3 Manus AI Systems That Make You So Good It Feels Unfair

0:00 Everyone you work with has access to basically the same AI that you do. And that's the strange part. The tools got evenly distributed, but the results didn't. And I don't think the gap is

0:10 intelligence or effort. It's the job that you give the AI actually because you know what happened? Most people have hired themselves a very fast assistant. But a small number of people have hired

0:21 an entire department. And 10 years of that gap will probably start looking like talent. So let me show you what hiring the department actually looks like. This is [music] Manis. I have

0:33 built three systems with it. One researches a decision from multiple directions. One actually works through my browser and the third stops useful corrections disappearing when the task

0:43 ends. And don't worry, I've packaged the setups we're using so you don't have to take notes. I'm going to tell you where everything lives at the end. starting with the research desk because there's

0:54 always more worth knowing than we have time to find out, right? So, I want you to imagine this. It's Wednesday morning and somebody says, "We are considering entering this [music] new market. Can

1:05 you come back on Friday with a recommendation?" Okay. So, what would we want to know? Who already sells there? What are customers actually complaining about? What happened to the companies

1:15 that tried this before? And already we have a problem because Friday is still Friday. At some point time will decide how much research we get to do. So I've created a fictional company for this

1:27 example and I am asking Manis whether it should enter this new market. So here's the prompt that I'm going to use. I'm not going to read it to you because that's not why you're here. And I have

1:40 my computer here so I can do it live with you. And now watch because we now have work with lots of independent research [music] threads. So Manis can split the

1:53 job into parallel research paths. I'll open a couple so you can see the different questions it is pursuing. So it's looking into regulation and competitors and alternatives.

2:05 Interesting. It's looking at failed entrance as well. And this is why I use the department metaphor because I asked one single question. Instead of following one research path from start

2:15 to finish, Manis can investigate different parts of the question at the same time in parallel. Now, let's see what this one's doing. Okay, that's interesting. I would not

2:27 have started there. And this one's actually challenging the idea, which I like. I think it's good. I want that because if all of this research comes back agreeing with the idea that I typed

2:37 into the box, I haven't built a research department. I just built a very expensive fan club. So, one of the things that I do here is what I call the disagreement check. It's

2:47 a deliberate look for research that makes my preferred answer not stick or less comfortable. Basically, what would make us say no? What assumption is carrying too much weight or the most

3:00 risk? Or where do two credible sources maybe [music] completely go against each other and disagree? Because once you can see where the evidence disagrees, [music] you will be able to see where

3:11 the decision actually needs more work and more thought. I mean, I've done this so many times where I've thought about an idea and I thought it was fantastic and then digging deeper, I realized that

3:24 actually I needed to really work on my judgment and my grounding of that idea. So, why I care about this is because I could open every one of these tabs myself [music] and obviously so could

3:35 you. The advantage isn't the access to secret information. It's that I can spend less of Thursday collecting it and more of Thursday deciding what it means to us and our decision. Most bad

3:48 decisions are not made because people can't think. People are smart. Most of the time, these decisions are made because people don't have enough time to look. Okay, so we're done. Let's see

3:59 what we've got. I'm not going to make you read this massive [music] research on YouTube, but I basically care about four things. The recommendation, the evidence doing most of the work, the

4:10 strongest argument against it, and what we still need to look into and verify. All right. So, basically, I could easily walk into a meeting with this. I would still need to check the sources carrying

4:21 the recommendation obviously because that's my name on the decision. It's not man is taking the risk and the responsibility. But I'm walking in having inspected far more of the problem

4:31 than I realistically could have done on my own within such short amount of time. That is the research desk and the move inside it. I call it the disagreement check [music] because you don't just ask

4:42 manis to prove your idea. You want to make it show you where the idea might hold some holes. Now, in a few minutes, we're going to move from research to something very different. I'm going to

4:55 let Manis into my browser where it can actually start doing the work around my meetings. But if you're looking at this research desk thinking, Lara, there is no chance I'm rebuilding that from a

