Hey, welcome back to Building

Stacklists. If you're new here, this is

a show about what it actually looks like

to build a startup in real time. Not the

polished version, the actual version.

Today, I have both of our co-founders

with me, Kyle and Martina, and we're

going to pull back the curtain on

something we've been heads-down on. It's

called ACP, the Atomic Content Protocol.

We built it while building Stacklists.

We're open-sourcing it, and this is the

first time we're going to talk about it

publicly. So, that's what we're going to

get into today. Kyle and Martina, let's

get into it.

>> No, I I see you're thinking bubbles so

on top of that.

>> bubbles?

>> You can just go for it.

>> I think his

I'm going to come at it from the user

experience perspective. Think about it.

We now, as of

you know, mid last year, I mean, it was

2 years ago that ChatGPT came out and we

all started sort of using it, but I I

really think like this year is one of

those years that more people adopting AI

into their daily lives. And one of the

things that I find is that as I'm

working with with

with Claude, let's say, is I work on a

hard problem, and then I go, "Okay, can

you just compress that into a final

output that I can take and go work on it

elsewhere or send it to somebody else?"

So, it writes this markdown file.

And I say, "Great." And then I download

it, and it goes into my downloads

folder, which is just a dumpster fire.

You know what I mean? And then all of a

sudden, like at the end of the day,

you've got So, I've got eight or nine

markdown files, which are like the

culmination of a bunch of work I've

done. And then you go, "Where do I put

it?" Do I just do a random desktop

folder that says stuff I'm working on,

you know? And then you've also got like

Notion or Obsidian, and a lot of those,

the same with bookmarks,

you know, was typically a like, "Here's

a place where you can sort of dump

stuff." But it takes a lot to sort of

like organize it and keep it up to date

and all those sort of things. And so,

you know, putting it in a Stacklist felt

natural as the as the sort of one place

that you can put anything and everything

about your life. Um but there's also no

standardization across anything. So how

Notion does it, how Obsidian does it,

how

um

you know, any of the tools that you use

when you're calling, you know, data back

and forth between all of these different

systems, they're all kind of speaking a

different language. So that was that was

kind of the goal is to what happens if

you standardize it.

>> Yeah. So I guess the problem isn't

really that like we're not saving

things, is that we're not saving them in

a way where anything else can find those

things.

>> I don't think anybody I don't think it's

the same problem that we had that we're

trying to solve with Stacklist, which

is, you know, I think people just get

used to the fact that like their stuff

lives in a million different places. And

they're just used to going, "Oh yeah,

hold on. Let me get that for you in my

thousand tabs on Chrome." And I'll just

like dig into and you don't realize in

that moment that you can't just sort of

search something at kind of a macro

level and recall it. You have to

manually go click and drag and find. And

it's it and

all of that sort of like organization

and recall actually happens in your

head. And I think there's so much mental

overhead from from doing that that like,

you know, it should be easy enough that

you should just be able to open a sort

of global search bar and be like, "Find

that thing I was working on." And it

finds it. And so that's, you know,

that's the beginnings of what we're

trying to solve there.

>> I was just going to say that that would

expand this a bit more from developers

perspective is that when you think about

it, and I think for me it unlocked when

we were thinking about, "Okay, how does

Stacklist content appears in any type of

elements?" And that's the first time

that I started to think about it on a

different level. Then I started to use

more of a cloud cost status bar where I

actually tracked, "Okay, how many tokens

in and out, you know, usage and

everything." And that got me thinking,

we are so used to, you know, just

spending all of those tokens and usages.

And yeah, just charge me more.

Charge

Charge whatever. Like I don't care. I

don't pay Stacklist pay. Like go

overboard, you know.

>> I pay for Claude on 20X on extra high,

extra Mac.

>> Extra Mac.

