Introduction: Who’s in the Room and What’s Hard About “Cloud Code”

So, I guess the first kind of thing here, this is a little bit of a different take on taking how many people here are technical. Hands up. How many business people do we have here from? Okay.

Now, question, how many would classify yourselves as being somebody who works in the giant corporal trade? Okay.

How many are startup type people? okay cool and now out of all of you how many how many have you have used

cloud code today this week ever okay so that's not bad I'm just trying to get out feel the room what's the biggest when you use quad code what is the biggest sort of problem uh that you guys find writing rate limits five hours yeah it's okay so it is

A Quick Reality Check: Tools That Don’t Waste Time

that's that's always everybody's kind of so i kind of want to start off with a video to kind of explain what people look like so let me just pull this up the cloud code does not ask quick question

that take 45 minutes delivers the code TypeScript error -free.

The cloud code does not talk for 30 minutes in stand -up. The cloud code does not schedule meeting that could have been email.

The cloud code does not post on LinkedIn. The cloud code does not reply all to company -wide email.

And ultimately, he's telling you everything. But at the end of it is, it will do whatever, does not, you know, it will just do more and more random things that people will say

that cloud has saved my life. But ultimately, the best part is if most of these people you ask, what does it actually deliver and most the time we're talking about all these

features but how much work is actually getting done right it looks really cool and yeah most people I don't know they don't have time for this interview right

exactly but what I'm trying to get out is all of us have an approach that first of all we don't want to be too technical but at the same time I think we want to try to get work done using this AI because in the end of the day we're all sitting we all just want to get work done that's actually meaningful or get work done for us so whatever moves needle

A Different Approach: An “AI-First” System That Gets Work Done

so I'm going to present to you a little bit something of a different because at MindStone what they've actually done is they've addressed this problem and I've been using this platform to give you some ideas this is more to give you ideas of what is possible out there a lot of the

the stuff I was blessed enough because working and taking some of the courses particularly with Joshua the CEO he was building stuff that was actually completing work it was weird he'd do demos where it was like making phone calls to people post event to see how did you like the meetup was it good bad whatever so that kind of got me thinking but as I got to know them a little

bit more they've been working on on this really cool interface and I wanted to kind of present it a little bit differently because what I'm showing you here is the equivalent of if you had cloud code but you could use it it could remember everything about you across everything that you do execute tasks for you and help you run

your business so here for example I have different conversations that I've had I I tried to set up the speaker but you know I'm talking to it but what you're basically looking

Example: Building a Talk by Profiling the Audience

at is I did a quick sort of exercise here so the first thing that I did was to give you an example I create a presentation for this group what the first thing that I did was I took all the people that were at a meetup at the meetup here registered on the Mindstone site and I put them in to get some sort of an idea of what the audience would be like so when i go up here if it helps me go

up here because it's it's not unlike claude code he does not like to be distracted when he's working so anyway if it's not showing up we won't go to it but basically

what i had to do was look at what kind of people are in the room what companies they work for Were they technical kind of going what you guys were showing? To get a percentage of who registered to try to get an idea Do I make this more of a technical talk when I'm showing this or do I make this I'd get things done type talk So I decided to go through more of a get this stuff done type talk

So I got to prepare a presentation Basically just create something from based on the group based on the companies that they were at to give some sort of very sort of general conversation that wouldn't be too over the head of people or too low too low maybe that I already have heard about this I found nothing interesting so I created this 14 point 14 slide presentation I'm basically kind of going over what it is just going to the six different tools

and I'd also taken kind of like the examples like Albert was giving I had given it some previous Talks that were done to get an idea what some presentations might look like so I didn't use Presentations, but what I did was I took the transcripts from those put them in to kind of give it Here's what good looks like

It's on a folder on my Google Drive that combined with some of those are examples That it can reach and fetch for and it remembers in long term if it's doing a presentation For whatever good looks like even if it's a presentation for let's say anyone individually let's say i was doing one for you individually or for the room it would remember what would be

kind of good practice best practice when people seem to be reacting thinking about it think about it like intent or sort of like inferring what good looks like to get a better picture of it so some language here is some of the stuff that people may have reacted to not only for conversations of previous uh talks about rebel but what also the room might based on people's job titles what

