Transcripts
1. Lesson 1 - Intro : Finished running a research task in Chap ChiptV agent mode, and the results
have blown me away. Think a ranked list of blog post topic ideas
that are low competition, high demand and
worth writing about. All tailored for a beginner
friendly vegan cooking blog. I've only used regular
ChaphipT before. Agent mode is going
to feel like you've switched from bicycle
to an electric bike. The core idea is the thing. You're still asking questions
and getting answers, but Agent mode can handle multi step research projects on its you can browse
public sources, cross check data, and present results in
neatly formatted tables. Now in this class,
we're going to break down exactly how to get
agent mode to work for you. I'll show you how to define
the task, set boundaries, and give it a
scoring system so it knows how to prioritize
the results. By the end of this course, you'll write a professional
research prompt, run it in agent mode
without interruptions, interpret the results, and
decide what's worth keeping. Example will focus on finding great vegan cooking blog post, a topic that I personally know
nothing about, I'll admit, but you'll leave with
a framework you can reuse for any non
commercial research. And before we dive
in, remember to aim for momentum, not perfection. This is a skill that gets sharper with practice.
Take your time. If you enjoy this
session, I've got other beginner friendly
classes on mastering the basics of Chat TPT or online profile so feel
free to check them out. And if you have any questions
along the way, reach out. I enjoy hearing from students
and seeing what you create.
2. Class Project: For the class project, I'm
going to ask you to write a structured agent
mode prompt with roll, constraints, research methods, scoring criteria,
and output format. Then just share your prompt and your top three results that came up in the project gallery. Remember, aim for
momentum, not perfection. The point is to practice and
3. Lesson 2 - Understanding Agent Mode: Here's the thing. Agent
mode, chap chibit, same chat box, same layout, but under the hood, a
completely different beast. Think of standard Chap chi BT is a very knowledgeable friend who can answer your
questions instantly, but only based on what
they already know. If you ask something
that requires checking multiple sources, they might add confident, but I have no idea where
any of that came from. Agent mode, on the other hand, is more like a project
assistant who can take a brief and research across
probably available sources, and come back with a
fully formatted report. Citations, comparisons, and even tables,
it's not rushing. It's happily taking
20 to 30 minutes to pull together
something comprehensive. Autonomous research. You can plan and execute
multi step searches without you micromanaging.
Fructured output. You can tell exactly how
you want results presented, tables, rank list, summaries, the work, evidence links. It can show you where it found its information so you can check the credibility for yourself. Public source focus. If you set it up that way, it'll only pull from places that don't require logins,
meaning no interruption. For example, researching vegan Cooking blog
topics, this is a dream. Instead of trolling through endless search results itself, you can have agent
mode, scan food blogs, trend reports, search
interest data, and come back with top topics ranked by demand and
competition level. We'll still need to give
it a great prompt because agent OD is only as good as
the instructions you feed it. That's what we'll work
on in the next lesson, how to plan the research task so you don't waste a single run.
4. Lesson 3 - Planning the Research Task: Okay, before we hit
Go in agent boot, we need to give it
a job description. I can actually follow. If we just say find me blog topics, it's going to shrug, pull
together something to the egg. So let's slow down and plan. When I'm setting up
a research task, I think about it in three parts. You've got Role. Who is AI pretending to be? In our case, it
could be as follows. You are a blog content
strategist who specializes in identifying
trend low competition topics. For new writers in the
vegan cooking niche, this frames how it should
think constraints. Only use publicly
available sources. Avoid oversaturated
topics like vegan, chocolate cake that are already
most likely everywhere. Looking for topics with
long term relevance tends to be what works. This is how it knows
what makes a topic good. You might ask it to rate its
idea based on search demand are relatively few
high quality articles on it, e for beginners. Could a new blogger write about it without being an expert chef? Taking the time to set
these three things role, constraints, and scoring, we're basically giving
Agent mode a map. Without it, it might
wander all over the place. In the next lesson, we'll take
these ideas and turn them into a structured prompt that's
ready to paste into AGBT. That's where it starts
to get really fun.
5. Lesson 4 - Structuring Your Prompt: Take off this like
writing a recipe. You wouldn't just say Mk dinner. You'd get ingredients, steps, and maybe even plating
instructions. Same deal here. Here's how I like to break a
prompt into clear sections. One, le definition.