>> And and then it got me like that could

really be dropped if there is some kind

of a uniform language. And I actually

loved that one point that I think Kyle

made the made the black mark about it

that really stuck with me was imagine if

we were on this podcast and speaking

different languages. Like we could

understand each other, you know, through

I don't know, various hands gestures and

so on, but we we would spend a lot of

energy. And the same stuff is is with

agents. If you speak with different

languages, they spend a lot of energy

tokens in that case. So that's when my

bill started turning, basically.

>> I think this makes me want to get into

sort of

>> Yeah.

>> like the reasoning of

>> Sorry, one one little piece on there is

I will say like like think about I think

we don't realize we get so excited about

something like like I'm working with

Claude and it just figured out how we

should talk about Stacklist at a macro

level. This is so great. And it says

download MD and I do it, but I don't

stop in that moment to go like am I

actually then putting this anywhere

where anybody can get to it? And even if

I do sort of like hook up to Notion or

to Obsidian or how and where I work and

my team works, like that is the the

goal. And one of the things we did that

Martina built an awesome MCP basically

connector into Stacklist that Claude can

read. I can just say now go put that in

my Stacklist and it just does it. And so

now it is in a place that is

re-findable, searchable, shareable, you

know, and so I'm not again putting it

into my downloads dumpster fire. I'm

putting it into a place that now I can

go find that. Like Claude can find it.

Chat GPT can find it. Stacky, the agent

inside Stackless, can find it. So, we're

we're putting it into a place that is,

you know, um can be searched and

organized really easily now.

>> Yeah. I mean, and I think what I sort of

want to get into now is like So, we've

been talking about this problem about

like knowledge and where to put it and

how to find it and have it in one

centralized place and all these markdown

files and sort of like Stackless is the

answer to that, right? But, ACP is is

not an app. It's it's a protocol. It's

what Stackless is built on or of or

with. So, what what why when when you

So, you guys sort of solved it in the

Stackless way. Here's this app, but you

thought to yourselves, "Okay, but that's

not enough. We need a standardization."

And I want to get into like that of like

why ACP and what is our point with it?

What do we want people to do?

>> Yeah. I think from a from a like from a

from a lay person's perspective, I would

say think about So, SMTP is what we use

to send email back and forth. It's a

simple mail transfer protocol. If mail

and Gmail and Hotmail and Yahoo had all

had their own different languages and

ways of speaking, it would be so

inefficient to try and send mail to each

other. But, having a simple protocol for

email to be able to go back and forth

made it that everyone can send email to

everyone else. And so, like that that I

think, you know, what how I would

explain it to my aunt would be that sort

of version, which is like you have to

have some sort of way to talk between

these systems. And if you don't, then

you're just wasting time having each

system try and learn each other's

language for every simple like call.

>> Okay, so a bit more maybe technical from

my side, but

think about it that internet was

designed for humans. And the fastest

growing users of the internet today are

AI agents.

And AI agents read the same pages that

we do over and over again. But, they're

wasting a lot of tokens and they're

asking what is this content about?

That's the end goal. Like what is this

content? They sometimes need to read a

lot of data to figure out if that's

actually the answer to the user

question, if that's something that you

want to serve as answer and so on.

Imagine Wikipedia for content, but on

the agent level. So, that you enrich

once, you store it and every other time

agent re-reads it it's like 90% lower

token usage because it exactly knows

what the content is about. That's I

don't think we will

I don't think we will solve internet,

you know, I don't think we will we will

do I don't know what in my head

>> internet. I think we solved the

internet. That's that's shooting for the

moon.

>> But that itself is pretty, you know,

when you see it actually the output when

you enrich something and when you see

the the content being reread second time

and it actually consumes

200 token instead of 20k tokens, then

it's like okay, this makes sense.

>> So, you ran a test recently, right? And

I think it was like 84% token savings

with the following the ACP protocol. Is

there any way we can do this live? Like

can we see it?

>> Oh my god.

I didn't prepare for that, but yes.

Let's me Kyle

>> This is the loading state is loading

demo. Which usually demos like 99% of

the time won't work when you try and do

it live, but we'll let's let's see how

it works.

>> edit it out.