5:05 YouTube video. Good. I've put the exact setup for this plus the other two systems in a free downloadable inside our founders community. [music] I'm going to link it below at the end and

5:16 you can use this QR code here. But don't open it yet, okay? We have to finish Friday's meeting first because nobody wants this on a slide one. And this is where I use Gamma. And Gamma is

5:28 fortunately partnering with us for this video. So clearly I've got the research. I've got the recommendation. [music] Now I need to turn it into something that the people in the meeting can actually

5:38 understand. So I'm taking the research into Gamma. And the distinction I care about here is that I'm not asking Gamma to invent the [music] recommendation from scratch. We've already done that

5:49 work. I'm only asking Gamma to help me turn the recommendation into the thing that I can stand behind in the room. What I did do, however, was that I told it who the audience is, what decision

6:01 we're making, and that I want the recommendation early. Also, the evidence needs to come underneath it and the biggest risk and the next step after that. That was it. And now that I have

6:11 the first draft, I can start working on it visually. I can change the structure. I can rewrite a section. I can change a [music] layout. I can swap the visual and all of that with AI assistance to

6:23 make sure that we get the best results [music] and that everything looks super professional. And what I really like is that I can present directly from here when I'm ready because research is only

6:33 finished when the person who has to act on it can understand it. So if you do want to give Gamma a try, I'm going to make sure to put the link down below. Thank you again, Gamma, for partnering

6:42 with us on this video. Now the research is done, the presentation is done. So now let's give manas something harder. Let's move to system number two where we give it the hands. So this is next week.

6:53 I've got a lot of meetings happening. And the meeting prep sounds like one task until you actually watch yourself doing it. I don't know if you've paid attention to what you do, but as far as

7:04 I'm concerned, I check the calendar and then I remember there was an email and then I need the document we talked about last time and then I look up what changed since we last spoke and then I

7:14 probably end up with six tabs open and somehow I'm reading an article from 2024 that I absolutely did not need to. Let me know below if that's you as well. So clearly I want Manis to prepare me to

7:27 help me be more prepared, but I don't want it to give me advice about preparing. I want the preparation done. Right. So this is Manis browser operator with my browser enabled. Manis can

7:39 operate in an authorized local browser session. The one that already has the tools that I am signed into. And that means that it can work with the things that I'm already logged into within what

7:52 I explicitly give it permission to access. So don't worry, it's not going to go and roam through your inbox without you [music] giving it permission. And I can see what it's

7:59 doing. I can interrupt it. I can take over. I can stop it immediately, which I absolutely would if it wandered somewhere I did not expect. So this is the job. I'm giving it the calendar, the

8:10 prep document and very explicit [music] boundaries. It can research and it can update this document, but it cannot send, RSVP, delete or change any source records. So again, the prompt is going

8:24 to be here. I'm going to type it directly in my computer. Now let's watch. It's looking at the calendar. Okay, so there's meeting one [music] and now it's going to the email

8:36 and it found the thread. Okay, now it opened the document. And I think this is an important distinction that I would like you to notice. I did not copy the calendar event into a chat. I did not

8:46 copy the email in. I did not open a document and create five sections. Manis is moving through the places where the job already exists. Okay, let's take a look at what it put together. The

8:58 objective, the previous context, what's changed, the open questions, and what I need to read. Okay, and now it's doing the same for the next one. And I know watching somebody prepare five meetings

9:09 is not exactly peak entertainment. So, let's speed this up, and I'm going to come back once it's done. All right, so we've got the output. With five meetings, everything I need is

9:20 sitting in one place. And this is the category difference for me. A chatbot will give you an answer. This one gives you the thing done. It works on the work. I can always ask a chatbot, how

9:31 should I prepare for this meeting and get an excellent answer and then I still need to find the email and open the document and collect the links and put everything somewhere useful and prepare

9:40 myself. The answer was one piece of the job. And as you can see, the answer is becoming the least interesting part of AI because this was almost annoyingly simple. I just authorized the browser. I