Here's an example. You are blog content strategist who helps vegan food
bloggers find trending, low competition, high demand topics that are easy
to write about. Two, constraints.
Here's an example. Only use publicly
available sources, avoid topics with thousands of existing recipes on
major food sites. Prioritize ideas that will stay relevant for at
least six months. C research methods.
Here's a quick example. Review recent food trend
articles, search interest data, and popular under
certain log categories, cross check findings to confirm demand and low competition. Number four, scoring rubric. Rated each topic 1-10
based on search demand, competition level,
beginning friendliness, and number five is output.
Here's a quick example. Present results in a table
with columns for topic, demand, score,
competition score, beginner friendly
score, and source. When you combine all of this, you get a prompt that leaves
nothing to guesswork, agent mode will know the role it's playing, the
rules to follow, the way to research,
how to judge results, and how to package
it up for you. Now in the next lesson,
we'll actually run this vegan Cooking blog
research task in agent mode so you can see it
working in real time and get a few tips to avoid
getting stuck halfway.
6. Lesson 5 - Running The Agent Task: Okay, welcome to Lesson five. We're now going to
run the prompt and see what results we
can come up with. Fingers crossed, can find
something that we can use, but if you don't get the
results that you are happy with first time around,
do not despair. Keep trying the second
or third attempt might fit your
goals even better. And here's a quick tip. If you're not sure how to
structure the prompt exactly, you can utilize ChachibT itself, take your basic idea
and change that into a highly detailed
usable prompt. That PT five agent
will be able to saw, understand, and work with without too much back
and form. All right. Let's jump into my hattPT screen and see what we can conjure up. Okay, I've got my prompt
I'm going to paste it here, but what I'm going to
do is I'm going to ask ChachiPT to turn
it into a prompt. ChachPT agent can use first, and then I'm going to
go into CathiT agent and use that final prompt. So this is what I'm going to do. I'm just going to paste
this prompt here, and I'm just going to say, is for GPT agent to work with. Just going to go ahead
and run this and we'll see what comes
up. There we go. So here is your
cleaned up version, ready to copy straight into Chap JPT agent for consistent,
high quality results. We've got our role
definition, constraints, research methods, scoring
rubric, output format. I'm going to copy it
pasted into a new chart. You just click on
this little plus, and you can see there you've got different modes you can use, but the one I'm looking
for is agent mode. Just go to go ahead and
paste my pump here. Now, agent mode will
take some time to run. It can take a minute, it can take 20 minutes. So what I'm going to do is just let you watch the initial run. I may cut some of the
more monotonous areas up until the point where we
have actually got a result. So currently, it's setting
up its own virtual desktop. The way that works is in order for it to go in open browsers, browse websites, which it
actually does, which is crazy. It's interesting to watch an employee or a colleague
doing actual work, but it can't do that
on my computer, set up its own virtual computer so it can execute these tasks. Now, you as the prompter, you can see it's virtual
browser that it's opened, and you can actually see it
running different tasks, going into different websites. And what's good
about it is agent is actually letting you know its thought process
as it's continuing to go with these tasks.
