businesses and side businesses and their skills for those businesses.
And
so
we're gonna go over, like,
some
ways to kinda build some comfort with using new technologies, some terms,
why employers are valuing AI literacy. We're gonna go over chatbots, conversational AI, generative AI for images, audio, and video, some AI based tools, talk a little bit about AI agents and automation.
And,
you know, if you are
watching this,
then you will see kind of a breakdown of time, and that's not the actual timeline. That's, like when we're in person,
hands on doing a workshop, answering questions.
This actual run through is not gonna take nearly three hours. So it's much shorter than that.
I think
fundamentals
of how AI works, this is an oversimplification,
of course, because we don't have time to get into the whole thing.
But basically, we take a lot of data, and when I say a lot, we'll get into how much is a lot,
And that data is used for training. And training
is what essentially makes the what we'll call the neurons.
Usually, they're called parameters in AI. Those are like the brain neurons,
and and the training data trains those neurons and those pathways
that make the system able to actually think.
And it correlates
information.
That's how it puts output. And then there's a post training system,
and post training is actual information
or software that the AI has,
and then it has the ability to make an output.
There is another layer in there of kind of tooling layer, which would be,
like, if I want my AI system to be able to correctly do math, then I would give it a calculator program.
If I want it to,
you know, be able to analyze a document, then I need to program for that. So there's another tooling layer in there, but we're not really gonna cover that.
So
training data, a lot of it is,
like,
a lot of data that is very similar.
And
if we're talking about something,
like, that's like chat GBT and things like that are trained on tons of text as well as other stuff if they're multimodal, which means, like, they can understand images or video or something. But like an image model, say a program like mid journey or something,
what they're doing is they're showing you, like, millions of images of cats, for example.
Yeah. It kind of correlates all of those things and says how far apart are ears,
my cat, like, what the cat's face look like,
You know, how does his body work? How does its tail? All that kind of stuff. And then they showed a bunch of other pictures that are sort of similar, but they're not a cat.
And they have the system trying to determine if it's cat or not, and then they tell if it's right or wrong.
If it gets it right,
there is a reward mechanism,
which essentially is like a circuit saying, oh, you've done it. Here's a little bit of, like, the human equivalent of, like, a little bit of dopamine. Yeah. You got the answer right. Here you go. You're happy.
But modern AI systems need a lot of data.
And if you could ask, well, how much is a lot?
GVT five is trained on approximately
70,000,000,000,000
tokens,
which is about 281
terabytes of data.
If if we convert all that into actual words,
then
if you read 250
words a minute,
reading those 52,500,000,000,000
words would take
four hundred thousand years of nonstop reading.
So twenty four hours a day from four hundred thousand years, that's how long it would take to read all the data. And then additionally,
there is something called synthetic data,
which is AI models coming up with data to train the AI models.
And then there's also
code, multimedia,
audio, all kinds of other code that goes into these systems.
So we're talking about tons of data, like, every encyclopedia,
every book ever written, the entire Internet, that kind of thing.
AI systems run tokens.
Tokens are essentially
like a symbol, a syllable.
It could be just a single character like a number,
and those tokens are how the system communicates
inside like internally in the software.
And so
you can usually say,
know,
70%
of the number of tokens is roughly the number of words.
So if something is gonna be a 100,000 tokens, it's 70,000
words.
And the more tokens you give a system,
the more that it can kind of spend those tokens on thinking.
And there's input tokens,
there's thinking tokens, and there's output tokens.
And I know it's a bit confusing, but the easiest way to think about it is
the input tokens you pay for, the output tokens you pay for.
So if you're writing a piece of software or using a program where you gotta put your chat GPT API key or your cloud API key in,
you pay based on a million tokens.
So if we do the 70% rule,
every 700,000
words is like $15
or $5
or, you know, it depends on the model.
There's also a different price for input tokens versus output tokens, so that's a good thing to check on. It's a lot cheaper
for input tokens and output tokens.
So
if you put a novel into a system and have it spit back out a few pages,
it's probably one third of the price of putting in a few pages of information and having it spit out a novel.
So output tokens
cost more than input tokens, but essentially all AI systems
run on these tokens.
