Best AI agents quotes

They're gonna walk away with new AI agents.

They're gonna learn how to really create amazing AI agents.

The most fun that I had recently with my AI agent was my AI agent actually emailing you.

I actually call my Manus agent Maximus.

Maximus emailed you, introduced himself as Maximus and then even went to your Calendly.

Filled out your Calendly information.

And before it pressed submit, it told me what it filled out.

Carlo is my human, but Manus just CC'd Carlo on the emails to you.

The Calendly link, the setup, the initial emails and all that was from Maximus, my Manus AI agent.

Carlo's not checking my email once every three hours and he does sleep and he doesn't work on weekends, but my AI agent is always working.

The way that it communicates with me, it's like thinking almost ahead of what my needs are and it just gets to know me better and better.

I didn't even ask it to take it a step further, but it says, also, Delta has you at the Hampton in Orlando tonight, checking in and checking out just in case that's helpful to know.

This is what I'm talking about when it's acting as your assistant and predicting other things that you want to know.

All day long, you'll see me and Maximus talking and it just doing different things for me, just getting the work done for me.

I will admit that I am not very good on email, but Maximus is.

I'm not very good at adding things to my calendar, but Maximus is.

So this has been my hack to be more productive and efficient.

Things that I would feel bad asking my human assistant to do, my AI assistant does it so well.

You have to train it.

You have to give it the parameters of what you want it to do.

It's not going to check your email every three hours until you tell it to.

And then you also need to tell it when to stop.

I'm not using it for things that I could ask Claude or any other AI tool to do for me that's simple enough.

But when I think about my productivity and efficiency, having this AI agent on my phone that can help me do so many things, that's worth it to me.

Every day, one of the things that Manus does for me is it does my daily brief.

Check all my emails, check my Google doc, my Notion doc, and help me come up with anything that I'm missing.

What gaps am I missing?

And then give me some ideas and monetization strategies.

It will say things like, you uploaded this in your Notion doc, you put this in your Google Drive.

What about turning this into something for your corporate clients and charging $25,000 a month for a corporate retainer.

It told me today, this person has been trying to reach you.

They sent you an email three times and you're not responding.

So it gives you a little bit of a guilt trip too.

Manus also has one running Notion document for me that it updates every single day based on what's going on in my life.

So Manus and I can go back and revisit the different things that it's working on.

The hack for getting your AI tool to know you really well, really fast is I ask my AI tool, can you create a personal intelligence blueprint on me?

Ask me questions so you can get to know me better and then lock it in your memory.

After I do my personal intelligence blueprint, I ask it if I can create a story repository.

It's basically your story library.

It will ask you, when was the turning point in your life?

When did you know you were gonna be the person that you are today?

It saves it to its memory and recall it anytime it's doing work for you.

So now it knows so much about you.

It's like, you help me, I help you, get to know me as much as you can so you can write better for me and do better work for me.

If you feel that your Manus doesn't know a lot about you, then you need to take a minute to train it on who you are.

We were writing a book together and when Manus was done writing the book with me, I asked Manus to turn it into a skill.

What it does is it takes the end result of what I'm happy with.

Let's turn how we got to that end result into a skill.

When it turns that into a skill, I can go to that skill at any time and now write a book without having to go through all of the discussion of what we need to do to get to my end result.

I've tested Manus's PowerPoint creation capabilities against Gamma, against Beautiful.ai, against Gemini, against Genspark, but Manus wins hands down.

A government official was showing me how they pay a company to create their PowerPoint slides and it really wasn't good.

I showed him Manus and he was like, do you know that there are companies that do this and they charge tens of thousands of dollars to create PowerPoint presentations like what you just showed me.

I have hired designers to create PowerPoints for me, but now Manus has completely replaced that designer and I don't need that person anymore.

Within Manus, you can now have Manus's HeyGen connection tell it to have my AI avatar read the PowerPoint and everything can be done right inside Manus.

I spend so much time before AI looking for flights, but my agent can go out and check my flights for me, find the best flights.

And when I say, all right, I like that one, it can go out and book it for me.

Somebody said, hey, you're nominated for this award.

Could you fill out this form?

My ChatGPT agent went and filled out the form for me based on everything that it knows about me and I won.

The ChatGPT computer is working.