>> You know, Kyle just talk for 10 10

minutes until I figure something out.

>> So, when I was born

uh

>> Maybe can we show while Martin is

looking for that, can we show people

what like the the atomic thing like the

all the components?

>> Yeah, I was also going to say just a

good analogy like the difference between

basically like if you if you walked into

a library and you had never been there

before and you had to sort of wander

through all the sections to figure out

where everything was. If you left, when

you came back the next time, like, you

know where the history section is.

You're going to sort of cut right to it.

Every time, you know, if there's no

instruction manual, it's kind of like

going into it with amnesia every single

time and just being like, "Let me go

read. Let me go scan every single

section." versus having this like this

map right to where you want to go.

>> What's actually fascinating is that with

Claude, now you can build pages like

this in in a no time. So,

there is actually know-how about a

protocol and how it works and the

problem and everything and how to So,

connect to Claude. I actually, what I

build as well is the I would say

point zero point point zero first alpha

version of MCP that anybody can use and

is this URL and basically, if you put it

into Claude or Claude code or wherever,

you can start using it and see how

enrichment works and and do it for

yourself. There's also playground where

people can go and try with their own

Anthropic API key. It's not stored

anywhere. It's just locally to the

session and try to see how the enriched

data comes back.

>> [snorts]

>> But, what you mentioned about token

usage, so I'll share my chat uh

uh

when you load it, you can actually ask

what tools are in there and it will say

which which tools and what each one

does.

And then, what I basically asked was

like, "Take my stack list stack and

enrich my nine cards inside." So, it

took nine cards and enriched it and it

said per card breakdown uh how many raw

tokens was used first time and how every

sequential time when it re-reads the

same card. This is the benchmark. It's

not 100% correct, this 84% because you

see it's the same on every row because

uh it truncates the body.

Maybe some maybe this one is higher,

this one is lower, but you get the gist

like how many tokens it actually can be

reduced if you enrich it.

>> So, essentially if all of if if

everybody started formatting their

content with ACP, then everybody would

be saving a bunch of money with AI.

>> [laughter]

>> Is that

>> Essentially, I mean okay, message is

definitely something that would drop.

Imagine how much faster you would get

reply if there is like a open database

with content that AI can just go through

and say hey, confidence of this content

is

95% and here it is in a second.

>> Very cool. Having that little it's like

having you know, lots of little treasure

maps saying exactly what something is

and if you just need to read what is

this about? This is about books, this is

about like favorites, this is about

restaurants, whatever. If you can just

read little bits of the headers of

things, you don't have to read the full

contents. Like you don't have to read

the full magazine article. You can get

the TLDR on everything, which then means

it's not only less expensive, but it's

also faster to just sort of really

quickly scan the headers of everything,

go okay, I found what I needed. And if

you standardize that also, then you have

every system that comes in knows exactly

what headers to scan and what they're

going to get, which is just going to be

it's going to be faster, quicker

lookups, you know, and and less tokens.

>> And so, Stacklist is basically like the

first app built on ACP, right? So, I'm

wondering like what does that actually

look like? Like for a real estate agent

or like honestly like just for me trying

to organize my stuff. Like can we can we

>> I think at a macro level like imagine if

you basically you know, the goal is I

want to search for restaurants or I want

to search for clothes or I want to

search for new AI apps to use between

you and me and Martina and our team or

whatever. As I start to search that on

Stacklist, now I'm not the the system

doesn't have to go into everybody's

profile and try and read everything. We

can really hit only the relevant parts,

which then sort of pulls it back to me

better and that sort of enriches the

whole platform. So, really what we're

trying to do is make this quicker,

better, faster, bigger, faster,

stronger, you know, and so imagine that

at a network scale. I think that's the

sort of thing that like, you know, it's

almost also prepping us for being

Facebook, Instagram size and thinking

about that early versus trying to

act through that now and then having to

to go think about it later. It's just

doing it doing it early is helps us with

scale.