So a lot of people say, everybody's going broke in the AI world. They're running out of money and yada yada yada, but
that's because they don't really know where the money is generated.
So everybody thinks that paying
the subscription fee for ChatGBT is where all the money is,
but probably,
you know, ten, twenty, who knows how many times that amount of money is used through the API, which is an application programming interface.
So if you're using something like Salesforce
or you're using
like, any SaaS program and it has an AI function,
that function, they don't make their own AI.
That goes through an API and application programming interface
to another AI,
sends them data,
tokens are used, they send back information,
and then they pay for those tokens.
And the reason everybody thinks they're going broke is because they're making these massive investments in data centers and chips and compute and all this stuff.
And the reason that they're doing that
is that it takes about two years to three years
to get the amount of they call it compute,
which is basically computing power,
to get the amount of computing power that they need built
in the future.
So they gotta pay for it now,
and they have to
try to estimate how much they're gonna need two and three years from now.
So
if the current kind of hockey stick growth of AI continues,
they're gonna need,
you know, 10 times the compute next year and another 10 times the follow gear and another 10 times. So now we're at a thousand times the compute of what they're using this year.
But they also if you order
too much compute, you go broke. You order not enough, then you can't make any money.
So there's this kind of
investment game
where they're trying to figure out how much demand there is gonna be for their service and at what cost.
Because once they train the model, which is the expensive part, using the model is relatively inexpensive and that's where they make profit.
So
companies
basically paying for tokens
is how AI's companies make money, and that's why people think they're going broke is because they're spending tons of money and not earning that money amount of money now. That's because they have to invest in future compute.
By 2030, it is estimated that
AI companies
will have revenue of about $4,000,000,000,000,
and
that does sound like an absurd amount of money considering probably the top three companies in the world together are worth $4,000,000,000,000,
but
the thing you have to think about here
is where do they think that revenue is gonna come from?
And not all of that is generating new wonderful things.
$4,000,000,000,000
is roughly 10% of the entire world's
labor market. Meaning, if you combine all the salaries of every worker on the planet,
10% of that is $4,000,000,000,000.
So
I think we can all estimate of where that revenue is gonna come from or at least some of that revenue is gonna come from would be the labor market.
So
I think now that we know a bit more about
AI
and,
you know, that it's not going away anytime soon,
there's a good chance
that trillions of dollars will be spent on it and it's also going to displace
at least a portion of the labor market. If not a lot of it in some areas, it depends where you work.
If you wanna find out more about that, can always get my book, Will AI Take My Job Too. You can get it on Amazon.
And
so I think when we talk about skills that are in demand right now or skills that are useful for your business,
I think you need to know five things essentially. Let's say six if we really wanna push it.
You need to know how to properly prompt an AI chatbot. You need to know how to generate an image
and hopefully to also edit that image.
You need to know how to upload documents or data into a system.
You need to be able to use computer vision.
You need to be able to know how to do research
and
studying and kind of deep research tasks.
And you also for number six is, let's call this the bonus one, you need to know how to create software on demand.
So we're gonna go through them one at a time.
I think the first thing is how do you prompt an AI system?
And
the example that I used when I was teaching this in Yarmouth is that
I'm creating a presentation for Nova Scotia works to teach people how to use AI for work. I am making it in Canva.
So that portion is context.
I'm giving the system context around what it is that I'm doing.
And then the next one is, I want you to create a background for the slide leaving the center area light colored so the text will show up. I'm telling it what I want it to do
and I'm putting like a little bit of constraint on it. Right? I want it light colored so don't use other colors. I want the center area light colored.
And then for additional information, I put do not put text in the slide background image.
It should use Nova Scotia works colors, make it snazzy.
And I used meta AI for this, but you can use pretty much any AI that will generate images.
I just wanted to give them some examples and ones that they can use for free since, you know, they're trying to build a business.
I know a lot of people are kind of used to
the idea
of giving an AI chatbot a role and then asking a question.
And I think there's there's two big problems with how people prompt AI systems.
Number one is they have one way that they prompt the system and they use it for everything,
which there should actually be different ways to prompt it based on different tasks that you're doing.
