ALL ROADS LEAD to AI CODING: Cursor, Aider in the browser, Multi file Prompting
Summary
TLDRThis video script showcases the transformative impact of AI coding assistants on software development. It demonstrates how a single prompt can generate significant codebases, highlighting the productivity gains and efficiency of tools like Cursor and AER. The script illustrates multi-file editing, context-aware coding, and the potential for AI to handle boilerplate, allowing developers to focus on higher-level tasks. The future of coding is painted as increasingly automated, with AI assistants writing a significant portion of the code, aiming towards full agentic tools that create value autonomously.
Takeaways
- π AI coding assistants can significantly boost productivity by generating large amounts of code from minimal prompts.
- π The script demonstrates an 8X multiplier in productivity by using AI to generate code, highlighting the efficiency gains in software development.
- π AI coding assistants like Cursor and AER are leading the way in AI-assisted coding, offering powerful features for developers.
- π» AER, a free open-source AI coding assistant, is praised for its explicitness and ease of use, especially with its new browser-based interface.
- π Multi-file editing is a standout feature of AI coding assistants, allowing simultaneous edits across multiple files, which is a significant advantage.
- π οΈ AI coding assistants can handle boilerplate code and minor adjustments, freeing developers to focus on more complex tasks.
- π AER's use of a REO map and its lightweight approach is contrasted with Cursor's deeper integration into the IDE, showing different strategies in AI coding assistance.
- π Cursor's blog hints at future developments like next action prediction and bug detection, showcasing the potential for even more advanced AI-assisted coding.
- π AER has achieved notable success in the S bench light, beating out competitors and demonstrating the effectiveness of its approach.
- π The future of engineering is envisioned with AI coding assistants writing a significant portion of code, potentially reaching up to 99% automation.
Q & A
What is the main focus of the video script?
-The main focus of the video script is to demonstrate the capabilities of AI coding assistants, particularly Cursor and AER, in generating code and enhancing productivity in software development.
How does the video script illustrate the power of AI coding assistants?
-The script illustrates the power of AI coding assistants by showing how a single prompt can generate a significant amount of code, such as 70 lines of code from a sentence and a half, and how these assistants can handle complex tasks like multifile edits and updating code across multiple files.
What is the significance of the 8X multiplier mentioned in the script?
-The 8X multiplier signifies the productivity gain achieved by using an AI coding assistant. It represents the ratio of the amount of code generated (208 words) to the number of words in the prompt (25 words), indicating an eightfold increase in productivity.
What is the role of AI coding assistants in handling boilerplate code?
-AI coding assistants help developers by handling boilerplate code, which are repetitive and standard parts of the code. This frees up developers to focus on more complex and creative aspects of coding, thereby improving productivity and reducing mental effort.
How does the script demonstrate multifile editing capabilities of AI coding assistants?
-The script demonstrates multifile editing by showing how AI coding assistants can make changes to multiple files simultaneously. For example, moving resource functions from one file to another and creating test functions in a separate file, all with a single command.
What is the significance of the browser-based editor for AER mentioned in the script?
-The browser-based editor for AER is significant because it provides a user-friendly interface that makes it easier to work with AI coding assistants. It allows for quick addition and removal of files and enhances the visibility and digestibility of the code editing process.
What are the future features hinted at by the Cursor team in the script?
-The future features hinted at by the Cursor team include next action prediction, which involves predicting the next steps a developer might take while coding, and bug detection, which could be triggered during different stages of the development process like commit, push, or PR.
How does the script highlight the importance of context in AI coding assistants?
-The script highlights the importance of context by discussing how AI coding assistants can utilize the context of the codebase to make more accurate predictions and edits. It also mentions the potential of using large context for better performance in tools like Cursor.
What is the significance of AER writing 7% of its own code as mentioned in the script?
-The significance of AER writing 7% of its own code is that it showcases the potential of AI coding assistants to not only assist in coding but also to generate parts of the code autonomously. This is seen as a step towards more advanced AI-driven development processes.