You can see here. I'll search the article about the most Google
vegan recipes. You can see it's going through different social media posts, different blog posts,
different searches in order to get all
this information. Whereas before, during the
initial stages of ChatBT, you would have had to
manually keep prompting it, guiding it to go and do
all these different tasks. Now you're essentially
just ship this away and letting agent do
all of this for you, which is absolutely crazy. What's worth noting as well is if you click on
the three dots here, you can see activity monitor. All of this activity
that it's doing, you don't have to keep up
with it as it's going. Once the task is complete, you can monitor what
activity took place, what it did, where it went. And the report that
it's going to provide, it will provide an summary where it tells you
what it's done, it's come up with, why
it's going to work, those types of details, linking back to the initial
prompt that you included. You can also stop
this at any point if you feel like the
agent is going onto website star really relevant to the result that
you're trying to get to, you can stop the process
or give another prompt, and it'll rerun that process. You can also take
over the browser. This is especially useful
if ChachiBT agent needs to go into sites that
are behind a login wad. An example, I tested
this with Canva, hook up my password, so it
was highly embarrassing. Went onto Canva, found the website and it asked
me to then take over and put in my login
details so it could go into my Canva account
and do some work for me, create thumbnails,
presentation, whatever it is. You can do that
with a huge amount of desktop based websites. Another site that it did
this task for was YouTube. Again, it went onto the desktop version,
asked me to sign in, and it went to my
YouTube studio, looked at my analytics
on pointing side, it is still early stages. This agent is slow. I asked it to do an audit
on my SEO on YouTube, and it was taking ages. Each video took about 10
minutes to analyze it to stop. It was doing the
task it's meant to. It was opening up
the right windows, opening up the right tab. It's just taking a while. I think we've still got a
few iterations until we get to a point where the
agent is lightning fast. It goes onto the
side you sign in, and it comes up with what you're looking for within
a minute or two. Was asking you to complete
some pretty exhausted, difficult tasks and high
volume of them as well. So I wasn't surprised it
was taking that long. If you ask you to do
a very simple task or a set of very
simple quick tasks, chances are it is
going to be quick, so just something
worth keeping in mind. I'm going to grab a quick
drink of water and come back once it's completed its
tasks and review the result
7. Lesson 6 - Reading & Validating The Results: We're going to look at
the executive summary. Then we're going to look
at the table of results. This is where the information
that we're looking for. Hopefully. Then we're
going to check if there's any source ling to make sure if
they're reputable, recent and actually
match the topic. Yeah, I does need a human
eye to confirm these things, even though it can actually tube 70 to 80% of what
needs to be done. And sometimes low
scoring idea might be a better idea as opposed
to a higher scoring idea, thinking of your
ethos, your brand, and what you want to
represent with what you do. Okay, we're going to jump into my chatter bite now and
see what we've Okay, so in total, our ATVT agent worked for 5 minutes much
quicker than I thought. The pump wasn't as complex as a would be. So it
says here below. Emerging vegan topics, balance rising interest with low competition and beginner
friendly preparation. Scores one to ten reflect estimated demand, ease of frank, lower competition scorer
is better and how approachable the topic
is for a new blogger. Not only is it bringing
us these results, it's also educating us in what
it feels is going to work. We've got my analysis. Let's quickly run
through the table. We've got homemade potato milk, creamy dairy free
milk from potatoes. We've got a demand score
seven competition is three, beginner friendly
score, one to ten, is a nine, low competition. That's good. A lot of
people looking for it. Not too much competition,
decent score. Why it's good rationalizing here why this result is good,
what I'm looking for. And it goes on there to give us a blurb on why it's
such a good pick. We've got source links as well. This is where the
human eye comes in. You just click on
those links, verify this information
is still relevant, is recent and is from
a reputable source, the result fermented cashew
cheese for Gut help. We've got a demand
score low competition. Not as good as the top one. Become a friendly,
slightly more challenging. We've got why? It's a good. Well as the website where
the the information from. I think this is key because
this goes to show for those people that believe
that with AI being here, it's going to make websites blogs irrelevant and so forth. This shows blogs that's
important because agents are referencing that information to
bring the results. If someone runs a blog
or email newsletter, this information is still
relevant because AI needs that human touch
to get us these results. You can now copy this table or paste into a Word document, favid got a summary
analysis here, 2025 food trend reports show a clear shift
what sustainable, trend, dense and globally
inspired plant based foods. This is why I've chosen
these blog posts. And finally, we've got
top three quick wins, home made potato miled, upcycled juice pop,
veggie burgers, sea moss, smoothie balls think this is what
you should go for. But ultimately, that decision
is still going to be down you can see how you could take this
process and apply it to any type of information. You need to research online, let the AI agent take
over and do all of this. It still will need
your oversight. It still will need
your decision making. B is not making the decision
for you in terms of, okay, don't do anything else. I will write all of
these for you report. You're still there to oversee
the different stages, much as you would with an
actual human employee, which I think is really, really important
to keep in mind.
8. Conclusion & Next Steps : We've covered how to
plan, structure, run, and validate a research task in agent mode using vegan Cooking
logging as our example. Now you can apply anywhere you need solid structured research. Feel free to check on my
Skillshare profile for more beginner friendly
hat TPT classes and reach out if you
have any questions. Now, it's over to you.
Victor Loyiso, Ex-Project Manager, AI Geek, Content Creator
