The core theme is that the effectiveness of AI agents is directly proportional to the quality and structure of the context provided to them, and building a robust "context infrastructure" is crucial for leveraging AI for personal and business productivity.
Mind Map
انقر للتوسيع
انقر لاستعراض خريطة الذهن التفاعلية الكاملة
AI agents on cloud code, co-work, and
codecs can now automate workflows, write
code, and run entire business processes
autonomously. But no matter how good
they get, they'll only get as good as
the context you provide them. And the
people and the businesses that will get
the most out of these tools are the ones
with the best context infrastructures.
So in this video, I'll explain all seven
levels of using context in cloud, from
chat to projects to a second brain to a
full business OS. Show you what changes
at each level, why it matters, and how
to transition. and I'll show you how I
set up the final level, a full business
agentic OS with shared context and
skills across my team with permission
settings. Now, before explaining the
levels, let me quickly explain why
understanding this context
infrastructure is key. As said, we're
all seeing these AI agents becoming more
capable by the day and with fast
developments in MCPs, connectors,
browser use, they're starting to become
really good at using our softwares and
the internet. and with skills, plugins,
sub agents, uh, schedule tasks, and
other developments, they're getting the
capabilities to actually start executing
work for us more and more autonomously.
But in order for all of this to actually
be useful for us or our business, agents
need lots of context um, to actually
know how to do work for us. And that's
why context is the fundamental layer to
get right in the upcoming months. I can
tell you from my experience that having
this context layer set up well has made
a huge impact on how I and my business
work together with AI and is allowing us
to make co-work and cloud code become
more and more of our primary operating
system. But it is a topic that is not
straightforward and requires some effort
uh and takes some time to develop. But
the earlier you start with this the
better because the context compounds. So
if you start the day your agents will be
far more useful for you and your
business in a couple of weeks or months.
Now, I'll explain all the levels of
using context from simple chat context
all the way to an entire shared context
infrastructure across a business and
show you exactly how to level up no
matter what level you are uh at right
now. You can also jump to the level
that's most relevant to you. But even if
you're a bit more advanced, I think this
breakdown will hopefully help you wrap
your head around it a bit better. Now
level one is where unfortunately most
people in the world are still stuck
which is by providing context to a
language model in each chat manually or
even worse not providing models with any
context at all. Again no matter how good
these models get if you don't give AI
context using AI usually becomes a
frustrating experience. If I give a
prompt like write me a LinkedIn post on
why setting up a context infrastructure
is the key to making AI agents useful. I
get a very generic piece of content with
m dashes that screams AI. Now, of
course, most people know this and have
figured out that you need to provide
some context to actually get good
outputs. But what context do you give
it? This of course depends on the
specific task. But an easy way you can
think about what context you need to
provide to get better outputs on most
types of tasks are first giving context
around who you are and what your
business does. Second, context around
who this is for like your ICP, your
customer or the recipient. Third, which
is probably most important, which is
showing AI what good looks like for this
task. for example, through examples,
references or descriptions. And lastly,
defining what the roles and guardrails
are for this task. So, if I give that
same prompt, but with that context, we
instantly get a far better response. No
am dashes, formatted like a LinkedIn
post, more my tone of voice, and more
relevant to my ICP and an aligned call
to action. But of course, the limitation
here is that it's extremely inefficient,
timeconuming to copy and paste or
rewrite context in each new chat. Which
brings us to level two, which is cloud
chat projects. Again, most people have
figured this out, but projects in cloud
chat were of course developed to solve
the copying and pasting and rewriting of
context problem. In a project in cloud
chat, we can just add the context files
once together with a system prompt here.
We can also add a broader scope of
context. And each time you want to write
a new LinkedIn post or do a repetitive
task, we can just give a simple prompt
here while still getting good output.