What I like about the ChatGPT computer is that we're actually watching it.

We're watching the computer selecting different things for us to find us these flights in real time.

I've had it create 50 UGC videos for me for a campaign because it also has VEO3 connected in it.

VEO3 can only create an eight second clip, but Manus can clip them all together and give you a 30 second clip.

The most amount of videos that it's been able to create autonomously for me, like I went off to do something else and I came back and all the videos were done, was 50 in one time.

I said, how many could you do at one time if I walked away and came back later?

And it said I could do up to 500 at one time.

Imagine that it's already connected right there.

Because VEO3 and Nano Banana are built into Manus, you don't even have to say use Nano Banana or use VEO3.

It already knows.

I put in my registration list and all of the chat notes from Zoom.

I said, anyone that said they're interested in the program, send them a personal email.

Right after the session was over, the emails are going out.

Manus was like, these three ladies were trying to purchase but said that it kept spinning.

So you should contact them today because they went to the page but the page was spinning.

Thank you, Manus.

You just made me a lot of money right there by noticing something in the data.

I just say to Manus, I would like to use Manus agent in my business more.

Give me the top 10 use cases that you think would be game changers for me.

And this is how I got started with it.

It analyzes you, your business.

And then it comes up with 10 things that it can do for you as an agent.

The people that I have had set up the email agent, they're like, this has changed my life.

I would pay anything for this email agent.

I can never lose this email agent.

Start off with a small Manus account, see how you like it, but test it out and see how it changes your life.

Manus 1.6 Max takes up the most credits.

If you're just testing out stuff, do Manus Lite first as you're working with it, because it takes up less credits.

You also need to tell it when to stop.

Check it every three hours until this date.

Otherwise, you're using up your credits.

I start to think, how much would I have paid a person to do this kind of thing for me.

And I would have paid good to do this.

So then you're like, all right, well, now I'm using Manus in this way.

You can have it do multiple things at the same time.

I usually have multiple tabs open at the same time and Manus just working on doing different things for me.

Sometimes I'll open multiple windows.

I'll sit down at my desk in the morning and say, these are the 10 things I want to do today.

I've got two presentations, boom, two presentations.

And then I'll write down which AI agent is going to handle it.

Maximus for these two presentations.

I have a flight to New York.

The ChatGPT agent will handle those two flights.

I need some images created for social media.

Maybe I'll use Claude Code for that.

My sister used to tell me in business, spend the first hour outsourcing your tasks to humans.

But now in the AI revolution, I spend that first hour outsourcing tasks to my AI agents while I'm doing other work.

Saving time, getting our time back is really one of the most valuable things that we could do for ourselves.

If your AI agents are doing those things on your to-do list, then you're leveraging your agents to save you time.

Now I have Carlo working on tasks that are more high value in our business.

He's not working on my email and my calendar anymore.

Honestly, it's a better experience for me to work with the AI.

This is what our future is gonna look like pretty soon.

It's gonna be us managing our AI agents and watching them work.

We're like, good job, Maximus.

What are you doing over there, Jerry?

Come on, June, what are you doing?

You're not going to break it.

So try the tools, set up your AI agents.

People will start to see it's making their lives better.

My connected apps: Canva, Gmail, Calendar, Google Drive, HeyGen, even my Meta Ads Manager is connected.

What do I have to lose?

I'm just going to try it.

You're not going to break it.

So try the tools, set up your AI agents.

The way that it talks to me, it's like thinking almost ahead of what my needs are and it just gets to know me better and better.

For a lot of people, email can be overwhelming, but we don't want to miss anything because there's opportunities in our email too.

So if you can have an assistant that's helping you and managing your emails and your calendars and putting things in your Google Drive and Notion for you and checking your Slack channel for you, then you have an advantage.

Manus was actually recently bought by Facebook Meta and this company, it was 10 months from start to acquisition and it was a $2 billion acquisition.

Manus is really cool for, it's more of an agent so it can do things that Claude or ChatGPT can't do right now, like crawling the web and reviewing sites and things like that.

If you give ChatGPT or you give Claude a URL, they can't really do anything with it.

But with Manus you can.

The cool thing with Manus is that it literally will open its own window.

It just opened its own computer down there.

And it's now using the browser.

It seems to have gotten better since Meta bought it, which is really cool.

