The Official BMad-Method Masterclass (The Complete IDE Workflow)
By BMad Code
Summary
## Key takeaways - **IDE Workflow is Key**: The BMad-Method can now be fully run from within an IDE, eliminating the need for a web UI and offering a more streamlined development experience. [11:15], [03:12:00] - **Analyst Agent for Brainstorming**: Mary, the business analyst agent, is highlighted as potentially the most special agent, capable of aiding in brainstorming for any type of idea, not just software development. [12:04:00], [12:37:00] - **Advanced Elicitations: Push the LLM**: Advanced elicitations are crucial for challenging the LLM to provide better results, acting like a 'cattle prod' to ensure the best possible output beyond average responses. [30:52:00], [30:56:00] - **Sharding Documents for Context**: Large documents like the PRD and architecture are automatically split into smaller pieces using the 'shard' command to maximize agent context and prevent overwhelming the LLM. [05:43:00], [59:55:00] - **Developer Agent Needs Strict Guidance**: Developer agents are described as 'very, very dumb' and require clear, sequential user stories with detailed acceptance criteria and context from architecture documents to perform effectively. [38:14:00], [53:35:00] - **QA Agent for Code Validation**: Quinn, the QA agent, performs a deep examination of the developed code and story implementation, checking for compliance, potential errors, and ensuring adherence to project standards. [01:12:46], [01:13:17]
Topics Covered
- Your AI agent should be a coach, not an oracle.
- Use 'hindsight' prompts to stress-test your product ideas.
- The key to reliable agentic workflows is context engineering.
Full Transcript
Hey everybody, my name is Brian and I am
the creator of the BMAD method. This is
going to be a long video, but I want to
show everything involved with using
Claude Code, all the agents, all the
options, all within the IDE. This is the
first time I've done this, and I want to
show you so many details and just how
cool some of these features are that
I've never demonstrated before in a
video. And I've never seen anybody else
show some of these techniques also. But
there are chapters here, so feel free to
jump ahead to the chapter that you need.
But honestly, I really suggest watching
this through from beginning to end.
There's so much good information that
I'm going to be sharing with you today
that this is this is just going to be so
beneficial for you. You might not be
using Claude Code, but there's going to
be a ton of information in here. Trust
me when I say if you can follow this,
this applies to all of the other IDEs.
So, install it for whatever IDE you
want. A lot of what I'm going to be
showing applies to all of them. Also,
please give it a thumbs up. It helps the
channel. Subscribe, tell your friends.
Share this. Spread the word about this.
So many people are using this in their
day-to-day lives. Um, people tell me
every day they're using this at work.
They're using it for so many different
things that I never even imagined. Let's
dive right in. You always want to come
to the BMAD method and see what's new.
If you come to the BMAD method, this is
also the easiest way to get started. And
I really do suggest that you start out
here. you can come to the read me and
you can see here's the link to the
YouTube where I'll have this video and
other tutorial videos posted. Also,
while you're here, if you're a fan of
the BMAD method or if you try it out and
enjoy it, I will say come up here to the
top and then over to the right of the
sponsorship button. There's a star over
here. Please click star. This helps
other people also find out about this
project on GitHub when they're looking
for agentic tools and helpers to help
take their vibe coding to the next
level. And this is going to be showing
you how to do a green field development
of a very simple to-do application. As
always, I say take this method and make
it your own. This is not meant to be the
only way to do it, but it is meant to
give you a framework so you can mold it
to your own way of working, your own
type of project. So, you'll want to
subscribe to the channel, hit the
notification bell so you know exactly
when that video comes out. But first,
let's get to it, everybody. All right.
So, aside from uh these links here, if
you scroll right down in the read me,
you're going to see this workflow user
guide. This user guide is very simple to
follow. You can read this, but if you
scroll down, look, there's two diagrams
here, and they're very easy to follow.
This shows you exactly how to use the
BMED method, whether you're doing it in
the web or in the IDE of your choice.
First of all, you have a project idea in
mind, right? You have two choices. You
can use the analyst. The business
analyst is here to help you further
refine your idea before you go to the PM
and start making your PRD. So, I'm going
to show you how to install the VMAD
method and actually get set up in cloud
code to start using the analyst. And I'm
going to show you the optional
brainstorming and then also how we
create a product brief which will then
make the rest of the flow so much
easier. And what's so exciting is you
can actually do everything with the BMAD
method now all within cloud code, all
within the IDE. So if you so choose, you
do not have to use the web and the full
team stack anymore if you do not want
to. I actually enjoy and have had great
success using the BMED method fully in
the IDE and I'm going to show you how to
do that today. Really quick though, we
have the analyst, right? And then it's
going to go to the product manager which
we will then go to the PO
and then we will come down to the
development phase. But take your time
really understanding this. It might look
complicated but it's actually a very
simple flowchart. This is if you want to
use the BMAD method for a brownfield or
green field development to take it from
ideation all the way to actually
developing it in the code. Right now you
see the only thing I have here is the
BMAD method. Now, you do not have you do
not have to install the full GitHub repo
BMED method locally. You will only clone
that if you want to create your own
expansion packs or maybe make changes to
the core system or contribute to the
project, which I would love by the way,
but you don't need to do that. I'm going
to show you how to use the BMAD method
in cloud code without even cloning the
GitHub repo. So, first of all, let's um
create a project. But what we can do is
from any directory. I suggest doing it
from the dev directory. We are going to
type npx and then we're going to type
bmad- method
install. We are going to use the bad
method 4.33.1. We will just hit enter to
accept that. Let's enter the full path
of our directory. Since we're in the dev
folder, I'm just say dot meaning this
folder. And then I'm going to say slash.
And now I'm going to put the name of the
new directory that I want to create my
project in. Since we're doing this from
scratch with no starter project, I'm
just going to say simple
to-do. That's the name of our project.
Just a very simple to-do application.
And hit enter. It's fine if that
directory doesn't already exist. If
you're doing this in a brownfield or if
you already have started your project
maybe with a template, you'll just give
it the directory to your project and it
will add this to an existing project. So
then right here, this is a multi select,
but we're just going to go with the
default and select the BMAD core. The
BMAD core is the agile workflow AIdriven
that we're showing here. So just hit
enter. And then we are going to shard
the PRD, which means we're going to
generate a large document, but then
we're going to split it up into smaller
pieces. So just accept the default here.
And then the same thing with the
architecture. The architecture file can
be very large depending on what type of
project you're building. And so you'll
want to shard the architecture into
smaller documents and this is all
automated. I will show how this works.
Um, but that is really going to help you
maximize the context in your agents so
they only load what they need instead of
having to load and read through a whole
massive file. They'll know which files
they need in the moment to load and it
will help them with their context. So
we'll say yes to that. Now, here's a
warning and this is just pointing out
this is also a multis select. You can
select one or multiple IDEs that you
want to install to. For example, let's
say that you are a fan of cursor and
cloud code. Well, you can hit spacebar
and cursor and spacebar and cla code.
Today, I'm just going to be doing this
in vanilla VS code. So, I am only going
to be using cloud code. So, I have
highlighted cloud code with the radio
button. Make sure that you actually
spacebar to hit the radio button. And if
you just highlight it like I have right
here right now and hit enter, this will
actually not install it for any idees.
Uh, one other quick tip is depending on
the size of your screen, you might not
see all of these options here. So, just
know that you can arrow up and down and
more options will appear and we're
constantly adding more idees for support
here. Hitting enter. And now this is
asking if we want to include the
pre-built web bundles. The web bud the
web bundles are not needed for working
in the IDE. This is kind of warning us
in case we thought it existed and we
typed in the wrong directory. For
example, you know, we could choose a
different directory location or we could
cancel and start over. But in this case,
we're going to say create the directory
and continue.