What is the ultimate goal for AI coding assistants as described in the script?
-The ultimate goal for AI coding assistants, as described in the script, is to achieve full agency in coding. This means developing tools that can create value autonomously, potentially reaching a point where they can write 99% of the code, allowing developers to focus on higher-level tasks and strategies.
Outlines
π€ AI Coding Assistants: Power and Productivity
The speaker introduces the transformative power of AI coding assistants, emphasizing their ability to generate large amounts of code from minimal prompts. They demonstrate this by creating a TypeScript application with documentation using just a sentence and a half, achieving a significant productivity multiplier. The speaker also highlights the importance of using AI coding tools to avoid wasting mental cycles on boilerplate code and instead focus on more complex tasks.
π§ Multi-File Editing and AI Assistant Features
The speaker showcases the multi-file editing capabilities of AI coding assistants, using AER (AI Engineering Assistant) as an example. They discuss the explicit nature of AER's interface and its ability to quickly edit multiple files in response to a single command. The speaker expresses enthusiasm for this feature and its potential to further streamline coding workflows, while also noting the desire for Cursor, another AI assistant, to implement similar functionality.
π The Future of AI Coding: Predictions and Developments
The speaker delves into the future of AI coding assistants, discussing the potential for next action prediction and automatic bug detection. They highlight Cursor's development roadmap and AER's achievements, such as their state-of-the-art performance on the S-bench light. The speaker also emphasizes the importance of optimal context in AI coding and the complementary nature of different AI tools, advocating for the use of large language models to handle complex coding tasks.
π Embracing AI in Coding: The Path to 99% Automation
The speaker concludes by discussing the end goal of AI coding assistants: achieving near-complete automation in coding tasks. They mention AER's self-written code percentage as a testament to the progress in this field and encourage engineers to embrace AI tools to maximize productivity. The speaker also hints at upcoming resources and courses to help engineers understand and leverage AI coding technology effectively.
Mindmap
Keywords
π‘AI coding assistants
π‘Productivity
π‘Boilerplate code
π‘Multifile edits
π‘CLI (Command Line Interface)
π‘AER (AI Engineering Assistant)
π‘Cursor
π‘Code generation
π‘REO map
π‘Agentic tools
π‘Bug detection
Highlights
AI coding assistants can generate 70 lines of code with a single command, significantly increasing productivity.
AI coding assistants can handle boilerplate code, freeing up mental cycles for more complex tasks.
The productivity gain from using AI coding assistants can be quantified, with an 8X multiplier demonstrated in the video.
AI coding assistants can update multiple files simultaneously, streamlining the coding process.
AER, a free open-source AI coding assistant, is highlighted for its browser-based features and ease of use.
AER's ability to handle multi-file edits is a significant advantage over other AI coding assistants.
Cursor, another AI coding assistant, is noted for its potential in next action prediction, enhancing coding flow.
Cursor's integration with IDEs provides a deep dive into coding assistance, potentially leading to higher productivity gains.
AER's lightweight approach and use of the REO map allow for a hands-off method in coding assistance.
AER has achieved a significant milestone by writing 7% of its own code, showcasing the potential of AI in self-improvement.
The future of engineering is predicted to involve AI coding assistants writing a much higher percentage of code, up to 50% or more.
AI coding assistants are positioned as essential tools for engineers to stay competitive and maximize productivity.
Cursor and AER are highlighted as the top AI coding assistants, with unique approaches to coding assistance.
The video discusses the importance of embracing AI coding tools to avoid being left behind in the engineering field.
AER's recent achievements in the S bench light are noted, showing its competitive edge over other AI tools.
The video emphasizes the need for engineers to focus on leveraging AI coding assistants to generate more value.
Upcoming features for AI coding assistants, such as bug detection and optimal context, are discussed as areas of future development.