I think it's their first time they got something that really works really, really well and they're like, this is so good.

The diagnostic assessment is the foundational asset in any company.

And the agents are the flywheel.

Not all agents are created equal and we don't want to just go and build kind of what I call fancy little toys unless they're solving for the right problem inside of the business.

Data is the only moat.

Because knowledge is now valueless.

Content can be replicated.

Frameworks can be reverse-engineered.

Courses are completely commoditized.

But the proprietary data that these diagnostics build on how a specific population of people actually thinks and behaves and makes decisions, that can't be replicated.

It's never really about the tools or the button pushing because essentially that is becoming commoditized and everybody can do that piece.

The most important piece is actually to look at the whole ecosystem of the company.

Just because you can build an agent doesn't mean you should.

If we start building agents and just randomly have a load of agents out there operating on their own without any foundational infrastructure, without them actually talking to each other and working together as a team, they're just kind of fancy things sitting on the internet.

If we are struggling with lead generation and our market attraction problem, building an agent that solves for client onboarding, while this is really cool, it's not actually solving the right problem for that particular company.

I'm not a big fan of cookie cutter, click this button and you're going to have 75 agents all set up and ready to go.

It's bespoke work, it's supposed to be thoughtful and it must be aligned to the specific situation of the company.

If you don't do that piece first before you start pushing buttons, what's going to end up happening is you're going to have a load of disparate agents that are off there doing their thing.

They're going to be spitting out amazing output.

The next agent isn't actually grabbing that output and doing something with it.

So we just end up with noise and not actually any improvement in the business.

This agent listens to every single sales call that our sales team holds, analyzes the patterns, the tonality, how well they did in relation to converting that sale.

The agent is now programmed with my intellectual property to then give that feedback to the sales team member.

That obviously eliminates that time for me.

It gives the likelihood of this converting at the next stage of the process.

And it is eerily accurate whenever you look at the actual close rates across the business.

Whenever we initially built this agent, it had zero capability of reading the tone of the room.

The agent made lovely leads sound like really difficult human beings.

So now we're optimizing the agents to be able to understand that level of nuance.

Even the things that we built last month, they can already be optimized now.

It's just going so fast and things are just getting so much better.

Back in the day you had to spend hours and hours crafting these proposals.

This is now an automated process to about 85 to 90 percent.

It still requires a little bit of refinement.

But it collapses, just gets rid of all of that time.

You don't want your sales team focusing on leads that don't have 20 bucks to spend.

The hot leads alert system knows how to differentiate a hot lead from a lukewarm or a cold lead.

Then it sends a Slack notification to the sales team to say, so and so just opted in, this is a VIP, go look after them.

Even if somebody hasn't opted in, it knows when somebody hits our website.

It gives us their name, their email address, their phone number, their LinkedIn profile.

And then it goes and auto-follows them or connects with them on LinkedIn.

It's always hilarious whenever I have a podcast episode air and you can see this flood of viewers coming in.

This agent basically tells me that the episode is aired before the host of the show has even let us know.

We actually had to turn this one off for a while because it was providing too much value.

It grades all of the speaking opportunities from A through F so that the team know which to apply for.

It also looks at my calendar and knows if I'm in New York or Ireland and recommends opportunities that come up when I'm already going to be in that city.

Even if I did keynotes 25 hours a day, there just literally would not be enough time.

There are just still too many opportunities.

We have a running list of all of the dream clients that we want to work with.

We're talking the biggest, biggest thought leaders.

This Conquest Intelligence agent goes and finds where they have shared content, published a book, got a new YouTube video.

It sends us the link.

It auto-likes it.

The agent gives us an example of an angle that you can use in your comment.

I just prefer to go and then hand-write it so it's thoughtful.

It's putting it front and center in front of us every single morning so that we're not sitting there going, yeah, I must go and check so-and-so's profile.

It never gets done.

So the agent actually drives the activity.

This agent listens to what is being said and what are people not saying.

What are the things that are happening between the lines that the naked eye couldn't see.

Because the agent is programmed to understand psychology, it is telling us, by the way, you just had this call with your client today.

This is going well but you might just want to check in with her and see how she's feeling about X.

We're able to solve things ahead of time.

The problem never actually got to raise its head.