Okay. So, if we scroll up here, we can
see what did it do here. It installed
some commands for cloud code. So, these
slash commands will be how we activate
our different agents. And then we've
also installed some commands and these
are great because sometimes we want to
do something without necessarily
invoking an agent. We are now done. That
is the install of the BMAD method. I
know it took a few minutes to explain it
because I was going through the options,
but let's see how fast this literally
is. I'm going to do it one more time.
I'm going to do n px vad- method
install.
And I'm just going to give this a
different folder this time. So I'm just
going to say uh let's say dash demo one,
right? Just demo one. Enter. Enter.
Enter. Enter. Select my IDE. Enter.
Enter. Enter. Boom. That's done. 5
seconds and you're ready to go with the
BMED method. It's really that simple.
All right. And just to see that our two
projects are there. If we do ls, we can
see that we do have the demo one, the
one we just did, and also simple to-do.
Simple to-do is the project we're going
to use. So, I'm just going to do rm-rf
demo one. We do not need that anymore.
And now we can see that we have our
simple to-do. So, I'm going to go into
simple to-do
and I am just going to open up uh VS
Code. Now, right now we have no project
here, right? All we have is what we've
installed for our agents under BMAD
and our custom tasks or commands for
cloud code. And then we have the full
BMAD core system here. Before we go on,
I do want to say one other thing though,
and that is I want to thank everybody
that has actually taken the time to buy
me a coffee. I've talked to a few of you
and just know how much it really means
for supporting me and the BMED method
and the community here. So, I want to
thank everybody that's come and
supported the BMAD method. And if you
would like to uh you can go to buy me a
coffee, look for BMAD code or the link
in the description. Any amount helps and
I truly thank all of you. I like to use
claude code in the terminal inside VS
Code, but it doesn't matter if you use
claude code outside. You could use it
inside the warp terminal or your own
external terminal, uh, you know, T-Max
or anything, whatever you want to use,
but I just find it convenient to kind of
keep it all in here. It's easy to, you
know, just have the, uh, keyboard
shortcut to show and hide the terminal.
Um, but we have no other project, right?
So, as always, if you've used cloud
code, you should know how to do this.
You just want to type claude.
And we are going to launch Claude Code.
Of course, you will have to have
installed Claude Code and set up your
authentication, but I'm assuming you've
done that. If not, go to the Cloud Code
website, see how to install it, figure
out whether you're going to do the $20
plan on the 100 or the 200, and then
you're ready to go. Oh, by the way, I
have been talking to a lot of people
that have just been using the $20
version of Claude Code. If you're fine
not having the sonnet, excuse me, the
opus model, the $20 clog code will
actually get you very far depending on
how much you're doing. Uh, so just say
yes, we do trust this folder. This is
the first time we've opened this in a
project. You could run the slashinit
command, but I'm not going to do that
yet cuz there's not really anything in
this project for it to know about. After
we build a little bit of our project and
get some stuff in place, that's when I
would do clait. Well, if we go back to
the diagram,
we can see that the first person or the
first agent that we want to talk to is
the analyst because we're going to do
some brainstorming. And if you just hit
slash, if you selected cloud code in the
installation, you're going to see all
these all these options here. And if you
don't want to arrow through this list,
what you can do is you can just start
typing the part of the name of the agent
that you want. So, for example, analyst,
you can just start typing it like that
and it will highlight it.
Or if you know what it is, you can just
type the whole thing or you can arrow up
and down to the one that you're looking
for. So, here's the analyst. So, we're
going to select it. This is how simple
it is to use clawed code with the BMED
method. You saw it takes 5 seconds to
install, maybe 10 seconds if you want to
read everything. So, this is Mary, our
business analyst. I know a lot of people
think the power in the BMED method is in
the development of actual code,
but I think Mary the business analyst is
probably the most special agent in the
whole BBA method. If you are just trying
to think of something to build or maybe
you have a little bit of an idea, I
really strongly suggest that you try
using this business analyst with the
brainstorming method. This goes beyond
software development. You can use this
for any type of brainstorming, whether
it be in your day job or, you know, life
questions. I've tried many different
brainstorming tools, including
professional brainstorming tools that
have been developed by Google. For
example, in the Gemini gems, there is a
default brainstorming gem that you can
use. It's it's nothing like this. So,
you know, let's let's do that. So, we're
talking to Mary right now. Mary is
active. As always, you can do star help.
And so, now these are all of the
different things that are available to
you. Now you can either select a number
or you can type out the name of the
command or you can rely on the fuzzy
matching and usually the LLM will get it
right. So I'm just going to select
number five though which means
brainstorm and I'm not going to give it
a topic. I'm just going to say number
five. Even though Cloud Code supports a
context window that seems to allow you
to go on and on and on without ever
getting close to compaction, that
doesn't mean you should. So, when we're
done with the analyst brainstorming, it
will give us a brainstorming document.
It will save it to a folder called Docs,
I will start a new chat. I like to start
a new chat every time I have an output
document, then I'm ready to go. So, even
if I'm going to talk to the same agent,
I will clear Claude or start a new Cloud
session and then reload that agent and
start going. So, here we go though.
We're going with uh Mary and first she's
going to ask us four questions. So, we
are going to use our Super Whisper. What
What is this brainstorming about? I
would like to create a simple to-do app
to demonstrate the BMAD method. So, this
should be a very simple NodeJS
background application. But what I
really want to brainstorm about is some
of the ideas we can do to at least make
it a little bit interesting. So let's
play some of your brainstorming games
and see what we can come up with. So
we'll let it spit all that out. So
question two, are there any constraints
or parameters that I should know about?
Well, I already mentioned that we're
using Node.js and I want this to be a
very simple, straightforward small
project just to demonstrate the BMAD
method with Node.js and TypeScript and
then we will submit that. So this is how
you work with the agent. One of the
magical things about the BBN method is
it not it is not just like a taskmaster
or some of these other things where you
just throw out your idea and then it
does all the thinking and working for
you. As a matter of fact, what you're
going to see with this brainstorming is
it is actually going to push on you to
get your ideas out. It is going to be
your coach. So that is what is special
about this. It is not checking your
brain at the door and doing it all for
you. It is going to work with you and
it's going to bring the best out of both
of you. The BMAD method, by the way, is
all about elevating yourself, learning,
and also elevating the LLM. So, as a
whole, you are both better collectively
than each one on its own. That is the
special sauce of the BMED method. That's
that's the secret everybody. Are we
aiming for broad exploration of
possibilities or more focused ideiation?
Let's go broad. Let's go crazy. I want
to show just what you are capable of as
the amazing brainstorming agent that you
are. Nothing wrong with flattering your
LLM every once in a while. They're sure
going to do it to you. Yes, definitely.
And that is also the default. But when
we're done here, this will take all of
our brainstorming and give us this
amazing artifact. This is one of the
coolest artifacts in one of the most
recent additions to the BMAD method.
This was not even in the original
release of V4. I thought this would be a
V5 thing, but I love the brainstorming
document output, and I think you'll love
it, too. We haven't gotten to the
brainstorming yet, but here we go. You
have four options here again. So, you
can pick the techniques, and there's a
whole list of 20 brainstorming
techniques built into the BMED method,
or you can let the agent recommend it
based on your project. Or you can do
random creative chaos. This one is fun.
You never know what it's going to get,
but sometimes it picks some of the
wackiest ones. or the progressive
creative journey where we start super
broad with wild what if type scenarios
and there's five of those and then
gradually go down to focus. I just want
to show you all of the different options
that we have here. So let's go to number
one this time. All right, so here we go.