But the downsides of these projects are
that first they live in isolated chat
windows. So we have to hop between
different projects for all our different
tasks. Second, clot can't actually
create, update or edit any of these
context files or the instructions
itself. So anytime you want to adjust
the process, update a context doc or
anything else. We need to manually
update it inside of the cloud project.
Thirdly, these chat projects are usually
limited to a specific task and work best
when I have separated projects for each
task. I'd need to set up a separate
project for YouTube packaging, YouTube
intro writing and ideation because I'll
follow a different process and have some
different context files. And lastly,
it's hard to actually test and improve
these projects without you extensively
using it and updating it manually. Which
brings us to the next level, skills,
which instantly resolve all of these
limitations and why I highly encourage
you to start using co-work or cloud code
if you're still in level two and working
in a cloud chat. Because for skills, we
need to make this transition. skills are
very comparable to these projects. We
have the skill MD this and this is
essentially the instruction or the
system prompt just like in projects
laying out the process and when to use
the different context files. For
example, here I have my LinkedIn writer
skill that includes a skill MD that lays
out the process it should follow and
when to read the context files and
inside the skill we have a references
folder with all the different context
files. But skills in contrast to
projects can be used in any chat at any
moment you want. This means for example
that mid conversation here in a YouTube
ideation chat I had I can turn a good
insight into a LinkedIn post instantly
by just telling claude write a LinkedIn
post based on the topic we discussed
using the LinkedIn writer skill and he
went ahead followed the SOP and wrote me
a LinkedIn post. Skills are also far
easier to build than projects. With
Enthropics built-in skill creator skill,
we can build them by simply telling
Claude to build us a skill. For example,
help me build an infographic skill. We
can also build them out of any
conversation you had with Claude by just
clicking here and select turn into skill
or by just telling Claude in the chat,
make me a skill based on the process we
followed in this chat. We can also
easily share these skills with our team
through zip files by just asking clot,
can you create a zip file out of the
LinkedIn writer skill which anyone in
your team can upload by going to
customize skills and clicking on the
plus icon here and create skills upload
skill. Or if you're on a team plan in
co-work specifically, you can just add
them to your organizational skills. We
can also easily adapt Enthropics
built-in skills, for example, these by
just telling CLA you want to customize
them or clicking here on edit. And we
can easily import skills from other
people and businesses by just going here
to browse plugins and then going to
entropics and partner. There are also
dozens of skill marketplaces around the
internet. And with built-in evals, we
can immediately test our skills and
improve them fast to make sure they
actually work and give us good outputs
consistently. For example, here I
created a newsletter writer skill and
then just told Claude, "Please test this
skill." The criteria for the test are,
"Is this skill functional? Is the word
count similar to my newsletter examples?
Is the sentence structure similar? and
is my tone of voice similar? It then ran
pre-ests in parallel and gives me an
eval report with a summary of the
results and suggested fixes which you
can apply immediately. You can even
autonomously let them improve themselves
through an auto research loop which I
recently did a video on which I'll make
sure to link in the description below
too if you haven't seen it yet. So, if
you haven't yet, you really want to
start building out these skills around
your repetitive tasks and processes.