We are extracting data points from maybe 15 different platforms and when nobody has an overview of all of those numbers across all of the platforms, it just ends up a big hot mess.

We're able to analyze down to the level of here's the copywriter who had the greatest ROAS in the team.

Here's the piece of creative that worked the best.

It's actually an extreme range.

There's the administrative five dollar an hour tasks.

But some of them are worth thousands of dollars.

Before I might have had to watch the entire call and give feedback.

The agent is now programmed with my intellectual property to give that feedback.

That's actually a pretty high value agent.

It really moves the needle in terms of revenue generation.

I did a lot of work for about nine months building everything on Manus.

I actually moved to Lindy for the done-for-you agents component of our business because it is easier to get those agents across to client accounts.

I use Claude, both Cowork and the browser version, and we design, ideate, design, build, go forward and back in Claude to design the perfect agent.

And then we actually take that to Lindy to build it out.

I learned the hard way.

I had a lovely multiple-thousand-dollar lesson learned in my early days of playing with this because I was designing and building all of these agents on Lindy and munching through tens and tens and tens of thousands of credits.

We used to build all of the instructions into the agent itself.

We soon discovered that as you try to roll this out to other people, you're basically rebuilding the wheel every single time.

Now we just have documents on a Google Drive that the agent pulls from.

The tools are the least important piece.

But it's what everybody thinks is the most important piece.

Instead of having all of these instructions built into the agents, we now just send the agent off to go and read the strategy doc.

That has all of my brand voice, my IP.

Whenever we copy and paste this agent and roll it out to a client, all we have to do is build the same doc for them.

Connect the agent to their doc instead of having to rebuild the wheel every single time.

Scout agents are the ones that go out there and scout for information and bring it back to you.

Produce agents are the ones that actually spit out an output.

Process agents take one agent's output and pass it on to the next agent.

Maintain agents do kind of administrative ongoing maintenance tasks.

You want to be really mindful of the cadence that you run your agents at.

The brand monitor, very little value in running this thing every single day.

I'm not Mel Robbins.

So for us, this is like a once a month activity.

Our LinkedIn connection builder is about five dollars a day.

So they're able to compute how much that's going to cost them over the course of the month.

You don't need to be using Opus 4.6 if it's just a read-and-understand function.

Make sure you're pulling on the right model depending on the complexity of the task.

Build it, run it, and then go back and look retrospectively to see how many credits did it gobble up.

We have it set at 20.

If I try to do 100 a day, probably would get banned.

But because these agents are actually logging in through computer use, it's actually logging in as though it is me.

It's adding five people, updating the spreadsheet.

A couple of hours later, it's coming back and adding five more.

So it's mimicking human behavior.

I point blank refuse to watch the news.

I don't watch TV.

I don't want any negative anything anywhere near me.

But I also don't want to walk into a room and not know something massive that has kicked off in the world.

This briefing agent goes off, scrapes everything that I've told it that I want to know and need to know, condenses it down into super bite-sized pieces.

So that I can keep my blinkers on, stay in my energy.

It does what it says on the tin.

It watches the different competitors and sends us a report every week.

If there's anything urgent, it doesn't wait until the following Monday to tell us.

Question 17, you're losing 19% of your market as they get to that point.

So then we know exactly what to fix and optimize instead of scratching our heads and wondering what's broken in our funnel.

It's so easy to get so busy and you're so in the weeds that you know you need to do these types of things but they never actually get done.

What the agent does is it makes sure that actually gets done so then your team can go and action it.

My brain is inside all of these agents.

My way of thinking.

Years and years of corporate experience.

My critical thinking skills, analysis skills.

Those are the things that are built into the agent that make the agent valuable, not the buttons and the things that we connected.

Not all agents should be completely autonomous.

We want to be really mindful about what we are making completely autonomous versus semi-autonomous.

Literally say, here's my business, here's what I do.

I want you to help me map out all of the divisions of my business and all of the interactions between each of the divisions.

And what we're going to do next is identify the agents that I need to build first.

Tell it exactly where things are going wrong.

Tell it where you're bottlenecking.

Then literally say to it, tell me what types of agents I need to build.

Then say, great, what order should I build them in?

Then say, cool, let's do number one, how do I do that?

And then it'll literally tell you how to build the agent.