You can see that there's quite a few
here and there's actually more coming in
the future. So we have what if
scenarios, analogical thinking, reversal
inversion, first principles thinking,
structured frameworks such as the
scamper method, the six thinking hats,
mind mapping, collaborative techniques
such as from the improv world, there's
yesand, there's brainwriting, roundroin,
random stimulation. There's deep
exploration with the five W's, a classic
brainstorming technique to really do a
root cause analysis. There is
morphological analysis, advanced
techniques such as forced relationships,
assumption reversal, role- playinging,
timeshifting, resource constraints,
metaphor mapping, questiontorming. But
with the power of the BMAD agent, Mary
here, she's going to guide you through
these and help you come up with stuff
that you never even probably imagined.
You're going to see how with
brainstorming, we can even make a to-do
list super interesting and exciting. If
we would have selected one of the other
techniques up here, those also involve
multiple selections from here and it
will work through each of them one at a
time. Usually 10 to 15 minutes is what
it takes for each one. So this can be a
lengthy process, but look, if you're at
the beginning of embarking on a really
large project or you have some ambitious
ideas, it is worth doing this
brainstorming to really come up with,
you know, super creative just
interesting ideas. take the time here to
build for the future. Let's just pick a
couple. Somebody in the community was
suggesting that they would love to see
six thinking hats and also the five W's
built into the BMAD method. And then
they realized that they were actually
already in here. So just for them, I'm
going to select number six, the five
W's. So we'll do 11. Uh let's do role
playing. Role playing's always fun. So
now the agent will remember our choices
and it's going to work with us one at a
time. And remember here, we are not
asking Mary to brainstorm and come up
with this for us. She's going to guide
us through this exercise. This is your
own personal coach to brainstorm on
complex, creative, or interesting ideas.
Folks, this is worth I don't even know
what. You know, you can pay experts to
teach you how to brainstorm or be your
brainstorming coach and it's going to
cost you thousands of dollars and you're
getting this for free and we're doing
all of this in claude code. So, this is
amazing. So, okay, what do we got here?
Well, excellent choice. So, first we're
going to start with the first selection
number six, which was the six thinking
hats, five ways, and then roleplaying.
This is going to generate some seriously
creative ideas. And again, this does not
have to be about apps. You can
brainstorming about everything. Maybe
you want to brainstorm ideas for a
vacation or brainstorm how you're going
to retire in 10 years or brainstorm, you
know, what you should do with your life
or how you should unscrew up something
in your life. I mean, this can be used
for anything. It is amazing. But, okay,
so what are we going to do here? White
hat. So, the six hats, there's going to
be six different colored hats. So, the
first hat is the white hat. What factual
elements could make your to-do app
interesting from a technical data
perspective? So, think about it. What
kind of data could it track? What
metrics, what information could it
capture that most to-do apps ignore? So,
I would have never asked myself that
question on my own. And I'm already
thinking now, wow, I was just thinking a
simple to-do app from the white hat
perspective. It's now getting me to
think about some really interesting
aspects of a to-do app. So again,
there's no right or wrong answer and I
don't even know if I'm going to come up
with anything interesting. This is not
the actual product that we're going to
create. We're just throwing ideas out
there. When we get the report at the end
of all of this, it's going to help us
distill down what is interesting to then
go to the next agent with. I think it
could track how much time elapses
between when they enter a to-do and when
they mark it done. Also, it would be
interesting to correlate the length or
size of to-dos and the complexity of the
to-dos and how that affects how soon the
to-dos get done. I love using Super
Whisper, by the way, or Whisper Flow or
one of the other many voice to text
ones. So, I'm just going to say yes. And
one nice thing about working with clawed
code in the IDE is that as you work
through each section of whatever
whatever document you're working on, it
will start creating the document for
you. So you don't have to worry about
being lost if you if you lose the chat,
for example. I know sometimes there's
anxiety when working with the web agents
that you might lose it and it's hard to
get the full document out, but here
you're building up the document as you
go. And we already have our
brainstorming sessions here now. Not too
worried about what's in there yet. We're
just going to keep going. So now we've
gone to the second hat. This is the
yellow hat. Optimistic benefits. So look
friends. Um this would take too long to
go through all of the brainstorming. So
I'm going to go through all three
brainstorming techniques. All right. And
look at this. We got our brainstorming
session complete. We did all three
exercises. Let's now look at the
document that it produced. So we will go
to open preview. This is actually going
to give us the output and keep a record
of the brainstorming that we did. So
first we did six thinking hats. critical
concerns, creative alternatives, and we
found some fun ones in there. And then
we went into process control. So, it
really got me thinking about a lot of
different interesting things. This led
to so many different things. Then we did
the five W's. The five W's was a lot of
fun. The first why is why is complexity
completion correlation valuable? And
then I gave it an answer. And the answer
was because people consistently
underestimate how long complex tasks
take. And the way the five W's works is
then it's going to say okay well why
does that or why is that the case and
then I gave an answer and then it's
gonna say well why is that the case why
is that it's a way to really dig down in
a somewhat annoying way but it's kind of
fun doing it with the AI so then we get
into technique three which is role
playing this is one of my favorite
because the AI will just invent some
sort of role playing and you never know
what it's going to give you so here I
came up with the overwhelmed freelancer
the context is You're juggling 12
clients, 47 open tasks. You have a
current to-do app, multiple of them,
making matters worse. So, how do we
triage intelligence? Show me what
actually needs to happen today versus
what I think needs to happen. Client
workload balance, visual dashboard
showing task distribution across clients
to prevent overcommitment. Imagine the
project that you really want to launch
in the market and what this
brainstorming agent can do for you. This
is next effing level. Okay. And then it
gave us more because I wanted to do more
brainstorming. So, we didn't just start
with the overland free overwhelmed
freelancer. We talked about the
perfectionist student and what they
would get out of a to-do app, the
executive assistant, the ADHD creative.
So, I wouldn't have even necessarily
thought of all those personas. uh the
analyst Mary helped me come up with
these different role playinging
scenarios and we talked back and forth
and then it gives us an executive
summary because okay so it's great so we
did all those things we're not going to
use all of that right it then we kind of
talked about what was important to me
and I kind of like gave it some of the
key insights and we came up with some of
these key insights together transforming
from task management to behavioral
intelligence your simple to-do app
becomes a personal productivity research
lab that helps users understand their
actual work patterns versus their
assumptions
Tell me that is not powerful. And that
is again just from like let's brainstorm
about a to-do app and this is what we
came up with together. And now here's
what's really cool is it breaks it down
into what can we do right now? What
might we want to research and do in the
future and then what are like the
moonshots or just the creative ambitious
things we want to do later. And again
this could be like any aspect of your
life. So use this use this in your
personal life use it for work. You can
see now just how powerful the BMAD
method is and you don't even need to be
in the web. We're doing this all with
cloud code. And as a matter of fact,
this is just on the cheapest model for
claude code. If I do slashmodel,
you can see that I'm just using sonnet.
I'm not using opus and I'm not using the
default. I'm just keeping it on set.
You'll learn over time which model to
use. It's kind of an intuition that
you'll develop. And trust me, even if
you don't consider yourself a strong
developer or if you're new to this,
you'll pick it up. You'll understand.
It's it becomes very intuitive to know
which model to use. So, I would
recommend though just so you can learn
that. Keep it on sonnet for most things.
We are done with this. We've saved
everything to a document. This is the
most important thing with the workflow
in cloud code or really any agentic IDE
that you're using. we are going to start
a new task with either the same agent or
a new agent. So what are we going to do?