Why Context and Long-Term Memory Matter

what might work right so one of the things you know was saying six different tools that forget what you did yesterday kind of what i was mentioning kind of the slogan everybody's kind of using a bunch of tools they've got chat gpt whatever cloud code open going through why this is better what the problem is if this is something that works with your workflow so let's say an

agent that can fix your workflows or think for you to solve them and persistent memory and tool connections across all of your machines that could work all this is just basically context capture

of course i kind of did the live demo this is more of the slides so don't there is no live demo i'm not doing the slides but i'm just trying to give you guys an idea of what it captured for what type of people it's for what are the type of technical teams that might use it that were kind of in the group um and then give you some sort of quick overview of what how rebel differs

from let's say other platforms like cloud code now all this can be done but I also of course it's here we've got all of it you know what it might be so the presentation was prepared for me because it understood based on examples other things context is very important now I'm gonna try to do one more thing here on a

Live Workflow Example: Persistent Chats That Execute Tasks

live example as well because Albert I'm gonna go see what your conversation was about so i've got an automation here let's see what essentially i'm doing here is because i'm able to capture all the conversations i'm able to have any conversations also in other like so

you have your conversations these aren't conversations like with a chatbot tasks actually get done in these different tap in these chats so some people might think about it like a project that you would do on clod but on steroids with no context tokens whatsoever this is just plugged

into my max i'm paying the 90 euros a month um and using this heavily probably about four to three to four hours a day uh just doing basic admin tasks in the background um i burn about my usage is about 10 a day so it's quite well engineered in the sense that it doesn't just

eat tokens because i know some of these things look great uh everybody's it's all fun and games until you get your cloud code api bill right everybody's had one of those that's then you wonder if you're going to have to mortgage your house because you wanted to really do that deep research on those five clients right um so it looks like it's done here but what i'm going to

go to now is the same conversation and what it should be able to pull now is on this after this presentation normally i talk to it but sorry i know everybody's doing talking right now i'm old school a little bit but i'd basically say something here to my whisper flow but in this case

so this here will look at the conversation that i had we'll actually do the research through all of it as you can see it actually working so a lot of this stuff uh just to give you a little bit

From One-Off Prompts to Reusable Workflows

of background on this this full system why this is important because what does it mean to be ai first i've been giving this question since october since i started since february even more since opus 4 4 .5 came out i started thinking okay this stuff can do stuff for me now but i still see people doing one one trick ponies they're still using ai uh in such a one use case at a time way

so you've got 25 workflows they're sort of like you use them and then you throw them away well why I mean wasn't work we promised a better future where we'll wake up have a coffee while something's working for us what you're seeing here

is a combination of here as you can see though it's giving a skill but it It completely hallucinated in this case. I should have waited till tomorrow, but That's not the point.

No But usually what would happen if this was done at 8 in the morning would have done the ingestion Normally when I wake up in the morning, I have a conversation and it will give me not only on my to do's What I have to answer

give me drafts of emails that I have to prepare for certain things because it's learned on previous things that I do as it picks up every day and and also give me context on things that I might have missed like hey look you had a call with this guy so you sent him an email did you send him an offer or did you prepare for this workshop

or whatever it might be at future lab for for whatever it might be 1so the real capture of AI first is really how do I take and what Albert was kind of covering how do I take all the know -how that I have and from the first moment working with whatever I do, how do I use AI? And this

The Goal: A Chief of Staff That Finds Information and Acts

system here is the best attempt that I've seen to do that because really what AI should be for you is a direct chief of staff. It should be something for you that you understand where it knows where the information is.

It's like, and not an assistant because it can execute stuff for it. it doesn't think, remember, don't forget that as well, but it should be your chief of staff that will be able to find whatever information that you need, hand you, let's say in the morning, a morning brief.

So you know what you have to do for the day. Help you plan your day. Like here, for example, if I miss something, it might have things like, hey, here are some suggestions what

Morning briefs, follow-ups, and drafted emails

what we can do for you um let's say business briefing uh post follow -up and commitments post meetings for example this is something that it did on its own i didn't think of that it did it for me so the really cool thing for me i found was i don't have to think about a lot of these things it will capture pull it out for me and say like look this is stuff that you have to do

and then it will also look like hey look you've got things that you might have to worry about you You might have to send an offer to somebody. There's hyper, it picks out things for me that I need done and prepares it for me.

Suggested automations and tool integrations (MCPs)

And on top of that, it also looks at automations and certain things. These things that you see here, it suggested these to me. It said, hey, I noticed that you're using this thing called blue dot transcription.

I don't have a connection. I don't know what the hell's going on. I'm flying blind.