The video concludes with a call to action for engineers to join the journey of AI coding and agentic engineering.
Transcripts
if you are not using AI coding
assistants to write code I'm going to be
real with you you are running out of
time watch this we're going to run a
single prompt to generate the beginning
of an entire code base
so with one command with a sentence and
a half we've generated 70 lines of code
and let's go ahead and run this to make
sure it's real right this is a bond
typescript application I asked for docs
I asked for Cle documentation all Ron
Bond index and you can see there we got
uses docs all Ron Bond index view you
can see this file doesn't exist let's go
ahead and make sure that it doesn't bomb
when the file doesn't exist so just
highlight we another prompt don't bomb
when file doesn't
exist let that run it's going to add the
existence check it's going to run
through every single line I'll accept
run this again no error this time let's
go ahead and add an item if we check
that API via our docs we can just say
add and then we pass in a link so let's
go ahead and add a link to add a new
resource imagine we're building at a you
know small tool that allows us to store
AI resources maybe we'll read them into
an llm ADD and then we'll pass in some
link we'll use the AER website URL we'll
hit add and then we've just added
resource with i1 let's go ahead and run
our view command again and you can see
we have AER in there let's go ahead and
add another link We'll add
bun and let's go ahead and add cursor
since we're at
it okay and then we'll run our view
command again and we now have our three
resources great so if you've used AI
coding assistant before you already know
what's going on here you already know
the power of AI coding assistance you
already know how much they can multiply
productivity in this video I want to
share and show off some of my favorite
AI coding assistant features I'm going
to give you a couple tips and tricks on
how you can get started and to progress
your AI coding abilities and then at the
end I want to look over some of the
incredible blogs that the cursor Bros
and AER are working on so AER shout out
and cursor shout out the NYS spere team
you guys are building great products
these two products are by far right now
the best AI coding assistants we've
talked about them on the channel before
it's important to bring up every now and
then because the space is rapidly
evolving let's go Ahad and just
backtrack a little bit right what just
happened so we ran two prompts generated
70 lines of code I just want to contrl Z
this a little bit and I want to look at
the actual multiplier that we just
unlocked by using an AI coding assistant
right I'll just control Z all the way
back I'm going to copy the prompt that
we use
and I'm just going to go ahead and open
up a node instance paste this in as a
string and type length so you can see
there that was 125 characters now let's
go ahead and split it and what we're
looking at here is the number of words
that we use to generate 25 words to
generate what what exactly did we
generate let's go ahe and look at that
first generation here you can see I have
a word count enabled here this 25w
prompt generated 208 words so if we take
that let's just go ah and do some
rounding right we'll round down so 200
over 25 that's that's an 8X multiplier
this is incredible right I just want to
really shout out how important this is
It's very rare that you end up in a
scenario where you can directly analyze
your productivity gains as an engineer
by using a tool I think that when you
look at a coding assistants and you look
at what you can do with them it's
blatantly obvious how much you can get
out of your AI coding tool right it's
just so easy to measure the productivity
gain um adex right we had another prompt
in there let's say we had a couple or
round it down right let's say it's only
a 5x that's still insane it's still
incredible and most importantly I think
one of the big highlights of a coding
assistance is you're not wasting your
mental Cycles on the boilerplate code
right you're just you're just not
wasting it let me go ahead and show off
another feature that I'm really excited
about with a coding assistant the
ability to update multiple files so
let's go ahead and switch over to AER
the best free open source AI coding
assistant what I'm going to do here is
create another file I'm going to call it
um I'm going to call it
C.S and I'm going to open up another
browser instance and run AER B this is
my AER Shand command to kick off one of
aer's brand new features you can now run
AER in the browser so this this wrapper
makes AER a lot more pleasant to work
with so let's go ahead and just look at
this let's go ahead and add some files
so I'll add um see if that CR file
didn't get picked up so let me go ahead
and just commit my code here so ad can
pick up that file go ahead and run that
refresh and I'll go ahead and add that