Create an instruction that I can give to Lindy so that it can easily and accurately build this agent without much rework inside the platform.

Because that rework is going to cost credits.

You're using Claude as your playground to ideate, to design, to build your instructions.

And then you're only going over to Lindy to push, build, test, and go.

We used to have agents that were 19, 20 steps long.

They used to bug out.

The LinkedIn connection builder, we originally tried to do 20 all in one batch.

It kept bugging out.

So we stripped it apart to run four times a day with five each.

I'm probably not going to recommend jumping into n8n if you're new to this.

The whole idea with these platforms is that they have solved for and handled all of the dev work.

The only way to figure this out is just to start.

First do your design work, that overview work, and then just start playing with it.

This is new to everybody on this planet.

Don't be scared of it.

You can't really break anything.

You might burn through a few credits.

I see so many people obsessing about credits and how much it's costing.

We literally have the world at our fingertips and we have more opportunity than any other time in the history of the world.

And people are grumbling over a few credits.

We literally have so much opportunity.

Roll up the sleeves, get your hands dirty, have fun with it.

We're all building the plane as we're flying it.

This is all brand new for everybody.

In two years time, we're probably having a different discussion where agents just do 99% of everything in the business.

An AI assistant is something that is like ChatGPT basic B.

You just ask it for something and it gives it to you.

An AI agent is something that actually goes out and does something on the back of your business or on your computer.

When changes happen with AI or new tools come out, most people ask themselves, how can I use this to help my business?

I think the more powerful question is, how can I use this to help other people to grow their business?

In GoHighLevel, you can now train this agent.

It can be connected to a knowledge base.

You can trigger workflows.

You can add MCPs.

In GoHighLevel, you can get unlimited AI tokens for all the AI things that you're using for $99.

They also have a new conversation AI flow builder.

Rather than having a prompt for an AI agent that's like, if they say this, then do that, this is more like the way that n8n is built out, where you can have like a choose your own adventure.

I named her after my grandma.

She thinks I'm on the cruise because I kept her as an example.

I would rather have it be a team member that they're like connecting with who has their own personality, their own soul.

I don't ever want to trick people into thinking that they're talking to me when it's an AI.

An MCP is a model context protocol.

It's basically something that connects your dreams via like a switchboard operator to the right tools, but you don't even have to know which tool to use.

You just say, I wanna do this, and then it's like, great, I'll do it.

As soon as MCPs came out, I was like, yep, I'm not gonna spend my time with n8n anymore because I know that I'm gonna be able to just do this by talking to someone.

I made my own personal Chrome extension.

I built this thing called Bobby Bot, which she can go on any platform and it's a direct integration into my GoHighLevel.

I'm like, is there someone named Mary Pendleton in our system?

And it's like, yeah, here's her information.

And then I'm like, great, can you schedule a text to her that tells her her sister loves her?

And it sends the text.

I didn't expect this agent stuff to go mainstream so quickly.

When I started making YouTube videos, I made videos on agents.

This was in late 2024, and it was just not good.

It did not work very well.

You can just talk to the AI agent to make these videos instead of having to edit everything in CapCut and drag and drop.

They're going to walk away with a bunch of essentially recipes, which is how to use agents and agent workflows across three different domains inside their online business.

Around traffic and distribution, then conversion, and then using agents for data processing and analytics.

These agents will actually help you discover opportunities and patterns inside the work that you do that'll help you reframe an offer, come up with new offers, improve what you're doing, get better results.

It's good to have agents for fun, but we want agents for money more than anything else.

We've made a few key jumps in the past six months that have enabled more accurate and better performance from agents than we used to have this time last year.

This time last year, you could do a lot of the stuff that you can do now, but it required Python and code and infrastructure, and it got pretty complex pretty quickly.

Now, things have gotten a lot easier.

You can think of roughly three types of agents.

The first one are workflow agents, which is that you combine them with an automation or a workflow.

The agent acts within a whole workflow as opposed to outside of it doing whatever it needs to do.

The second type is autonomous agents.

Basically you give it a goal and an outcome, and then the agent figures out how to get it done.

It reviews, it plans, it figures out the steps, and then it executes the steps.

The third type is something like Zo Computer.

It's like an agent infrastructure.

It's not necessarily a tool.