We should create this for me. And it created it for me by itself because it's got all the right sort of agents with MCPs to connect all these tools to make all these things happen so I've been doing this for

about this is probably about the eighth day that I've been using this quite intensively but I've already noticed that my I'm saving about two hours a day and just admin work that I don't have to do because I just click if I know for example here at night I'll get a more I get a warning usually saying hey like here's what you've done today here's what you have tomorrow do you want me to

prepare drafts for this this this did you send this out hey by the way it might just be notes and some things will actually execute for you so if i wake up in the morning i will have drafts ready for me for whatever i have to do so this isn't one off this is a fully functioning ai first operating system for anyone who's working every day now you can talk to it there's parts of

it you can and you can hook up more things like for example uh you can hook up whatsapp to it you can use telegram of course this is a little bit of customization but what i found so far it will tell you how to do it so it will tell you how to build it because it seems to be able to search for the documentation so it makes it a lot quicker for yourself

and it also allows you to take all these things and make your life a lot productive so that's kind of it um and yeah do you does anyone have any questions or anything you've seen here yeah good

Costs, Usage, and Practical Limits of Today’s AI Tools

Good. I can show it to you. I'm on a Max plan because Claude just switched this over. So this is, let's go back here.

It's like they instantly want me to pay, just jump up with this subscription page. Basically, I use the Max plan. It's the 5x.

But I can show you the usage even for today if it decides to let me jump out because Claude is not but basically it's been all within 10 to 15 percent i use within five business days i and this is like running parallel agents that some of the jobs could be one job the longest job that i ran it for to my shock was three hours and i thought i always had a heart attack when i saw three hours but it was well within the max uh token usage because think about there's a couple

What’s broken: fragmented “surfaces” and weak shared memory

things that Claude doesn't do that this does one of the big problems with Claude Claude code if you have the three surfaces so you have desktop right you have chat you have co -work you have uh code at least in the moment today Claude does not have carried memory across all three surfaces so if you fire up co -work de facto each session that you fire up unless it's a project and

And projects have limited space for memory that they have, which also, very important, you're not sure if that is being used to train Claude. They say that, oh, yeah, you mark off something. I'm sure that's how it works, right?

What changes: local-first control and stronger context capture

This is fully on my desktop. Claude, it's a layer one. All this stuff is my proprietary information. Any markdown skills, anything that's created, it's on my desktop. Or in whatever place that I feel comfortable with. Claude never has access to it.

and the second part is the problem is these the three surfaces that cloud has do not work well together this like fully transparently this just sits on top of cloud code it's not like it's a completely new platform what it is is just an optimized way to solve the problem of how do i capture context so the system gets to a know me better and can talk to me get to understand me better do tasks for me and not make it super complicated in the process by spinning up virtual machines which is essentially creating a new instance for each new task that you say in co -work

which means you can only do potentially one task at a time and the agents might not be aware that one might be executing this exact same task and with cloud code having the same potential of having that memory carry over so that's that's a like at least for me i know in practice it doesn't

think about it because like what i was saying most people are building one -offs so these things like what you're using for example but imagine if that was remembered and then claude code came to you and was like hey look that seems like a great project have you thought about doing this for maybe uh google ads i've thought about this based on your conversations right that's that's what i think is the big game changer for a lot of people and unlock and i haven't seen anything

else come close to this i tried to build this system myself i gave up because the people at mindstone already had it figured out so does anybody have any questions any thoughts is yeah

Not One Model: Using the Best Tool for Each Job

sure i have i have um gemini i use for certain things um chat gpt i don't know where i stand with chat gpt right now it's a love hate sometimes i'll use it but i said but i find myself being kind of like it's just not execute i don't think it's as well designed as claude um and then on the

end with obviously perplexity like this is another thing like perplexity is plugged into this so claude's not doing the research if there's voice models let's say chat gpt has a better voice recognition than claude does so i'll just use that research is much better done by perplexity because it has way better capabilities to do deep search so i'll use that for research so it's not

just tied to one model but it's basically think about it as a operating system that you can log into get things done learns from you and help you can execute so it's more to show you guys how to inspire you guys to think about where this is going and try to really think for yourselves

themselves how do i take the ai that i have and stop using it for one -off tasks so you know the kind of n8n stuff that you see zapier those type of things the next level kind of cloud code executing some one thing um thinking in terms of an ecosystem or an operating system for what you

Conclusion: Think in Systems, Not Single Prompts

do every day as a starting point and that's it i think for my end if anyone's got any questions happy to take them if not now we can always take it while we talk upstairs or or sticks around yeah all right thanks a lot guys