The other thing is that people forget
a couple of the important bits,
which are context and constraints.
So
if you want kind of a general rule for prompting your chatbot, you say, I want you to act as whatever the role is. So I want you to act as a poet. You know, I want you to act as a professional
book editor for fiction books. I want you to work as this. I want you to be a data analyst.
Whatever the role is.
The idea of that is you're you're telling it
where in its, you know, AI computer brain it's gonna be thinking from.
And then I want you to do some kind of task. Right? That could be write me a limerick,
help me with some software, analyze this data,
know, yada yada yada, whatever the thing you wanted to do is.
The important parts come after this, and these are the parts that everybody freaks.
You can say, here is the context, put a colon,
and just tell it all the context you have on that task.
And don't be scared to like put a lot of context in.
I know that we've
used prompts for, you know, software as well as some of the ones that I kind of reuse for tasks and stuff that could have as many as, you know,
with documents and stuff, which we'll get to in a bit. But just in the chatbot itself,
sometimes, you know, half a page, maybe even a page of context.
I also like the idea of putting
the audience is
and then who is the audience.
What you're telling it is, like, if the audience is experts or the audience is beginners or, you know, the audience is
professional painters or whatever it is,
it's gonna
generate the output with that in mind.
Now the with that in mind, you could put the output should be and then tell it what format you want it in. Like, the output should be a PDF file. The output
should be a list, should be a table, should be a story, whatever.
And lastly, I would put constraints.
And the constraints are the rules that you wanted to follow when it does the thing.
So
this is act as role,
do a task.
Here's the context.
The audience is blank.
The output should be format
and constraints.
Give it the rules.
So a simple one here is
act as a poet, write me a short limerick about prompting.
The context is to include Nova Scotia works,
the audience is people learning about AI,
and the output should be standard poetry format, and the constraint is to keep it under a 100 words.
And what it gave me was there once was a prompt with some quirks that wandered through digital works. At Nova Scotia works, they sharpened its perks and taught how and taught AI how language works.
So it followed, it used the constraints. It's not the best poem ever written, obviously,
but it did the task.
So another prompting
tip
is
you can you don't have to, like, write everything into one, like, run on paragraph.
You could put stuff on different lines,
which can be helpful. So you can say, like,
I need 15 ideas for whatever the goal is. Right? I need 15 ideas for my upcoming marketing presentation
for my small business. The context is this is the business, these are the customers, this is the situation.
The constraints are it's gotta be realistic, low cost,
executable by a small team,
and the output, I want them in a table
with columns for the idea why it could work, the amount of effort it will take, and any risk involved.
So that's like a brainstorming
prompt.
You'll notice on that one, I didn't tell it
to use a rule
because you don't have to use a rule for everything.
Another one is business analysts.
So I can say,
I want you to analyze the situation
like a consultant. You are a business analyst.
Your goal is to help me what to do next. Here's the facts around my situation,
and I want you to return what matters most, what's uncertain.
I want you to give me several options.
Give me your best recommendation and tell me the risk of each option.
And you could also upload documents or files or supporting information with that.
So there's different ways to prompt the AI, and the other thing I think is really important
is to remember that you can ask the AI
for help prompting it. So you can say,
this is the prompt that I have,
and, you know, by reading the prompt, you can tell what I'm trying to get you to do.
Do you have suggestions
for me to improve the prompt to get a better input from you? And it will help you do that.
I
think that
if you're gonna use a chatbot also,
you don't have to one shot everything.
So if you put all the information in and everything in and you get an output,
you feel like you should refine that output.
I think
what people forget is that just because it wrote something once doesn't mean that if it would write the same thing again because there's a thing in AI systems called temperature.
And temperature is essentially how much randomness
is added into the system.
So if it writes you a story, gives you an analysis, whatever that is, there is going to be some randomness at the output.
So if you want to reduce the amount of randomness,
one way to do that through the chatbot
is just to ask it to give you more than one response or open more than one window and ask it the same thing in three different windows or ask it the same thing three different ways.
You can even take all of the responses you can get from those, put them all into one prompt, and say, these are the responses I got from asking these things.