We're going to stop clear the context or
if you don't want to clear the context
and keep that conversation history for
later, we'll just kill the window, start
a new one, and restart Claude. Now, you
might not know this if you haven't used
Claude before. Let's say we close this
and then we realize, oh, we're not done
with the conversation that we were just
in. If you hit slash and just hit the up
arrow a few times or type slashres,
you can now see the chat that we were in
here before. So if you've had multiple
chats, you'll see a list here. There's
only one right now. If you highlight one
in blue in this theme, if your theme, it
might be a different color. And now
we're back in the chat that we were. So
that is how you get back to a previous
chat. But we want to start a new chat.
So again, I'm going to just close that.
I could have done slashcle, but I want
to retain it. So, I'm just going to do
cloud again. And now we're going to
start a new chat. Now, we're still going
to talk to the analyst in this case. And
now I want to show you how to talk to
the analyst to do basically the most
important thing out of the analyst that
you really want is the project brief.
Uh, by the way, the analyst is a totally
optional thing. You don't have to do
brainstorming. You don't have to do a
project brief. It is something that you
can do. Again, it really depends on the
complexity of your project and how much
you already know. If you if if we were
just doing the simplest to-do app, I
would go right to the PM, create a PRD
for a simple to-do app. You can go right
from PRD with its epics and stories
directly to development. I will show all
of that in the A to B full walkth
through, but right now I just want to
continue showing you this basic path of
using cloud code. So, we're going to
select number two right now. And this is
now going to help us create a project
brief. And you'll see right there it
loaded elicitation methods. That is one
of the other big awesome things that's
built into the BMED method. The BMAD
method is built up of YAML templates
which have two things. Basically, the
outline of the type of document, but
more importantly, and this is what makes
it, I think, more powerful than any
other method I've seen so far, but other
people are starting to catch on to these
ideas. Embedded in the template is
instructions for the LLM and how it
should work with you. Because as I said
a little bit ago, the beauty of the BMAD
method is not that you ask it for
something and it spits out a whole
document, but that it really works back
and forth with you coming up with the
best possible thing that you guys can
come up with together. So, first it's
asking, do we already have any existing
brainstorming results? Market research.
Sometimes it's best to look out in the
marketplace, see what's out there, and
build your own version of it, maybe with
your own twist. It's a big market out
there. So don't feel like you always
have to be the first to market with
anything. So maybe you want to do market
research and see what is out there or do
a competitive analysis and see what your
competitors are doing differently. But
what we're going to do is we're just
going to take our brainstorming. We're
going to drop it here and say here's our
brainstorming session results. We can
use this to kickstart our work in
interactive mode in creating the project
brief. You don't have to use these exact
words by the way. So don't don't you
don't have to memorize the exact words.
Talk to this as if you were talking to
your business analyst. Now this is going
to kickstart it. So the nice thing here
too is we started a new chat, but you're
seeing because we saved it to a
document. I'm able to use a brand new
context. It's not getting polluted. It
only has the output from the
brainstorming session. And so it just
keeps it lean. And so now it's going to
ask us some questions. It's going to
pull things out of our brainstorming and
probably get us through the PRD a little
bit quicker than normal because you can
see here, for example, it already knows
what the executive summary is because
we've told it when we were doing the
brief. I am only going to show this
first section and then I'm going to
finish the document and then we'll move
on to the PM. What you can see is after
you complete the executive summary, this
is just one part of the overall brief by
the way. It already figured some stuff
out from our previous brainstorming. we
can correct it or we can tell it to
change things. So, you'll want to read
this and obviously work with it. But at
the end of each section, you're going to
have multiple options. These are called
advanced elicitations. Whether you're
working with the PM or the architect or
the business analyst right here, every
section of a document, most of them will
ask you different advanced elicitations
related to the document you're working
on. Now, unlike brainstorming where it's
asking you, the user questions, this is
where another powerful aspect of the
BMAD method kicks in, and I haven't
talked a lot about this before, so this
is very important, but this is where you
push on the LLM to do better. When you
talk to any LLM, whether you're just
using the BMAD method or not, or you're
just using Chat TBT, if you picture a
bell curve, you know, you're kind of
getting the average slice of the entire,
you know, corpus of information in the
LLM. It's giving you the average
response. It's, you know, not
necessarily great. It's not necessarily
weak. The advanced solicitation is your
chance to really stick the cattle prod
to the agent and say, "Do better. Make
sure that you have put this through the
fire and you're giving me the best
possible result." Challenge the scope.
We can force it to brainstorm. So, just
like it was having us brainstorm, we can
now turn the torture back around on the
LLM and have it brainstorm different
ideas. There's actually many more
advanced elicitations. Again, just so it
doesn't overwhelm you, it's just giving
some of them. Please do number four and
number five. And by the way, while we're
waiting for that, notice that it did
already start our brief. This is our
project brief. So again, the beauty of
doing this in the IDE with cloud code is
that we are generating the documents as
we go. But first, it did critical
assumption testing.
So, it's basically, by the way, this is
just another great thing to do with any
product idea that you have. And so, I
love that the LLM thought about this.
Look at all of your ideas for your
product and really stress test them and
make sure that they make sense,
especially if you want to build an MVP
first to just get something to market.
And then we also did explore alternative
solution approaches. So at the very end
of it, it kind of synthesizes everything
and it gives some recommendations based
on what it was doing with itself right
there. Sure. This is just a demo. So
we're just going to apply them all,
update the section, and then we'll move
on to the next section of the product
brief. Let me say please uh subscribe to
the channel if you're enjoying this. So
much more content is coming out, my
friends. You know, I know these videos
don't come out as often as I would like
to because I'm also constantly working
on actually improving the BMED method.
I'm not here to just hype up new things,
new products, or, you know, even tell
you I'm using this, but I do want to be
able to use this forum or this platform
to share the changes with you. But also,
I spent a lot of time actually working
on this. But subscribe to the channel,
hit that notification bell so you do
know when the next one comes out because
it is a little sporadic sometimes. And
join us in the Discord. I'm going to
finish this up. We will look at the
whole product brief when it's done, and
then we'll move on to the product
manager. All right. So, I'm going to
pick up the pace here a little bit. We
are done with the brief. We've gone
through the whole thing. The brief is
now ready for the PM handoff. So before
we do that, I just want to show here's
our product brief. Since we're in cloud
code and not in the web, it just keeps
updating the document for us. We have
the executive summary, the problem
statement, pain points, why we're doing
this, what is the proposed solution, the
target users. So this says like the
demographics, who do we expect to be
using this? Why would they use it? How
are they each going to use it
differently? And why is it valuable to
them? This is so important. If you're
just wanting to put any kind of product
in the market and you're just thinking
you're just going to vibe vibe code
something, throw it out there and be an
overnight millionaire.
Maybe you will be, but you have to
consider all these things. You have to
understand the target audience that
you're looking for. So that's the beauty
of this product brief. It's going to
really help you figure out a lot of that
and make sure you're building something
that makes sense or might indicate that
you want to pivot on something because
it's not actually going to achieve what
you want to achieve. Why waste a lot of
money, effort, and time building
something that's not going to go
anywhere when instead you can do this
little bit of research up front, right?
So, we're showing this this whole
document. The point of the product brief
is to really work back and forth with it
and produce this document that helps you
understand what it is you're going to
build at a most fundamental level. As
always, what did I say? Even in cloud
code, even though we could go on and on
and probably talk to three agents in a
row without running out of context,
we're still going to start a new one. So
again, we can just kill it and start a
new cloud session. I'm going to show you
a different way this time and that is
just slashcle
and this is going to clear the
conversation history and free up the
context is it's it's basically the same
thing. You could also do compact. I do
not recommend it. And if you ever see
your IDE is showing you the warning that
you're almost ready to start compaction.
That means that means you've been in the
chat for too long and you probably want
to wrap up what you're doing. There's
different techniques for what you for
that to get out of there, but you
generally do not want to rely on
compaction.