>> Think about so, we all know how

basically everybody heard about AI

indexing, how AI crawls through our

page, how you need to have JSON-LD, how

you need to have schemas and so on.

That's all built for crawlers, for cloud

crawlers, for ChatGPT crawlers, but it's

not built for agents. I mean, we are not

replacing anything. We are just adding

one layer of enrichment of data that is

specific for the agents. So, imagine

having also the JSON-LD schema and

everything around crawlability, and then

plus the portion of the agent that it

can be easily found and imagine search,

how search will work. Search will work

like it will be blown out of proportions

when you search for real estate agents

that have, I don't know, local expertise

in Atlanta.

That's that's how we think about it.

>> Think about how efficient inefficient it

is right now when I connect up services

like I have, you know, this this

personal operating system agent Virgil

that I built just for me just because

I end up drowning in when I sit down in

the morning. I've got my coffee and I'm

like, okay, really quick, let me just go

check Todoist and Slack and email and

calendar, my granola notes from

yesterday, my Light Field CRM for

opportunities. Like, let me go scan all

this stuff. So, this agent essentially

connects to all that stuff, but that

agent has to go every single thing is

all built in different ways. And so,

it's trying to figure out piecing that

stuff together. If you imagine that if

it if when it connected up to notion, if

notion was built in such a way that it

could just hit the headers of everything

and sort of had a map of what to do, it

would be faster and less tokens for me,

less tokens for notion. Um and you have

that standardization that all would

happen, you know, more efficiently and

and inexpensively versus,

you know, that agent trying to basically

learn 12 different languages when it's

coming to

to do these these calls for me. Martina,

have that EMW those two EMW slides.

If you want to show those, just the

high-level sort of the card versus what

AI sees.

>> While Martina's looking, Kyle, do you

know that uh I once offered to send my

Martina a pizza on Twitter before we

met?

>> [laughter]

>> Did you send it?

>> She She She declined.

>> You should send her a pizza with

pineapple on it.

>> She was like, "I can get my own pizza."

>> What? [snorts]

That doesn't sound like me. I said you

will buy me a pizza in San Francisco. I

hear it is so cheap there, so we can do

it there.

Uh okay, here it is. Uh this is the card

that is a single [snorts] piece of

content and it has like the ID when it's

created, if it's parent or something, uh

is it can agent read it? Uh is it

public? How many tokens consumption? And

the confidence layer.

Um that's like the basic structure of

enrichment, which I added more data to

it to it to be honest.

Uh but unstructured would be this is the

idea of the card, I don't know, notes

and whatever is in there, but

having the enrichment and this is where

also Stackless MCP comes really handy

because if you use our MCP or stack here

whatever we have all all of those

enrichments built in. So whenever you

say save something from either close

session or URL through stack here

whatever it automatically gets enriched

and automatically picks up all of the

data. So our structure of stacks and

cards is of course normal ID notes and

author and everything around it but also

JSON-LD baked in uh this ACP baked in.

So we are really

uh what we like to call it LLM

delicious. So this is the structured

card.

>> Mhm, that's good Stackless.

>> And this is the structured stack. So

stack is like the container over card

and it has the same the same enrichment

but it also consumes like okay how many

children are inside uh what is the

structure and what is the relationship

what's the confidence of this actually

answer somebody's question. This is the

structured data.

>> So basically a Stackless card is like

basically already formatted the way AI

systems want to cite things without you

or me or anybody having to do anything

extra.

>> Yes, AI and uh agents.

>> AI and agents of course. This is so

cool.

>> Yeah. So this is like you know I think I

think what I love about this is you know

one thinking ahead and really getting to

a place of identifying we ran a big

research on this and it really seemed

like a a big white space that you know

people are still so enthralled by

building we built an MCP which is like

step one but really thinking at a macro

level when you get to a place where you

have these agents that are sort of you

know helping you run your life but

they're interacting with am I is my

agent going to interact with all 100

logins that I use throughout the week?