Combine these into which you think is the best. You know? So
use your judgment, but also don't be afraid to iterate. The only thing you wanna worry about is if you're kind of working with it on and on through the same conversation.
Eventually,
what happens is the context builds up because
the system has to keep telling it all the things that you've already had a conversation about, and as that gets to be too much,
it starts to lose the plot.
And so you might forget what your original task was.
It may start to forget about some of the details, or it doesn't think they're as important anymore because you've talked a bunch about something else.
And if you're gonna have a conversation
about,
like, something different than what you were talking about, you should always use a new window for that. Right? Open a new chat and start over.
Give it fresh context and and go from there.
Pretty much any AI, the one I used in my example
was was using Copilot,
but
any AI system, any modern one, you can upload documents with your prompt. And
I think
this is probably one of the best use cases
for businesses or business people.
I really my favorite example is you get an insurance document, and it's like 50 pages long.
So upload it to the AI and say,
this is the insurance for my business or, you know, whatever it is for.
I want you to let me know if there's anything out of the ordinary,
anything that stands out to you, if there's anything about this I should ask my insurance agent about.
You know, help me understand this document. Also helps with understanding,
you know, if you get legal information,
terms of service, all these kind of things, contracts with other companies, you can put them in and say, hey. Help me understand this.
And also you could say like, if it's a legal document, you could say, if this if you think this is beyond,
you know, my ability to to have an understanding, let me know if I should go get legal guidance for this. Right? And they can help you, you know, with those kinds of decisions. I think the other thing you wanna remember is
the more important that the thing is, the more you need to pay attention to it. Right? And this goes for input and output. Right? So
there's,
you know, a worry about hallucination,
which does happen. It happens more in some cases than others.
When it comes to things like data analysis,
generally speaking, the hallucination is pretty low nowadays.
I haven't run into much in the way of hallucination.
I think
one of the things that happens with hallucinating is if the system is trying
to give you an output basically to please you. Right? It's trying to to answer your query,
but it the data doesn't exist or it doesn't have access to it, it will make it up so that it can complete the task.
And
like an example would be if somebody said, tell me about the nine books that Matt Rouse wrote but I've only written eight.
Sometimes it's gonna give you eight books, sometimes it's gonna be like make up the ninth book.
And, you know, obviously that's a problem.
Most
newest systems
that you pay for don't have that problem.
It is more common in the free models.
But as I said, they are getting better.
Kind of a rule of thumb, if it's life or death, you better have yourself and several other human experts check it.
You know, if it's you throwing out a post
on, you know, LinkedIn or something, you're having it help you, you you put a summary in of what you wanted to say and have it format it for you,
and it's not the end of the world. Right? If something you post out on LinkedIn maybe isn't perfect,
but, you know, you still wanna keep an eye on it anyway.
There is
kind of a tendency for people to have
like, the AI is correct 10 times in a row, so they don't check it the next 10 times.
But the error rate might be one in a thousand.
So you don't know which one of those thousand are gonna have the error in it. Another trick is you can take the output of of your AI, put it into a different AI,
and have it check to see if it agrees.
So
this
example that I used here is
our company
has and and we thought of this years ago, was kind of a funny thing. Now lots of people done it, but
we had some early AIs write a romance novel
and it was a romance novel about an AI.
And
it's
it's called Circuits of Desire. It's it's kinda comical. But anyway,
what we did is we used it to test the creative writing ability. But I took that document and I uploaded it into the AI chatbot, and I said, I want you to give me a summary of the document and the characters.
And I just uploaded like a word doc, and it was able to give me, here's a summary of chapter one, here's all the characters that were mentioned with the description of each, all the locations mentioned with the description of
each. And I checked it because I've, you know, read the book that we were working on, and,
you know,
it's correct. So I mean, from a data analysis standpoint,
I could not read,
you know, three chapters of a book and pick out one chapter and then to
summarize it and pull all this data out, check it for accuracy would have taken hours and, you know, Copilot did it in about forty five seconds.
One thing that I think is highly overlooked is computer vision.
So
the example I used in my talk was a picture of an oil burning
water heater.