It's just randomly going to forget
things that might be important. So,
there's other techniques, but basically,
if you follow my method and you just
switch and do clear, but sometimes
you'll forget and all of a sudden you'll
see that warning and then, you know, you
have to adapt. But let's um
let's talk to the next agent. So if you
don't know what agent to talk to, again,
you can go back to the diagram. So we've
done this, right? So we we talked to the
analyst. We've done our optional
brainstorming. We didn't do the optional
market research or competitor analysis,
but you can try those on your own. It
depends on what type of project, but we
did create our project brief. So now we
are going to talk to the product
manager. And if we have a product brief,
we're going to give it to the product
manager. it will actually ask us. So, I
mentioned you might have a similar pro
or a simpler project. And if you do or
you already know what you want to do,
maybe you even already have a product
brief, you can just talk to the PM and
it will just ask you more questions if
you don't provide the brief. But if you
provide the if you provide the brief, it
kind of gives it a kickstart and it will
get the PRD done much quicker. But it
just depends what you want to work.
Again, if you're just doing a simple
app, you don't have to go through all of
this. just go right to the PM, create
the PRD, even tell it you want it to be
a technical PRD with architecture built
into it. You'll be coding in no time. So
there's multiple flows and workflows and
ways you can work this system. But
again, come back to this diagram if
you're not sure. So let's go back to
cloud code. So now again, I hit slash
here. We could just type PM. That'll
take us right to it. If we're not sure,
we can also just kind of look through
here
and then arrow on it. The product
manager is going to basically create a
PRD which is a product requirements doc.
Building a PRD is important and also
I'll say the three most important things
with this PRD is first of all it's going
to give us all the functional and
non-functional requirements and the epic
that is made up to meet all of those. So
right there we can see everything that
is going to be encompassed by the build
that we're planning here.
Secondly, it's going to help work with
us to figure out what is in scope for
the MVP and what we can potentially pull
out to have post MVP. Because if you can
make your app the simplest version of
the app that you can make that will meet
the initial goal, that's less risk, less
investment in a product. Maybe you want
to get it out there with just a minimum
set of features to test it in the
market. That's why you want to build an
MVP or minimum viable product. what is
the basic thing of this core? But the PM
here will also capture your other ideas
and if they're not part of the MVP,
we'll still hold on to those and those
will be post MVP epics that you might do
after the main build. So again, what is
most important here? It helps us
maintain MVP scope. It gives us the
functional and non-functional
requirements with the epics. And then
the the third most important thing, but
really kind of the key to the whole BMAD
method is in those epics, it is going to
create the user stories, but more
specifically, it knows that we are
creating these for very, very dumb
developer agents. Now, as I've said all
along with the BMED method, you cannot
just check your brain out and leave it
at the door and rely on it 100%. Right?
So, we will use the advanced
elicitation.
We will read the information ourselves
and logically think to yourselves, does
the story sequence make sense. If you
don't see a story, for example, that
says scaffold and you know, set up your
project or set up your accounts or
whatever you need, you know there's a
problem and something's been missed,
right? And usually that should be the
first one like project setup. You know,
learn to kind of spot those things as
you go. And there's also some advanced
elicitations that will help us discover
that. So, let's move on here. So, we are
just going to do create PRD. And I want
to show you before we've been doing the
number, we can also just say star and we
can actually just give it the command
name also or you could even speak it. Uh
do notice that there are two Yep.
There's two different PRDs here. There's
the regular PRD and then there's the
brownfield PRD. Brownfield is a term
that you will hear and you'll also hear
green field. And I realize some people
don't know what that means. So I want to
explain it and it's very simple. Green
field means like you're looking out on a
clean green pasture. It's all open.
Pretty much sky's is the limit. It's
going to be easy to build your
foundation. Nothing is there. Nothing in
is in the way. A brownfield is where all
the has flowed. It's a cesspool.
Maybe it's an existing application
that's existed for months or years or it
was developed by other people. Maybe you
understand it. maybe it has a lot of
issues in there.
If you're going to work with something
like that, then you'll want to follow
the path create brownfield and that will
have a few more steps in it to also
provided the context of the existing
application that you might be working
on. Number two, we're just going to do a
regular PRD.
Now, the PM is asking us, okay, do we
have the product brief to create the
PRD? So, again, this will kickstart us
to the PRD. If you were starting here,
you would say no and you would just
start asking its question or you could
go back and do it. But we're just going
to drag it in because we do have the
brief. And we'll say
here you go, buddy.
There is a lot of overlap between the
product brief and some of the sections
of the PRD,
but it's really going to translate that
from the language of the product brief
into functional and non-functional
requirements and a few of the other
details that we need in there. Okay. So,
let's see what we got here. So, it gave
us some goals. So, it understands what
we're trying to build, why we're
building it. We got some detailed
rationale here. Tradeoffs made, key
assumptions,
and areas needing attention. We have
options to do advanced solicitation. And
by the way, I love this. Just like
before, as we produce our document, we
are going to generate each section of
the document one at a time. So, it's
always going to have access to
everything it's done. It's going to he
keep kind of a cohesive vision of the
document we're building and it's going
to build it up section by section.
Beautiful. I love this. Now we are doing
section two. We moved on to the next
section. So this is the requirements. So
this is going to give us the functional
and non-functional requirements. Take
your time and read through this. Again,
getting this right is important. And you
really want to try to look, is there a
functional requirement here that you do
need or don't need. I want to show you
my absolute favorite advanced
elicitation to use here. Let's see if it
suggests it. But I want to show you how
you can actually find what other
advanced elicitations are available for
the BMED method. But what you can do is
you can go to data. We're going to go to
data and here are the elicitation
methods. We can see them here. So
there's explain reasoning, critique and
refine, analyze logical flow, assess
alignment with overall goals, identify
potential risks, challenge from critical
perspective, tree of thought, deep
dives, hindsight is 2020, the if only
brainstorm, agile team perspective
shift. This will take your different
agents and have them all look at it from
a different perspective. It's a lot of
fun. Uh stakeholder roundt,
metaprompting analysis. So some of these
are very correct. uh very creative. Um
and then we get into some of the new
advanced 2025 techniques that have been
added in recently. Self-ont
self-consistency validations, the rewoo,
which is reasoning without observation.
I love that. Rewoo, persona pattern
hybrid, emerging collaborative
discovery, red team versus blue team,
innovative tournament, escape room
challenge, proceed and no further
action. I want to tell you, I've done a
lot of research into how to really
prompt engineer and get the most out of
LLMs, and these are some of the best
techniques that I've found from Google
and Anthropic. But I've also studied
multiple prompting competitions where
expert prompters just compete, and I see
a lot of them using some of these very
cool techniques, and some of these are
just crazy fun. So, let's let's pick a
let's pick a wild one. I like the what
if if only hindsight is 2020. So what
I'm going to do is I'm going to just
select this here. This is another thing
you can do. Eventually these will all
just be in the menu at one time. Close
this now.
This looks at all your functional and
non-functional requirements and then it
imagines that we've built the
application and that we're I don't know
sometimes it does six months out
sometimes it does a couple years out and
it's an analysis where the LLM does this
conversation with itself reflecting on a
board meeting or something else where
people are like if only we would have
done this something would have happened.
So let's see what it did here. Imagine
it's 6 months from now the MVP is
launched. You're looking back at these
requirements. What if only statement
might you be making? If only we had. If
only we had included basic team sharing
features from the start. Non-functional
requirement one might be unrealistic for
bootstrap budget and single developer.
What if we hadn't promised cross device
sync and MVP? Functional requirement 7
adds significant technical complexity.