We've been building out more than 60
skills across all our business
processes, which if you're interested,
you can also download and customize for
yourself if you check out my AI
accelerator in the link in the
description. And lastly, skills can also
be scheduled, which means we can now
trigger them autonomously through
Claude, which I'll show you some
examples of later in this video. Now,
skills are amazing, but skills are best
for pre-established processes of work,
and much of our day-to-day work isn't
actually a pre-established workflow. So,
this brings me to the next level, which
is using file access together with
skills. Because for many, if not most
tasks that AI can help us with, it
doesn't actually follow a
pre-established process or workflow. We
have one-off tasks. We have tasks like
ideiation, planning, strategy, or using
AI for decision-m. And for many of
these, we don't necessarily need or want
skills. But we do want AI to have more
context around you, your business, and
your goals. And this is where file
access in cloud code or cloud co-work
becomes powerful. Because with every new
chat we open in cloth coowork or cloth
code, I can now give cloth access to a
folder on my computer. For example, here
I selected a file with relevant YouTube
documents. And in that folder, I have
some documents about my YouTube
strategy, some old transcripts, a
hookbank, etc. And if I now just want to
ideulate or plan a new video together
with Claude, I'll get instantly better
outputs because it has context on my
strategy, me, my channel, and what's
important to me. For example, here I
wanted to ideulate on the video I'm
recording. And you can see it pulled
some data and context like my brand, my
ICP, my YouTube voice, and my YouTube
strategy in order to give me more
relevant ideas to plan out my video. And
because it has more context, it even
pushes back on some things. Because, for
example, now it knows most of my
audience is non-technical and I explain
something that's too technical. When we
start working with file access, we also
start working with the cloud MD, which
is basically an instruction on how to
navigate the folder, which becomes more
relevant if the context grows. But I'll
cover the cloud MD in more detail later
in this video. And if you haven't yet,
you really want to start using this file
access consistently because you'll be
surprised how much more relevant your
answers get. And it allows AI to become
much more of a strategic sparring
partner. And secondly, because it now
has access to a file on your computer,
it can't just read those files. It can
also instantly update the files. You can
save new files or assets like
presentations, Excel sheets, Google Docs
directly into the folder. So any uh
update you want to make in a context
docu, good outputs you want to save or
assets you want to save, cloud can
instantly do it. And essentially what
this means is the more you start using
file access, the more this folder will
grow with context naturally. Now, and if
you're just starting at this level, I'd
highly recommend putting in some effort
and setting up some of these important
context documents. I've added a free
resource also in the link in the
description below that's basically a
questionnaire where you can go through
and I highly recommend taking 30 minutes
with a tool like Whisper Flow where you
can talk to your computer and you just
do a brain dump and by answering all of
these questions you can then feed that
brain dump into Claude and he'll create
these structured context documents that
are important to have as an initial
start. I've also added in some example
reference files so you get an idea of
what these look like. in my AI
accelerator. We also have a full
step-by-step walkthrough on how to set
this up efficiently together with best
practices and uh unlimited one-on-one
life tech help. So, if you want some
help, you can also check out um the link
in the description below. Now, when
you're starting to use this more and
more and the context in your folder is
growing, it's natural to go into the
next level, which is using cloud co-work
projects. This can also be done through
cloud code. The same principle applies
if you use cloud code. and projects on
cloud co-work is essentially just a
better way to organize your context
across different areas of work. Now this
is different than a chat projects
because chat projects are very task
based. Co-work projects can be used on a
higher level for areas of work. For
example, I have projects here set up for
sales analytics, operations, agency
clients, community management and
YouTube. And projects are essentially
the same as file access but in this case
we just predefined the file here with
the relevant context for this area of
work. So now when I want to idate on a
new YouTube video, I can just directly
go into that YouTube project and the f
folder will already be selected. We'll
also have all our chats around this area
of work organized here below. We still
use skills for the repetitive task of
course. For example, in this chat when I
was planning the video and I got the
concept clearer, I used the YouTube
intro writer skill to give me some intro
ideas and variations according to my
framework. We can also see our scheduled
task that are relevant for this project.
For example, my YouTube ideation skill
runs every morning to give me new ideas.
But besides this better organization,
there's one more added feature to these
projects, which are instructions and
memory on the project level. And these
allow us to add in specific rules and
guardrails and specific memory for
specific areas of work. For example, my
YouTube project, I have a specific
memory that it needs to push back during
ideiation because I want to have
alternative framing and factchecking.