crud file and I'll add index.ts you can
see here with this nice clean UI it's
very visible it's very easy to digest
I've just added two files to the context
I like AER in that it's a lot more
explicit than cursor you really know you
really know exactly what AER is looking
at and then I'll just type move the
resource functions from indexs into
credts let it rip huge shout out to GPT
40 open AI really popped off with that
model it's been incredible to work with
look at how fast this is at gp4 level
with GPT 3.5 speeds uh pretty crazy
stuff let's go ahead and check out those
edits and so you can see here we now
have our crud TTS and we also have our
index.ts you can see here we're now
operating with just some 30 lines we
have that you know CLI functionality and
then we have our import for uh the crud
ttsa right let's go ahead and just run
this again make sure everything works
the same so bun index View and Bam so
you can see here we have our same files
everything's working as as previous
let's go ahead and remove um what is it
delete and we'll say two so resource ID
to deleted that should take out the bun
resource and if we run view again bam we
have one and three AER and cursor so
really incredible stuff right um
multifile edits is a huge huge Advantage
this is something that I would like to
see cursor bring to the table I know
they're working on a lot of really
important cool features but I think
multifile editing is a huge huge win uh
again big shout out to to ader and Paul
for having this feature and I've really
been enjoying using the browser editor
here for AER it just makes it a lot
easier cleaner to work with you can more
quickly add and remove files on the
channel we've been using AI coding
assistants for over a year now uh which
is really really crazy to say the Space
is really exciting this is the tool you
need to be paying attention to if you're
an engineer our productivity as
Engineers is directly tied to the
ability and the strength of our tools
you can be a shitty engineer with great
tools and get a lot done but if you're a
great engineer and you have shitty tools
you are limited by your tools you want
to be a great engineer and you want to
have great tools AI coding assistants AI
code editors these tools are where we
need to be focusing our time and
attention to right now there is no
bigger multiplier on your productivity
than to increase your ability to
generate more code in ranges in the 2 to
10x range you do not want to miss the AI
coding train and get left behind so
that's awesome uh we can push this even
further let me just show off a quick
example let's say we wanted to create a
test we'll call this cr.
test.ts and again uh for that multifile
capability what we're going to do here
is open up AER in the browser and we're
going to let me go ah and just commit
this again and what we'll do is refresh
crud CR test index and I'll do the
following right I'll say add a new
function delete all and add bun tests
for the crud module so I'm asking to do
two things in one here usually you'd
want to go one at a time add this delete
all and then afterward add BUN test but
I'm doing it together uh just because I
know that this will work so I'm going to
fire that off and AER is going to start
making changes to multiple files getting
all this work done for us we type one
sentence it's writing a bunch of code in
the right places for us automatic Al
it's so cool to see this come out and
you can see here edits applied so it
looks like we're not actually using
delete all resources so I'll just go
ahead and say I'll hop back over to AER
and just say uh create a CLI command to
run delete all resources let that run
and so while okay so nice it it added
that so quickly so I notice here it
didn't add that delete all would just go
ahead and highlight over this you can
see we have cop Plus+ doing some work
for us uh Delete all it's adding that
into the loging for us I'll just hit tab
that's awesome that's complete um let's
go and take a look at the test you can
see here we're using just I don't want
to use just here so let's go ahead and
just use cursor for this I'll just say
remove just and I'll also say um don't
mock anything I'm fine with this just
running against uh you know my local
file
system just going to make all the
changes for us here uh we're going to
ignore this we do have bun automatically
importing this is just a problem with my
lunter that I don't care about and we
can run BUN test you can see everything
completed properly let's go ahead and
bomb a test on purpose so let's uh let's
change this to a five just to make sure
that this is real and Bam you can see
there we got one F test revert and BUN
test good so we're all good we uh wrote
test we modularized our crowd
functionality we added a new delete all
resources function and we exposed that
via our CLI application all of this
happened in a hilariously small amount
of time we are rearranging code at light
speeds thanks to our two AI coding