What if we started with single device
and added sync in phase 2? So what's
going on here? This is actually helping
you figure out how to cut scope out of
your MVP. Keep it a lean lean build so
you can actually get something to
market. Get it tested and then layer
those features in later. This is the
real agile way of product development.
Super powerful. So I'm going to say
apply changes and update sections.
Revise the requirements based on these
insights. Uh before you do that, you can
talk to it. You can question different
things. You can tell it what you want to
keep and not keep, but I'm just going to
do number one for purposes of time. So,
I'm going to finish this and I'm going
to get this up to the epics and stories
and then we'll continue together.
There's only one epic right now because
I told it just CRUD. So, foundation and
core CRUD operation establish the
project setup. As I said, always kind of
look out for that, right? Uh we are
going to have local database. So, it'll
do the database initialization and then
implement all basic to-do management
commands. add, list, complete, delete,
update, basic crud. So, since there's
only one epic, it's going to go right
into the stories here. So, epic
foundation and core. You'd want to read
this. And as always, read the stories
and acceptance criteria and think,
does it make sense? Is there something
being called out that actually relies on
something later? This is great. This is
how agile actually works in story
breakdown is each piece of functionality
can kind of be its own standalone story.
Can we list and search for the to-dos?
Can we mark a to-do done, delete a
to-do, update a to-do? You can also ask
it questions if you don't see the one
you want here and you don't want to
look. So, I'm going to tell a deep deep
think please that the order and
granularity of the stories actually make
sense. The sequencing must be perfect
with no story dependent on a later
story. It's checking to make sure that
they all follow the correct progression.
You're absolutely right. Let's deeply
analyze the story sequencing and
granularity
as there are significant dependency
issues in the current approach. So, do
we accept it? We'll say yes. So, does
this sequencing now make sense? So,
let's say yes. All right. So, looks like
Claude now thinks we have seven stories
in perfect sequential order. Do we
believe that? Well, don't check your
brain at the door. Let's close that
though because we're done talking to the
PM for now. But you do want to, you
know, go through your doc. Imagine this
is a big ambitious project with multiple
epics. Do you really just want to accept
some slop and move on and start
developing? Maybe. But I would suggest
take your time and really understand and
try to learn from this. Ask the LLM
questions. The LLM is not going to
always tell you to do this. The reason
we have agents and personas is because
it's to help you also have a mental
model. Not not only does it help the
agents perform better and keep focused
on their domain expertise and you know
tailoring responses to that persona that
they're interacting, but it's also for
you to understand that as they are
serving this role, talk to them as if
they were in that role. So, you're
talking to a product expert or an
architect or a scrum master. Ask them
why they're choosing things. What do
different things mean?
The beauty of this too is you don't have
to be embarrassed about asking
questions. You can ask this what people
might think is the dumbest question.
Look, maybe you don't know, maybe you've
heard the word database before, but you
don't even know what a database is.
Okay, you can Google it. It'll give you
a definition, but you could also just
ask why are you using a database for a
to-do app. It mentioned a name of a
database. You could ask it why did you
select this database? Why not something
else? And it'll give you an answer. So,
you can learn a lot just by talking and
at any time you can just ask the agent
questions about this document or any of
the documents we're going to do. Now, we
can start a brand new chat with the
architect. So, I'm going to open up
Claude again. So, we're in the brand new
cleared context. Nothing here. And if
you're not sure, you can always clear it
again just to make sure. Now, we're
going to do slashchitects.
Now, again, this is a very simple
project. I would probably try to just go
from PRD to stories. There is a workflow
that supports that, but I want to show
you basically how the architect works.
So, same thing. It'll either show you
the commands. If it doesn't, type
starhelp and you can see what's
available here. Now, here we have a
couple different architectures. Don't be
confused, but it's pretty simple. You're
either going to do a full stack
architecture, which is front end and
backend, maybe is a monor repo, which
means all the code is in one project
such as Nex.js, or maybe it is two
separate repos and two separate
projects, React for your front end, and
maybe a cloud service for your back end,
and maybe in two different GitHub repos.
Um, you also might be starting with a
starter project such as NECJ, Nex.js
with Superbase and Shad Saiyan and
Tailwind. If you're doing that, you
would want to generate your project, get
your boiler plate and share that with
the PM or the architect. So, as they're
doing the architecture, they already
know what you have, what your
technologies are, what is in your
requirements.txt if you're doing Python
or your package.json JSON if you're
doing TypeScript or JavaScript, Node.js,
excuse me. Create full stack
architectures if you're doing front-end
back end. Create backend architectures
if you're just building like maybe a
service project or utility or anything
that does not involve a web front end or
a guey front end. The one exception is
you might have a CLI. You will still
want to select create backend
architecture.
And then there is a create front-end
architecture. If you're only working on
a web front end, maybe you're working on
a website or maybe you already have a
backend REST API and you just want to
build the React app to call that REST
API. That's where you would say create
front-end architecture. And then
finally, our friend the brownfield is
back. So we can create a brownfield
architecture. And again, that's going to
involve a lot of the research that the
analyst and the PM did to understand
your current project. Feed that into the
architecture along with some of the
challenges of architecting against an
existing project. Creating our backend
architecture because we are doing what a
to-do app only in the CLI. It's just
running in the command line for
demonstration purposes. So, we'll hit
number two. Now, let me give you a tip
on working with the architecture.
Regardless of what actual IDE or system
you're using, you want to use the best
model you can. I'm showing you that this
can be done in the $20 version of Claude
Code, but I would recommend, especially
on a more challenging architecture,
use Opus or whatever powerful model you
have because you will get potentially
better results. Or do this on the web
with Web Gemini. save your credits and
then just get into the IDE after we have
our architecture. But I actually enjoy
doing the architecture here because as
you can see yet again
we are producing our document in real
time as we go section by section. It
gave us a sequence diagram. It gave us
some highlevel architecture. You know
what we're doing with the database, how
we're going to handle error handling and
global installation and uninstallation
process. Do we want to focus on any of
these? We can do alternative analysis,
stakeholder input, risk assessment,
resource impact, user experience impact,
technical feasibility, market
validation. You know, read what it gave
you and think about if any of these make
sense or if you want to do any of these.
And if not, just move on to the next
section. Also, question in this next
section what technologies it comes up
with. Make sure it knows what the latest
versions are or what the best version
is. You can also tell it to use the web
to make sure that it's considering the
latest technologies for everything that
it picks is this is going to create a
table of all the specific versions and
technologies we're using. If you don't
have this and tell it what specific
technologies, what what testing
framework you're using, for example,
what'll happen is later on your LLM will
be developing. It might be using justest
just because it sees that it's already
scaffolded and then it'll run into a
problem and instead of trying to fix it,
it might be sneaky and just try to
install a whole new test framework
without you noticing. It'll tell it but
maybe you're not paying close enough
attention or you walk away and you come
back and all of a sudden there's a new
test framework or there's another new
package. And that's because it's not
forcing it to stick to any list. When we
create this table here and shard this
out to a separate document, the dev
agent later on will always be aware of
this technology stack and will not
diverge from it. It's a very powerful
technique to make sure your agents are
all using consistent libraries,
packages, and versions. So, I'm going to
go through the rest of the architecture,
but just remember section by section you
really want to use the advanced
solicitation. So let me show you right
now just one quick example example and
then like I said I'm just going to kind
of run through the rest and we'll pick
up at the end. So let's say in here
doing data models. So this is great
because if it makes data models that
means later on the agents they'll always
conform to the same model. And much like
our technology stack you don't want them
producing different models on the fly.