And you can make these memories or rules
by just telling clot in a chat that it
has to memorize this. Now when you're at
this level and you really start to use
projects skills, schedule tasks more and
more and consistently and really start
connecting it more and more with your
softwares, you'll start using AI more
and more as your operating system. And
honestly, if you use this infrastructure
well, you can already get a lot out of
AI for yourself and your business. But
when you start to use this more and more
and your context and your projects are
growing, you'll notice that even with
this project infrastructure, the growing
context will become harder to manage.
You'll have shared context files across
multiple projects, across multiple
skills, for example, common docs like an
ICP doc. And when something needs to be
updated, it needs to be updated across
all of these different projects and
folders and skills separately. So that's
where we want to start looking at the
next level, which is setting up a second
brain or a personal operating system.
Now, even when you're planning to roll
this out across a business on a
companywide level, which will be level
seven, I still highly recommend you
start with level six. Once you've set it
up and it works for yourself, then think
about level seven, where you actually
start syncing this across your team with
permission settings, etc. Now, in this
second brain setup, all we do is we just
add all of the context and centralize it
into one folder. And this becomes very
powerful when we have a lot of context
because we'll now have persistent
up-to-date context around an entire
business or life across any chat or AI
provider. We can do that by opening that
file through cloud code or just doing it
with file select in co-work or by
setting up one project connected to the
personal OS folder. But I can also do
this in codeex or any other AI provider
that allows for file access. And as I
said, we're still doing the same. We're
just adding all of that context into one
big folder. And this becomes an
advantage when you start using AI and
context around more and more areas of
work in your business or around more
departments because when you have a lot
of contacts, it's better because we now
just have one folder to structure and
organize. And this also means that
context docs don't have to be updated
across multiple projects or skills. I
can have all my projects or business
departments inside of the same folder.
And this is also the level where we can
start to add real-time context by
automatically adding your meeting
transcripts uh your daily task updates
or analytics through scheduled tasks.
This schedule task for example
automatically updates my second brain
with all of the uh meeting transcripts
across my team every day by using
Firefly connector. I also have a
schedule task here for team task roll up
which checks every day what my team has
been working on and updates that to the
second brain. You can also do this for
analytics. And then I can also do things
like a morning brief where it pulls
context from my second brain, knows my
priorities, knows our to-do list across
the business and gives me an overview of
what's important today. With this setup,
it also allows us to build better skills
and build them faster because through
this setup, we already have in-depth
context around our business, which we
can refer the skill to. So all we need
to do is lay out SOPs or workflows and
link them to which files in the second
brain it needs to read to get more
context. So in this setup, I highly
recommend starting to build your skills
a little bit differently. So instead of
adding context docks into the reference
files inside of the skill, you actually
want to make the skill reference where
it can find the reference files in your
second brain. For example, as you can
see, I did in this one, it only has a
skill MD with references to where it can
find the different reference files to do
its job better. And this means when I
make an update on my ICP document, all
my skills that refer to that file are
instantly updated to. You can do this by
just telling cloud I want to adapt the
intro scale. I want you to add the
reference files to the ben iOS and make
the scale reference the files instead of
having them in the reference files
inside the scale. Now a couple of things
become important at this level. Firstly
the setup and the file structure are
important to get right because of course
you're managing a large amount of
context. Now that's why I highly
recommend you use Obsidian which is
basically a free tool that helps you
visualize a folder on your computer with
some extra benefits. You can download
Obsidian for free by just going to their
website. But it's important to
understand that Obsidian is not a
cloud-based software. It's just a tool
that helps you visualize, organize, and
structure a folder on your computer in a
better way. As you can see here, because
of course trying to do that inside of
the actual folder with a growing context
like this, it becomes hard to do. We
also get a nice graph view here to see
all the relations and connections
between all of our context files. And
then for this file structure here, it is
a nuance topic. There are some best
practices, but it will depend on your
unique situation, your business, and
your way of doing work. Now, I recently
did a full tutorial where I show initial
file structure that I've seen work well
for most businesses or solopreneurs,
which I'll add in the link in the
description below, too, together with a
plug-in that we've develop developed
that you can install in cloud code or
cloud co-work that walks you through
setting up this initial file structure
with the context for yourself. Now, that
plugin you can download and use together
with all our other plugins and skills
we're building out internally in my AI
accelerator in the first link in the
description below. You also have more
in-depth step-by-step guides on helping
you set up this OS and one-on-one uh
live help and multiple Q&As every week.