You know what I mean? And they're all

talking you know different languages

that's where you know starting to

hopefully lead the way and and think

about this early is going to you know

hopefully will will help not only for

stacklist adoption, but hopefully this

solves a problem that that, you know,

other people are going to be thinking

about and trying to solve here soon.

>> And if ACP like really succeeds, like

what does the world look like?

>> Same.

>> This

>> But with hats and t-shirts and like it

says I only ACP and like there's all

sorts of branding and No, I mean really

open source it's giving back, right? To

the It's It's really giving back to the

community and and I would imagine

Martina that you you know, at some point

you would have

contributors and managers on the open

source project and and people are taking

it even further than than we've thought

about it.

>> Yeah, I mean that's the whole goal with

open source, but I imagine it if people

actually start to use it, I mean I would

be over the moon because that's like,

you know, for me for me personally that

would be a huge success. But

it would be like Wikipedia, you know,

you don't think about it, you just use

it. You have the open source can crawl

and and and read and I think essentially

it would be much faster and much

streamlined process, especially for

developers to be honest. But yeah, I I

think nothing drastically dramatically

in the world, but what is the saying?

Small moment for the world, but big for

me. So, think about it like that.

>> And where can people actually go and and

start trying this? How do they do that

today?

>> Atomic Content Protocol

>> Atomic Content Protocol.org. We'll put

that large in the video.

>> We'll put We should make a cool make a

quick stack on this and maybe let link

to the app and then also maybe link to

the NPM package to install it and then

it'd be cool just to We could also just

do a quick markdown file in that package

to put in as a stacklist markdown card

with either how to get started or or

maybe there's a quick like scaffold for

Clojure or something like that that we

could we could throw in there as well.

>> Cool. And then I just have maybe like a

last question, last thoughts, one thing

from each of you that you want people to

walk away knowing. Whoever wants to take

it away, go first.

>> Don't offer Martina pizzas because she

won't take them. Take them. Yeah, I just

want people to really know that, you

know, I think it's important for for

everyone to to take away.

>> Perfect. I think

>> I think for me it's like I just love

this moment that we're in and I think

it's important you know, it's

I think it's important for for everyone

whether or not you're dug into any of

this, you understand any of it, or

you're in it, or or you're not. Like

everyone's going to kind of need to be

in that place where you're going to be

interacting with these tools, you know,

at some point. Like this is going to be

as using Claude or Claude code in this

kind of way is going to be like learning

office at some point. You know what I

mean? Like you can't be like I don't

know Word and I don't know Excel, you

know? And so it doesn't mean that you

have to implement this stuff right now.

It doesn't mean you have to, you know,

you have to be a master at it, but

be around it and explore and be curious

because this stuff's changing so much

and this is how the world's going to run

soon. So, um,

you know, and also reach out to Martina

and I if you've got questions and we

love collaborating and helping and

pointing in the right direction.

>> Yeah, I would also say reach out if you

think this is all And if you

if you

>> Don't just keep your opinions [laughter]

to yourself. We don't want to hear it.

>> [gasps]

>> If it can be actually better in some

other way, that's that's also legit.

Like I would like to hear all all

opinions. All opinions matter. So,

that's why I open source it because you

can take it and do, you know, explore

and see see how it behaves and maybe

somebody somewhere has like like a

billion-dollar idea that's that will

unlock him. So, amazing. I don't know. I

built this and I'm really excited and I

can see the potential and I can really

see how it benefits everyone basically

and I'm really hyped about it, but at

the same time I need to ground myself in

a way like it's out there. If people use

it, fine. If not, also fine. We'll use

it and stake lists will be built on

this. So, there will be signs that it

actually can improve all the the agentic

stuff and reaching and reading and and

so on. So, that's enough for me.

>> The open source is great. Just please

don't rename it atomic content pizza.

It's just not what we're we're not like

we don't want to get that far into it.

Like try try and keep the the spirit of

the protocol, please.

>> Thank you so much, guys. This was really

fun. I am going to stop recording now.

>> All right. We're out.