But let's say for whatever reason
I don't understand why we're creating
these or I don't like them. So I'm just
going to ask a question. And this is one
thing you can do. You don't need the
prompt here always to ask just ask a
question. So say explain to me the
interfaces you created and I want an
explanation of why each one of them
exists and are we missing any? So the
architect is programmed to also be very
explanatory and explain in a very clear
way meeting you at your level. I
recommend talking to all these agents.
Even if you're not in the middle of a
project, you can load up these agents
and just talk to them. talk at them talk
to them about your ongoing project. What
did he tell us? He told us a lot. So,
let me explain each type script
interface I created and analyze if we're
missing any. So, the to-do interface
that's obviously important, but why it
exists? Type safety. So in other words,
an interface allows us to define a
object or a par or basically this would
be like the blueprint for an object that
we will create later and every object
must have a task string completed
completed at and an ID to make it
unique. Um so why does it exist? Type
safety. We just talked about that
database mapping. So that's interesting.
So what that's telling us is he's
setting up an interface here that will
later also map directly to the SQL
light. So we always make sure we have
the same properties and fields in our
local database. And it's also going to
help us define an API contract. The API
contract when it returns a to-do is
going to be guaranteed with each to-do
to return these properties. So if we
wanted something else in our to-dos, now
would be a good time to add it, not
later on when we're developing. So
that's again why you want to actually
pay attention and try to understand what
the architect is telling you. And if you
don't, just ask it. And if you read this
and you still don't ask it, play five
W's with it and ask it why you did this,
why you did that, well then why did you
do that? Well, why did you do that? And
on and on and on. Again, you can learn
so much just by engaging in discussions
with the architect to really start to
learn how software is developed in
reality. It's very powerful stuff. All
right. Right. And so the uh architecture
looks like it is done. We finished the
last section. It gave us the final
summary and the document is updated.
Let's just make sure that everything is
in the document. It's pretty large
document which is why we shard these
documents, but we have our success
criteria. Before we shard our documents,
just like with the PRD, you really want
to go through here and make sure that
everything makes sense, such as coding
standards. As a matter of fact, coding
standards is one of one of the most
important ones because this will be used
by the developer agent. Directory
structure. You want to make sure that
this makes sense. If you want to render
sequence diagrams, you can just
rightclick in VS Code and you can say
open preview and then you should be able
to see some of these nice looking
renders. If you do not have that option
in VS Code or in Cursor or Windsurf or
whatever you're in, you'll want to go to
your extensions, which is this icon
here. These are our installed
extensions. And you'll want to make sure
that you have something similar to
Markdown all-in-one. I like this one
along with Markdown preview mermaid.
This will give us mermaid documents or
mermaid rendering from this one. And the
markdown all-in-one will also let us
auto format markdown to make sure that
it's valid markdown. Here's our sequence
flow. Make sure that the flow makes
sense.
Here's our database schema. Again, if
you don't understand something here, ask
it. Ask it why I did this, but this
makes sense to me. Here's the source
tree that it's going to produce. This
controls where the LLM is going to look
for files, where it should put them when
it creates them. And again, we'll just
help it stay on the rail. Having a
source tree is a very important document
to keeping the LLM on track. I'm going
to open up this and I'm going to show
you how you can just talk to the agent.
In the coding standards, I really want
to make sure that the developer agent
will use Gab will use good Java do style
comments or JS doc style comments on all
public functions, public interfaces,
etc. So, please add that to the proper
section in our architecture under the
coding standards.
Okay, that was a little bit wordy, but
that's why I like using just the uh
speech to text because I can kind of
work it out in my mind as I'm saying it
and usually the LLM can get it right and
understand what I'm talking about. But
this should now add some documentation
standards so we will have good
consistent documentation. Now, now I'm
good with this. Now I think the document
is great and we are done here. So now
I'm going to close
cloud code. I'm going to close all our
documents and let's look back at our
project. We now have an architecture and
a PRD. The two key documents that we
need. We don't really need the LLM going
forward anymore. Looking at the
brainstorming or the brief. Every file
that you keep here is potential context
that the LLM might choose to load. You
can add restrictions. you can tell it
not to, but it's better sometimes to add
these to another folder. So, you could
add these to a project references file,
or you could just take them out and
store them, or maybe you store them
online, depending on how you're
organizing your project artifacts. But
whatever you do, you don't want them to
get pulled in and polluting your context
overall. Okay. So, we'll come back to
Claude use the shard command. So, if we
just type shard, we will see that
there's a task shard doc. And do we want
to use the MD tree? Yes, we do want to
do MD tree. So, okay. So, now it's
saying great, the MD tree command is
available. If it was not available, it
would suggest that you install it. So,
now I need to know which documents you'd
like to shard. Please provide the path
to the document you want to shard. And
I'll use the MD tree explode command to
automatically split it into smaller
documents based on level two sections.
So, first of all, we want to do our PRD.
So, I'm just going to drag it in there.
That's it.
And the default should be a folder
called PRD, which it is. And look at
that. It's already done. That's how
quick it is. We now have our epic list.
And here is the most important one. Epic
1, epic 2, epic 3 if there were multiple
with all of the stories because this is
what the scrum master is going to use to
build up our stuff for our developer.
So, that is great. So now we can say do
the architecture and we're just going to
drag it in.
So as you can see using the BMAD method
in cloud code is super easy. We don't
have to jump back and forth to the web.
It's super powerful and it just does a
great great job at following
instructions regardless of which model
you use in cloud code. Here's our
architecture again just in seconds. And
now this is the one that is really good
to shard because look we have coding
standards which has our documentation
and some of our other stuff in here that
we wanted with examples source tree. So
this shows us exactly what our project
structure is going to look like. Text
deck those three are critical and I want
to show you why so you understand this
regardless again of which tool you're
using. If you go into
Claude, since we're using Claude, we'll
look at the Claude version. And let's
look at what the developer agent has
here. The developer agent has a command
on here that basically tells it, look at
BMAD core, look for the core config.
And this is what tells the developer
agent what files it will load every
time. So the way the developer agent
works, I've explained this before but
I'll say it again. The scrum master is
going to take the highle epic and story.
It's going to read through various
architecture documents. It's going to
understand one story at a time and it's
going to create a very detailed story
for the developer, a lower level
developer story. This is a
self-contained file for the developer to
have all of the context it needs to
build the application.
We talked a few videos ago ago about
context engineering. This is the core of
context engineering. Giving the agent
exactly what it needs to build its
little piece of the kingdom, right? But
additionally,
not only is all of this information the
scrum master is going to put in there
exactly what it needs from the
architecture and other sources, the
developer is always going to load these
files that you have listed under dev
load always files. By the way, I didn't
have to make this myself. This file was
created on the install. But if you ever
wanted to customize it, maybe there's
another file that you always want the
dev to load, you can add it here. But
again, here's our key documents that
were sharded from the architecture, the
coding standards, the tech stack, and
the source tree. Take your time and
really go in there and make updates to
those if you need to and change them if
any of these things change. All right,
we've come a long way. If you're still
with me, thank you for sticking with me.
If you made it this far, I would love to
know who's actually still watching. Um,
because I bet this is a very long video.
I hope people are appreciating it,
though. So, if you've made it this far,
how about leaving a comment down below
and tell me what is the craziest thing
that you want to brainstorm with the
brainstorming agent. Give me some of
your wildest ideas. And if you actually
use the brainstorming agent to do it,
come back and tell me one of the
craziest things you came up with that
you never thought you would come up with
before. I would love to hear it. Maybe
also jump into the Discord and and talk
about your brainstorming adventures. So,
as a matter of fact, some people in the
Discord actually talked about spending
hours just doing brainstorming and
coming up with some amazing ideas. We
sharted our documents. We have our
architecture. We have our PRD. So, that
is great. Now, I'm going to do something
here. I'm going to create a new folder
that I'm just going to call for now
temporarily.