So, if that's interesting, you can check
it out in the first link in the
description. But you can definitely set
this up yourself. I think my last video
will help you a lot wrap your head
around the file structure. And it's also
important to understand that the file
structure and the context will grow
naturally and fall into place more and
more uh the more you use this. So the
important thing is to just get started.
Now secondly uh your cloud MD becomes a
much more important to optimize at this
level together with potential index
files. Now what is the cloud MD? The
cloud MD is essentially an instruction
layer between your agent and the
Obsidian vault or your OS folder. And it
basically makes sure cloud knows where
to pull context from in a situation and
where to update it. So it's just routing
it to the right place which you can
imagine becomes a lot more important
when the context grows. So you can see
here in this chat in coowwork where I
give it access to my OS folder. It has
an instructions document or the clock MD
here. And this basically lays out how to
use and navigate the folder.
instructions on what to do at the start
of every conversation, how to route
between knowledge with information on
how the folder is structured,
information on Obsidian syntax, how to
add wiki links, and rules on how to use
context inside of this folder. Now,
again, even the Cloud MD will naturally
evolve and get better the more you use
it. And Clot can create the initial
version itself. Our plug-in will also
help you with a cloud MD instruction
that worked well for us. And then Andre
Karpathy, one of the leading AI
researchers, recently added a new layer
to this too where if your context grows
even more, you can start using index
files in each of the subfolders. So your
agent understands better how to navigate
each of the separate subfolders. For
example, you can see on the shared
context file, I have another cloud MD,
which in this case we just called cloud
MD, but it could also be called an index
file with more information on how this
specific subfolder is structured, which
cloth can read to know how to navigate
this folder. And we have another one
here for each of the subfolders. Again,
this is something you want to start
thinking about when context is growing.
And Cloud can help you out with building
uh these documents, of course. Now,
thirdly, what's going to be important is
you probably need some of these
scheduled tasks to make sure your second
brain is up to date, just like I showed
you with the meeting transcripts, but we
can also do this for your daily
analytics, task lists, your CRM
pipeline, whatever is relevant to you.
And fourthly, uh which is an important
one, uh is there is a maintenance aspect
to this. I'd highly suggest going
through your files on a weekly basis to
make sure things are going right. Are
there no duplicates? Are the documents
put in the right place? Are there any
conflicts in the context? And you do
want to dedicate some time to this,
especially at the beginning because it
will take time to get this right. And
lastly, you will need to start using
this cons consistently. The only way
this is going to work for you is when
you use it a lot because the more you
use it, the better it will get. And
there is some learning curve attached to
this. I can tell you I'm definitely not
there yet, but it is getting better and
better and it's making a big impact on
the relevancy of my AI outputs across me
and my team's chats. But it does take
time to figure out what file structure
makes sense for you and to test its
capabilities. Generally, I try to
approach this with a mindset of trying
to let AI clock code or clockwork or
wherever you use AI become your main
operating system for work because if you
do that, you'll slowly but surely fill
in the gaps to actually make it become
your main operating system. And this is
where we're heading anyway. So you might
as well be early. Now if you want to
take it to the last level, which is
going to be a game changer for anyone
who runs a business, is to actually roll
this out and sync this entire context
data set and the skills across the
entire team. So all of your team
members, AI agents instantly become far
more powerful for your business. Now,
when rolling this out for teams, a few
things of course become important.