9:52 described the result and Manis was able to work through the tabs. But there is still one annoying part. I did have to remember to run it. So once the workflow works, I turn [music] it into what I

10:03 call the Sunday run. Basically every Sunday evening, it will [music] run the same instruction. Again, the prompt is going to be here. So you don't have to remember anything because

10:17 Manis also has access to something called scheduled tasks which basically can continue recurring work using the same setup that it already has. So instead of getting to Monday and

10:29 remembering that I need to prepare something like this is waiting for me. Now there is a huge difference between a reminder and a result. A reminder gives you the job back to you. But this gives

10:41 you the preparation ready. [music] And the Sunday run does not have to be meeting prep, right? It can be anything. You can make it totally yours. You can make it the update for the weekly

10:52 report, checking a recurring set of accounts, preparing Monday's open actions or reviewing customer feedback, whatever already happens every week for you specifically [music] in your work,

11:03 in your job, in your company. Don't invent recurring work so that you can automate it. Find the recurring work first. Okay. So, we've given the department more eyes and now we've also

11:14 given it hands. But both systems still have the same annoying problem. I have to ask for something. Manus does it. I make a correction and a week later I start something new and apparently we're

11:27 having the same conversation again. So, the third system gives those corrections somewhere to go. Let's talk about a workspace that remembers. So this is a manis project that I use for research

11:40 and recommendations. It contains the files, the instructions, the standards that belong to this kind of work. And I'm going to give it a regular absolutely normal boring task. I'll

11:50 basically just say review the sources and give me a recommendation. Okay, so this is good, but there are three things that I would still change. First, there's a claim that cites an

12:02 article summarizing the original research. The primary source exists, so I want the primary source. Second, I've read all of this and only now found out what Manis recommends. If I ask for a

12:14 recommendation, I need the recommendation early and then show me the reasoning. Right? And then third, this sentence sounds more certain than the evidence is. And the sources kind of

12:25 disagree. So, I need manners to tell me they disagree. I don't want it to smooth that away because the paragraph sounds nicer. [music] It needs to be factually accurate. So, let me share those changes

12:36 with Mannis. Now, I'm done. And normally the AI fixes this task. Great. But then next week, we meet again and I have to say, primary source is important. Recommendation

12:49 needs to go first. Tell me when the evidence conflicts. I've probably spent a measurable percentage of my adult life telling AI to [music] stop burying the recommendation. Just saying now

12:59 apparently this is my cause but manis projects can review what happened inside the work and propose useful updates that carry into the project. [music] So I'm going to be asking it to review the

13:10 corrections that I made to the task and ask it what should this project carry forward into [music] future work so it can propose updates and not add anything before I explicitly tell it to. And look

13:22 at this. it wants to carry forward. Prefer primary sources when available. Lead recommendation tasks with the recommendation. Make uncertainty and conflicting evidence explicit. These are

13:34 not just edits to the current answer. Manis is proposing these changes to the project and all I have to do is approve what actually belongs there. So [music] yes, yes, and yes. So that is the carry

13:48 forward. It's a useful correction that now has somewhere to go. live and become permanent. And I want to be careful with the claim here. Okay, Manis has not absorbed my entire brain

14:00 because I corrected three paragraphs. Clearly, what happened is much more [music] pragmatic. I guess I made the correction. It identified something reusable. I approved it and the next

14:13 piece of work can start with those standards already built in. That's it. So basically when the same correction appears three times, the correction is probably [music] not the problem

14:23 anymore. The system has nowhere to remember it. That's usually what happens. Now before we test whether that actually survives into different work, there's one more move inside this

14:33 workspace that I use a lot and I want to let you in on. Let's say that I'm happy with this research. Okay, but now I need two different things from it. Let's say I want a one-page leadership brief and

14:45 separately I want somebody to challenge the recommendation as aggressively as humanly possible. I don't want those two directions contaminating each other, right? Because they would feed on each