And the reason I use that example
is that the old house that I live in has a oil burning hot water heater,
which I hope to replace soon.
But aside from that, what happened was it stopped making hot water,
and so I had to figure out how to solve this problem.
And what I did is I took a picture of it, and then I uploaded it to ChadGBT,
and I said, do you know what this is? And
it gave me some information.
Like in the example that I used
in my presentation,
I just actually grabbed a picture off the Internet to use,
And not only did it figure out all the components of the system
and
what the unit does,
there was also, like, handwriting on part of the box.
And it read the handwriting that said reset one time only, and then explained
that you only wanna reset it one time
because if the burner fails to ignite,
pressing the reset button pumps more oil into the combustion chamber. And if you do it too many times, you can have a fire hazard or, like, a kind of puff back, like, a tiny oil explosion of soot that shoots out of the machine.
So it actually read the warning that was handwritten on the machine and then explained it.
In my case, when I did it with my boiler, I I took the cover of the box off, I took a picture of it, and I said where is the thing that I used to check to see if there's oil coming to it?
And it explained, okay, it's this model, it's this year. I went and double checked that because I know safety first. Right?
And it was correct and, you know, it said to turn this thing. So basically, I followed the troubleshooting steps that it gave me
and I was able to get it working. You know? Another time, I was with a friend who had a camper,
and they essentially,
like, blew a circuit breaker. But we checked all the circuit breakers, and they were all still fine.
But we still couldn't get power out of this one outlet,
and we thought maybe there's a switchable outlet somewhere, which we did find. We said hit the reset. It still didn't do it.
So I pulled out my phone. I took a picture of the outlet, and I took a picture of the make and model for the camper, and it told me that underneath the sink in the bathroom, there's probably another switchable outlet, which I pushed the button, solved the problem.
Right? So you can also use things like screenshots.
So if you're working in a piece of software,
especially some that's a little convoluted, one I like to use is Mailchimp.
There are stuff in Mailchimp that's super hard to find because the I don't know why their UX is so bad still, but it is. So
you could take a screenshot and you can be like, where do I find the workflow section? And it'll tell you where to click to go find that, and it'll explain how to do whatever the job that you're trying to do is. And you can as you go through that job on the computer, you could keep uploading screenshots into it, and it'll say, okay, well, you've gotten here, Now you need to do this. And I can be like, okay. Well, I did this,
but then I pushed this and it didn't work or the buttons grayed out. And it'll say, here's three reasons why the button's grayed out. And, you know, so you could work your way through by uploading screenshots
and chatting with your chatbot,
and it'll teach you how to use the software that you're, you know, having problems with.
One
more thing
that I really like that you should try out if you have not is is called NotebookLM,
and it stands for notebook learning model,
I believe. NotebookLM
is free. You could use it on Google.
It is also connected with Gemini, is Google's AI chatbot.
And
the best thing about it is you can give it a website, you can have it go out and do research on its own, you can mix and match, you can put in notes plus upload documents plus have it do research. So you could basically load it up with information on a topic,
and then there's a bunch of options that you can do with that information.
You could, you know, get an audio overview, a slide deck, a video overview, mind map, you can have it generate reports, you can generate flashcards,
it can quiz you on it, it can make an infographic,
it can make data tables.
All of this is in there and it's all free.
And
my favorites are the audio overview,
which essentially makes a podcast
with it can have one or multiple speakers talking back and forth.
And it also has this super creepy thing that you can do that's kinda fun, where you can click on a button and you can ask it questions like you're calling into a radio show, and it will change the voices and the, like,
the show on the fly.
So that's something worth trying out.
Also, makes great overview videos.
So
you could
like, I pumped in my book and I made a quick kind of overview video and then I sent it to someone
whose podcast I was gonna be on
because I knew that they hadn't had a chance to read the book yet. So I was able to give them an overview of what my book's about in, like, a little, like, five minute video instead of them having to read, you know, 250 page book first.
Lastly,
and this is like I said, this is for bonus points. Right?
I think everybody at this point is sort of vibe coding,
And I think there's there's a couple differences
between vibe coding and kind of simple software on the fly, which I call on demand software.