First, the file structure will probably
have to change a bit, and you'll need
some more files for using this across a
team. For example, in my business OS,
you can see I have a few more folders
like my departments, my team and their
roles and plugins and skills so they can
be easily shared across the team. Again,
if you want to learn more about the file
structure, I covered it in full in that
last video which will be in the link in
the description below. Now, for your
team members to get to this initial
setup, you can of course just share a
zip file with them uh of the entire
context doc so they can install it. And
then second of course when you want to
share this across the team updates need
to actually be synced across the team
and ideally in real time. Now we've
explored multiple options of doing this
and because of course they are local
files it's not extremely straightforward
to do but for syncing across a team uh
you have multiple options. First you can
use GitHub. Second you can use Obsidian
sync which is a feature of Obsidian.
Third you can even uh launch a
self-hosted solution to do this. But the
best option we found which we're
currently using is a plugin inside of
Obsidian called Relay. Now this is a
community plugin inside of Obsidian
which you can find here by going to
settings, clicking on community plugins,
go to browse, type in relay and from
there you can install it. Once you've
installed it, it'll be listed under your
community plugins. And through relay now
I can decide for each of the folder to
which of my team members these updates
need to be synced to. And through this
any change anyone in my team makes in
any of these contact stocks will it will
automatically be synced and updated
across anyone in the team in real time.
And you can use relay uh for up to three
people uh for free. But with this setup
of course uh permission settings become
important too. Not every team member
should be able to update every file or
not even uh every team member should be
able to see or access any file. So you
as a business owner of course need to
control some of these permission
settings. Now, unfortunately, this is
not very straightforward to do on Relay
yet. We actually talked to the founder
of Relay and this feature is in their
pipeline. But in the meantime, we've
built our own version or custom setup on
top of this relay plugin that actually
gives us these permission settings. And
see that we have installed here our Beni
relay plugin, which is just our version
of this app with permission settings.
And now, for example, I can still make
updates in this general context folder
uh with the important documents, route,
strategy, etc. But my uh team members
only get read access. So they can still
use these documents but they can't
actually update them. This is what one
of my team members would see a little
lock with this is a readonly file. Now
this setup is a little bit more
technical. Um so if you want to set this
up we have full guides together with all
the other guides on how to sync across
team members with GitHub and other ways.
So if that's interesting to you and you
want access to our customized plug-in
again you could check out the AI
accelerator. And then lastly, of course,
once you have set this up, it is key to
have one person in the business really
be the operator and the manager of this
context layer because it does require
maintenance. It requires effort to
actually keep this updated and well
functioning across the business and this
is going to take some time. So someone
needs to be responsible. Now again, I
think syncing and permission settings
become a lot easier very soon because
there are a lot of businesses trying to
figure this out and there will be more
of an infrastructure around this very
soon. Um, but this is the way you can do
it right now. Now, that's it for this
video. Thank you so much for watching.
Again, if you want more step-by-step
guidance on setting this up for
yourself, uh, multiple weekly Q&As and
unlimited one-on-one live tech help, you
can check out my AI accelerator in the
link in the description below. Thank you
so much for watching. If you got any
value out of it, I highly appreciate a
like and a subscribe. It really does
help me. And if you want to learn more
about cloth co-work and obsidian setup,
انقر على أي نص أو طابع زمني للانتقال إلى تلك اللحظة في الفيديو
مشاركة:
معظم النصوص تصبح جاهزة في أقل من 5 ثوانٍ
نسخ بنقرة واحدة125+ لغةالبحث في المحتوىالانتقال إلى الطوابع الزمنية
الصق رابط YouTube
أدخل رابط أي فيديو YouTube للحصول على نصه الكامل
نموذج استخراج النص
معظم النصوص تصبح جاهزة في أقل من 5 ثوانٍ
احصل على إضافة Chrome
احصل على النصوص فوراً دون مغادرة YouTube. ثبّت إضافة Chrome للوصول بنقرة واحدة إلى نص أي فيديو مباشرةً من صفحة المشاهدة.