14:56 other and that's not what I want. Okay? And I also don't want to start from zero. So in Manis, you can branch from this point. The new session can keep the files, instructions, and conversation up

15:08 to here and the original stays where it is. So, in branch one, I'm going to say turn this into a onepage leadership [music] brief. That's it. And from the same standing point, I'm going to go

15:19 also build branch two. And I'm going to say assume the recommendation is wrong. Build the strongest evidence-based case against it. [music] And now, essentially, we've taken one body of

15:30 work into two directions without rebuilding the standing point. Now, every time you start a fresh chat, you throw away all the small decisions that got you somewhere useful. Branch will

15:41 let you keep that starting point and go somewhere else from it. But now for the actual test. Did the carry forward change anything? Let's see. I have a different recommendation problem with

15:52 different sources, a different subject, and I'm not asking Mannis to repeat the job we just corrected. I just want to see whether those three standards are still remaining. They're still in place.

16:04 So, I'm going to say review these sources and give me a recommendation and we'll see if Manis knows how to apply the same principles. Okay, first check. Okay, I'm very pleased to see that the

16:15 recommendation is at the top. Good. Let's take a look at this citation. The claim goes to the primary source. Good, Manis. Now, third. And here, instead of pretending these sources agree, Manis

16:28 has explicitly flagged the conflict. All right. Well done, Manis. Honestly, that is super good given that I only corrected those [music] things once. But I'm surprised many times you have to do

16:39 it again and again. So, it really understands the pattern there. And that's okay. When you change the task, as long as the standards come with you, that's fine. This is the point where I

16:50 think our standard for AI needs to change. The best AI system isn't the one that gives you the best first answer. It's the one that makes answer number 50 better than answer number one. That's

17:00 why we talk about continuous and never- ending improvement. Every AI tool can impress you once, but what interests me after 6 months is whether the work that I've already done makes my Tuesday,

17:12 Wednesday, or Thursday easier than it was 6 months ago. I do want to mention one thing, however. Feature access, credit limits, rollout status can change. So, please check what is

17:22 available in your account before you build one of these. I know usually people comment saying that some of the features are not available to their account or in their country. So I'm

17:32 showing you what is available to me today. One last word of wisdom. Do not start with madness. Start with your week. Where would a disagreement check change a decision because you never have

17:43 enough time to investigate every angle? Or what repetitive job could become a Sunday run so you stop remembering to start it [music] yourself? and what correction deserves a carry forward

17:56 because you're tired of saying it twice. Those are the three systems. And if you want to build them, I've put together the setup for all three, including the instructions from this video. Like I

18:06 said, inside the founders community, you can join. You have the first seven days for free. You can [music] take it and leave if you don't want to stick around, but I think there is enough value in

18:15 there to stick around. I promise. But bottom line is I need you to pick one thing. Please don't go ahead and build seven complicated AI workflows tonight because they looked clever in a video.

18:27 The best automation starts with the job that was already happening and was annoying to you. And that's what really unfair means to me. [music] Everyone can open the same AI, but one person walks

18:37 into that Friday meeting having looked at five angles while somebody else looked at 25. One person gets reminded to prepare for Monday while somebody else already has the prep done. And of

18:48 course, one person repeats the same correction next week and somebody else carried it forward. None of those differences look dramatic on one day. [music] But if you give them a year or

18:59 two or 10, that gap will start to look like talent. So you if you want to see the same idea applied to Claude, I have several videos. This is a good one to watch next. In the meantime, thank you

19:10 so so much for watching. Like this video if you did. Be sure to subscribe if you haven't done so. Make sure to share it with anyone in your circle of friends or family or co-workers who probably needs

19:19 a little bit of help and maybe doesn't know about Manis and how amazing it is. And until next time, I hope to see you in our community, whether in the Trailblazers or in the [music] founders

19:27 community, depending on what is most helpful and serving to you. So, thank you again and I'll see you next time. Bye.