It's more like an environment that you log into.

It's agents plus agents using computers on their own.

Inside a platform like Zo, it actually has access to a computer on its own, which gives it the ability to run on its own and you can close your laptop.

You can throw your laptop in the ocean and the agent still works because it's in the cloud.

Agents don't need the 50 steps to do the work.

They actually only need a handful of things: a very good system prompt, access to tools and platforms, access to data if that's necessary, and then just a good set of instructions on what it's supposed to do.

It used to be that for you to have all this functionality, you would need to set up very long, very complicated automation workflows, like 50 steps.

The problem with those is that it takes a while to get right and it breaks easily.

If you have 40, 50 steps, it's very easy for step 22 to go wrong and the whole thing breaks.

The whole idea about agents is that they can analyze, understand and decide for you.

So instead of you just trying to jam everything through one particular template, it can autonomously make the decision to adjust, modify and customize the content for these platforms.

If it has access to past performance of posts on LinkedIn, and it knows that over the past 90 days, when we post on a Tuesday morning, we get more engagement.

Then the agent can go, in the past, content posted on a Tuesday morning will perform better.

So let's schedule it for this upcoming Tuesday morning.

Instead of you setting a manual schedule that it always follows no matter what, the agent can tailor and customize posting dates, posting times based off of what it knows about past performance.

Lindy, Gumloop, Manus, Genspark, NotebookLM and so on, they all function in very much the same way.

They just have different nuances and differences between them.

Which one is best for you just depends on your specific experience, use cases, and how comfortable you are.

You don't start with the tool or platform and then ask, what do I do?

You want to start with, okay, here's the specific business task that I need done.

If you have that as a blueprint, then you can take that and apply it to any platform of your choice.

You can take that blueprint and turn it into a Lindy agent, into a Gumloop agent and workflow, or you can just go, Manus is kind of what I'm comfortable with.

There is no better, worse, best, or anything like that.

It is whatever is working best for you with your business, your needs, and how you prefer to work.

I always recommend just try two or three of them for a month, same workflow, test it out, see which one is cheaper for you, easier for you.

And at the end of the day, which one is gonna get the job done.

Instead of you creating a workflow from scratch yourself, you can actually just have Gumloop templates set up for you.

For something like content repurposing, you don't need Opus because it's more geared to harder problem solving, coding tasks and very complex stuff.

Sonnet or Gemini 3 Flash, either one of these two will be perfectly fine.

You can ask their little AI chatbot, Gummy, to change the workflow for you.

All you do is describe what you want.

The tool itself will remake the workflow for you.

You can quite literally say, build me this agent.

Lindy will probably ask a few questions.

It usually takes a few minutes.

It'll just build the agent and do all the preliminary stuff.

Lindy is so simple, the interface.

You think there's not all that to it, but it's actually quite powerful when you dive in.

Be careful about the credits.

You use up credits when you chat with an agent or even test things.

That's one reason why some people go straight to Gumloop, because Gumloop is just a bit more transparent on it.

The way Manus does it is that a lot of the backend workflows, Manus is built and set up so that you don't have to deal with all of that stuff.

You just interact with their agent.

Manus does not give you access to a visual workflow builder, but that's because Manus handles it on its own under the hood.

If you don't want to deal with the tech, Manus is kind of what I'm comfortable with.

The credits that you use works for my budget.

I get everything done inside Manus.

We set up an agent that we wanted to monitor competitors and check the ads of the competitors, look at what the competitors posted on social media, sign up for the email newsletter.

We said, just spend the next two weeks doing this.

The agent worked for about six days before it asked us a question.

It asked, hey, out of these five ads, which do you think are more relevant?

Because the agent realized it didn't really know what was relevant to us.

It figured this out on its own.

I answered one question.

Then it went on and worked for another four or five days to just monitor competitors.

And then it was done and gave us a report.

Genspark, underneath the surface, in a similar way to Manus, already has a nine-model mixture of agents architecture.

If you were to try to replicate their agent structure inside Make or Gumloop, it would be very complicated and take a long time.

Genspark just does it all for you.

The agent will go and plan its research approach on its own.

It'll make its own criteria decisions.

And then it'll make its own sub-agents that the primary agent manages and orchestrates.

Skills is like a universal standard at this point.