And I'm going to say get init. put
everything in there except for the
architecture in the PRD sharded
documents. And now I'm going to add agit
ignore. So I'm going to say new
file.getit ignore.
I've added agit ignore file. Please
populate it with the common things that
should be in a git ignore and also
include the ignore folder in the git
ignore and the.getit get ignore those
files should also generally be ignored
by clouds so it won't clutter the
context. So there we go. So we just
populated a
get ignore with a lot of common
properties. Now if you're using a
starter template or a starter project,
it's probably going to set up your git
ignore for you. Or if you're using like
Nex.js, it'll it'll set that up for you.
In a complex project, you would want to
use the PO at this point.
And the PO has a command called run
checklist. And it's going to run the PO
checklist. What that does is it's going
to look at all of your user stories
fresh
and it's going to look at the
architecture,
make sure that they're aligned, make
sure that nothing came up in the
architecture that needs to feed back
into the stories because maybe something
changed or if there's any big gaps. It's
a really good idea to do that. Sometimes
it'll find things, but this is such a
simple app and like all of the BMED
method, it's adaptable. We do not have
to do that step. So, we're going to skip
that and we're going to jump right to
the scrum master.
Now, if you've never used the scrum
master, you might not know what it can
do. So, as always, I recommend doing
/help if it doesn't give you the help
commands automatically. Sometimes it
will. That's just the nature of LLM
sometimes. Um, so we have a few things
here. We can draft, which means create
the next story. We can do correct
course. This is a feature that a lot of
people have asked for and don't know
exists. Why does my chair keep sinking?
Um, correct course is something let's
say you've developed, you know, halfway
through some of your epics. You're mid
project in the middle of a story and you
realize you forgot something or you want
to make a big change. Wrap up what
you're doing and then talk to the scrum
master or the PM. You can choose either
one and run this correct course command.
It's going to ask you a bunch of
questions and it's going to analyze how
far you've gone and whether it's better
to, you know, revert back to a certain
earlier stage and produce new stories
and epics or maybe generate new future
stories and epics from where you're at
right now. Or your change might be so
drastic that it just recommends,
although it generally shouldn't, that
you start over. But not only is it going
to give you these suggestions, it is
going to help you. It's going to figure
out what epics and stories need to be
modified or changed or added or removed
so you can do this pivot, account for
it, and still get to your end goal.
Maybe it means making updates to the
architecture or as I said the PRD or
maybe you found a new API or a new
library that you want to do use, so it
might put it in the tech stack. All of
that's a possibility with the correct
course command. That could be a
lifesaver depending on the situation. Of
course, it's best to avoid that which is
why you go through a lot of planning.
What we're looking for here is number
one, we want to draft the first story.
Now, if you know what story it's time to
draft, just tell it. We know we want to
do epic one, story one. So, what we can
do is we can just say we can say we can
say star draft 1.1.
So, let's say we're halfway through and
it's time to do story 2.3 because story
2.2 finished. You can just tell it draft
the next story and it will figure out
what the next story is. Or you can tell
it which one to do. That'll just make it
a little bit more streamlined. It'll go
faster and it'll start drafting drafting
the story that it needs to. When it
drafts a story, it's always going to
check if there was a previous story. if
there was, it will also check if there's
any notes in that story that it needs to
carry forward into the new one. A lot of
times also these project foundations
will have steps in them. So, you really
want to read these stories especially
and see if there's any human basically
things that you need to do. Maybe you
need to set up an account or maybe you
need to, you know, visit a website and
provision some infrastructure that for
whatever reason cannot be done by the
LLM through a command to the remote
service. Maybe you have to get out a
credit card and pay for something. Okay,
so it's done. Let's look at our story
and let's see if it followed the proper
template. And it's always going to put
it in draft mode. This is standard agile
story practice. So as a developer, I
want to establish the basic noode.js JS
TypeScript project structure so that I
have a solid foundation for CLI
development. Acceptance criteria means
what does it actually take to say yes
this story is done and complete. The
tasks and the subtasks are going to
sound a little bit similar to this a lot
of times and it should because these are
the actual steps that the developer
agent is going to follow step by step to
implement our story. This is where the
scrum master agent goes through all of
the architecture documents, finds
information that is relevant to the
story and gives it to it here. Since
we're doing project setup, the agent
needs to know the structure of our
project that the architect came up with.
So that's why it pasted it here. So the
developer agent has this contextual
information and doesn't have to search
for itself bloating its own context. So
there are placeholders where the dev
agent can put notes and also where the Q
agent QA agent can put their notes.
Change this to approved.
We are done with the scrum master. We're
just going to close our context and
we're going to start a new one because
as always between every step even though
we don't have to, it's better practice
and it minimizes the overall context.
All right. And now we're here in the
home stretch. We got James, our full
stack developer is locked and loaded.
Let's do star help. Now you can just
tell James develop a story and it will
find the highest numbered story that is
set to approved or already in progress
that maybe wasn't finished by an
instance of himself from before. Or you
can just tell him. So why don't we just
tell him then he's not having to do the
searching. It will just be more uh more
streamlined.
So we'll say star
develop
story.
We could have also just told it develop
story. We could also just say number
five. Regardless, I'm just going to hit
space. And then I'm going to drag in the
file. And now we should finally see our
developer implementing story one. I
figure let's just do the first story and
then we'll commit. But we're just going
to say yes, let it do whatever it wants
to do. It's setting up our package.json.
JSON. And one thing we can do is as it
adds the different packages that we need
as it's going through the different
subtasks. We'll make sure that this is
aligned. By the way, we could turn on
the unsafe mode for cloud and it would
just churn through. But for right now, I
just have it. I'm sitting here. We're
doing this together. I don't mind just
telling it yes, continue. If you want to
roll the dice, you can turn off all the
safety features and let Claude go wild.
And look, we're uh we're almost done. It
it created a bunch of its own internal
to-dos mapped to the actual to-dos in
the story.
And then we should see that it will
actually go through. Now it's starting
to check them off. We'll set the status
to ready for review. So when it does
that, you have two options. I actually
recommend doing both, but it depends on
the story. If it's a simple story,
probably not necessary, but review it
yourself. Maybe test out the
functionality manually. Check it out. I
would start a new context yet again. So,
we're just going to exit. We're going to
say Claude. And now we're going to use
our QA agent, which is quality
assurance. So, let's do star help on
Quinn. Quinn the QA agent. So, let's do
number one.
And let's pull in
story number one.
Now, again, this is a fresh, brand new
context. So, it's going to be looking at
the story and all of the project source
code and other updates that it made to
really make sure that it did a good job.
So, this is going to this is going to do
a pretty good deep examination. This is
another place where you might want to
use Opus. Maybe even more importantly
than when we were doing the development
because this is the critical kind of
piece that makes sure that the agent
didn't go off the rails and stick files
in a stupid place. Um, and it looks like
it's already uh finding some things.
This was obviously a very small setup
story. So there's not a lot for the QA
to do for the QA to do, but it did find
a few things and it gave us uh these
compliance checks that it did here or
improvements that it made. So this is
great, right? I just want to say thank
you again. If you stuck all the way
through this video and made it through
here, you're going to be so far ahead of
the game. You're going to be a power
user using the BMED and understanding
the BMAD method, especially in claude
code. I welcome you to come into the
Discord forums and give me your shared
experience of how it's working for you.
Share with the community or if you have
questions still ask the community and
they'll be happy to help you. I'm sure
also it's a great it's a great place.
We'll see you next time. My name is
Brian. This has been a very long episode
on the BMAD code, but I hope it was
helpful and also worth waiting for. So,
thank you everybody and we'll see you
next time.
Loading video analysis...