So a good example is in Claude. Right in in just like the Claude interface, I didn't use Claude code or any of this kind of stuff.
I said, I want you to write me a program that will tell me when my car is paid off with a form for me to input the info.
And
in, like, about a minute and a half, it generated this form.
It was running in the browser,
and it had current balance, interest rate, monthly payment, and any extra monthly payments.
And I was able to type in the information
and have it give me a payoff date for my imaginary car payment that I don't have.
And it also gave me a balance breakdown over time about how much balance versus interest I had.
And it told me when the halfway point was.
All of this was written from that single one sentence prompt
into a working functioning piece of software that I can use.
And when you think about vibe coding, vibe coding is more like I wanna build an application
that's like a web based application or a website type application that I could use
repeatedly or maybe even have other people use.
And that's more like you're building a website or you're building a web application.
And this is like on demand kind of small software and scripts.
There are some other tools out there,
like SimTheory
is one, you can use Cloud Code or Codex,
where you can write software that will run-in the browser that you can securely run yourself that you don't have to share with anyone else,
and it's not shared with anyone by default.
And so you can make things like
an application
to maybe help you manage your own money or, you know, with some kind of internal business task that you need,
where you can maybe write a script that you would then export
and upload to a web server or use on a computer
where you could download the file to your computer, and then you could share with your other employees. They could use the same software that you're using. So software on demand is something where
as long as it's not production software, like you're not putting it out to the public,
you probably don't need a software developer to do most of this stuff.
Where you do need a software developer is if you're gonna have public facing software
or software that connects to other systems,
especially stuff like finance data or customer data,
and you need to ensure that it's secure
and also to ensure that it's not gonna like
f up all of your customer data. Right?
If you don't understand cyber security
at least basics,
you should not be making production software.
You can make a software
as like a prototype,
and then have your company's developers or go pay a development company
to take a look at it and review your code and make fix any problems with it. But I think to be safe, you don't want to release production code
if you're vibe coding, at least not in the state it currently is.
That's up to you. Do what you like. I'm just saying, in my experience, that's a poor idea.
So
that is how to prompt an AI chatbot, how to get better at prompting,
having the, you know, the AI chatbot help you with your own prompts, generating
images.
There's lots of information to find out online about generating images,
consistent characters,
how to make things more realistic,
which image generators are better at generating
images versus text versus,
you know, combinations of those things.
All that information is easily found online.
You should be able to upload documents
and
be able to,
you know, use analysis and stuff like that from the data that's in those documents,
as well as, you know, change document formats and stuff like that. Like uploading
actually, something I do commonly is I take screenshot,
I upload it, and I guess this is more computer vision, but it's
something that you can't copy the text, but you need to copy and paste that text somewhere. Take a screenshot and load it in and say, does this text say? And it'll spit it out to you and you copy and paste it.
Computer vision's great for solving problems,
real world problems, getting feedback
on, you know, mechanical
things or
also using screenshots,
so figuring out software and other types of problems.
Lots of ways to use computer vision.
I think super underutilized.
Research and study notes using
you know, we didn't talk about deep research systems, but you can click the the thinking version of your AI chatbot if you pay for it or the deep research kind of setting on it.
If you have one, and you can have it go use more tokens to try and do more research for you.
And I think
Notebook LM is something everybody should try out just to see it because I use it all the time. It's super useful. And then the last is the idea of software on demand,
and that's, you know, generating small scripts and and lightweight pieces of software you can use internally,
as well as all the way up to your vibe coding,
you know, an MVP
of, you know, software that you wanna test out.
Or, you know, if you're using it for investor meetings or something like that, you can say, hey. We've generated an MVP. Go around and play with the minimum viable product, see how this works,
or to test your own ideas out.
All those are great. If you're gonna make production software,
you should think about, you know, getting a developer.
So
I think
that's about it for now. I think, you know, all these things,
this is this is an overview
that
is gonna help you in the long run.
And, you know, if you're fairly familiar with AI already, hope this is good refresher for you. And
I think we'll leave it at that.
Slap like, leave us a review,
Tell your friends.
If you love the podcast, we got some great guests coming up,
and we'll see you next week.