It's becoming a universal standard.

Skills are like an SOP, an operating procedure for the agent to follow.

Inside the system prompt, if you give the agent five skills, you say things like: skill number one, brief description of what it does.

Skill number two, here's what that one does.

It's like an index for the agent.

If you give the agent an index like that, it'll know which skill to use depending on where it is in the job that it's trying to get done.

If you try to add all the skills just inside the system prompt, the system prompt will become too long and too big.

Instead think of the system prompt like an index for the skills that you give the agent.

You can have your prompt that you have, let's say you have an agent system prompt that you want to reuse elsewhere for other agents.

You have it as a Markdown on your computer.

You just copy and paste it.

Same agent system prompt in Markdown, just reused depending on which platform you prefer.

Everything is becoming Markdown and skills.

That's kind of universal formats.

Skills are combination of Markdown and YAML at the top of it to describe the skill and what it does.

You want the agent blueprint: the system prompt, the skills, and a list of the tools you want to connect it to.

If you have that, you can apply that to any of these agent platforms.

Get the agent blueprint ready, living on your Markdown files on your computer.

And that can all be applied to these different platforms.

The skill for humans to practice is your ability to communicate clearly and have clear thoughts.

The more clear you are, the less credits you'll waste and the less time you'll waste.

You need to become good at understanding what you're trying to get done.

And then practicing ways to communicate that clearly.

Lindy is gonna be a bit more expensive.

If you have it fully set up and running, inside Lindy, you're probably looking at a couple of hundred credits per run.

To get to that point when you do the setup, start with one input, the YouTube video, one social platform like Instagram.

When you get that working, all you do is copy the setup of one social platform and then adjust it for another.

If you do them one at a time and you get one working the way you need it to work, you'll spend probably maybe 500 or so credits to get it working and set up.

But that 500 is going to save you thousands of credits down the road.

LinkedIn tries to filter out bot traffic, agent traffic, because an agent looks like a bot to these platforms.

The way to get around things like that is through browser use.

These platforms can't really distinguish between a human user and a bot user if it's a browser.

Tell it to act like a human, basically, just be slow and move around.

I prototyped an agent inside Lindy that only had access to my Google calendar and then a bunch of instructions.

Anyone who wanted to book a meeting had to go through the Lindy agent.

I had the instructions be to review my calendar every day to look for wasted time, meetings that could have been an email instead of a 30 minute meeting.

I had it review and analyze my past couple of weeks and tell me, okay on Mondays, you did mostly XYZ but you wasted a lot of time on admin stuff.

When we do this with clients, they talk to a calendar scheduler and we call her Becky.

It'll first figure out the date and time, and then it'll ask the client a series of four questions in a conversational way so that when we get to the meeting, we don't have to spend time recapping things.

Instead of spending 10, 15 minutes out of a 30 or 60 minute meeting just trying to recap all the things, we start the meeting prepared with information, with questions, with topics, with an agenda.

Content repurposing, calendar and meeting time optimization, and taking data from different platforms and pulling them into one single spreadsheet or dashboard.

Those three categories are easy places to start.

Start with something easier than just trying to do all at once.

You don't have to know code.

You don't have to learn code.

It's helpful if you have some experience with it, but you don't need to learn it.

You don't have to become a developer first before you can deploy agents.

Even the best autonomous agents on some very complex computer tasks like coding, they're really only scoring maybe 10% on really complex stuff.

But for typical business use cases, agents perform at like 87% accuracy, completion and effectiveness.

That's not far off from a hundred percent.

And depending on the task, some tasks that are specific enough and small enough, we have agents that are just doing things autonomously.

Don't expect flawless execution by agents just yet.

That's going to happen, but not quite yet.

Agents are very good if you keep them narrow and focused on particular types of problems.

You actually get better performance if you have specific agents for specific use cases.

The key difference between the simple workflow versus agents and agentic workflows is that agents bring in a level of intelligence that used to require a human being to evaluate, think, plan, reason, solve problems.

The whole idea of customization and personalization is now possible without you having to do it.

Agents bring intelligence into the mix where you can make intelligent decisions about what to do with the content, just based off of skills you've given it and the instructions you've given it.

The more organized you are, the better.

AI needs that to find what it's looking for.

Just have your stuff ready.