assistants I'm going to go ahead and say
it these are the two best AI coding
assistants right now a lot of the AI
engineering the AI coding and the
agentic engineering content that we
share on the channel gets lost in time
to the YouTube algorithm I mentioned
this in the previous video but I'm
working on a resource a website that
that will be free that will allow us to
you know kind of build up this
repository of incredible AI tools AI
resources and of course you know some of
the most important content on there is
going to be AI coding because that is
the first really big use case for
engineering productivity gains in the
age of llms in the age of AI so those
are a couple features I wanted to call
out so I hope you see the value in this
if you've been using AI coding
assistants AI code editors AI pair
programming everyone has their slightly
own different terminology for for it in
my mind it's all the same these are AI
coding assistants they help you the
engineer write more code faster than
ever if you're using these yourself or
you're going to start thanks to this
video hit the like hit the sub it helps
other engineers get exposure to these
incredible tools as well one last thing
I want to do in this video I want to
look at where the cursor Bros and where
AER are kind of hinting at what's coming
next right so let's go ah and look at
this previous blog that came out from
the cursor team and uh they're talking
about the problems that they're you know
aiming to solve in 2024 2025 right and I
think it really shows where their heads
at and what they see coming next I love
this idea of next action prediction this
is when you're in cursor you're writing
code and it's basically copi Plus+ it's
this idea that once you start coding
something and you start making tweaks
there's a flow to it right there's a
path there's a pattern to it and just by
hitting tab just by knowing the most
likely case the most likely position of
the cursor you can essentially guess
what we want to do next right and let me
show you a concrete example of this if I
started making a change here right let's
go ahead and say I wanted to suffix this
right and I said read resources method
for whatever reason cursor in copat
Plus+ will start looking ahead and
Gathering context about what's the next
move he would likely want to make given
that change so if I just highlight the
next method and hit tab it's starting to
pick up on the trend right method method
and now it's updating the method call
inside here right and now it's looking
at the next method and now it's autoc
comp completing both the function name
and the caller right so you can see how
incredible this is right I'm not
thinking I'm not wasting mental Cycles
on this stuff my AI coding assistant is
doing it for me right and that's the
incredible part your AI coding assistant
right now can handle all the boiler
plate it can handle all the you know
weird small name changes variable
placements all the stuff like that
that's you should be really confident in
using AI coding assistance right so I
just want to shout that out um again I
want to shout out the uh you know cursor
team uh great work on all this stuff
please keep pushing this this is
incredible it's been great using that so
this is really cool I'm excited to where
they take this I do think this is the
right question to ask can we take this
idea to its natural limit there is no
reason why a natural logical flow of tab
completion like this cannot be solved
completely and I think you know C Is
Right on track to do that because they
have the full editor at their disposal
pretty incredible stuff there they talk
about some of the the ways they're doing
that um I'll leave that up to you to
read I'll link these in the description
of course um perfect edits this is
something that I think will come and get
kind of close My Philosophy around you
know models and Ai and predictions
you'll never get 100% but you should try
to get as close as possible I think the
real trick here is can they make it fast
and keep it accurate right um they
mentioned multifile edits here this is a
great call out I'm glad that they are
you know taking note of this I'm sure
that they you know pick this up from
ader AER does support out of the box uh
multi file edits which is amazing you
saw us we did a couple of those here we
have three files that edited all three
in one shot um so that was pretty
incredible it would be awesome to have
that rolled into cursor as well it's
great to see them looking at that
optimal context if you're a fan of the
channel you know that I am very anti-
rag I don't think that retrieval
augmented generation is worth anyone's
time
outside of like Brute Force pulling in
relevant files and throwing it into the
context I think that this is betting
against uh the development of the next
incredible generation models which is a
dumb thing to bet against take whatever
context you need throw it at the model
and let the model handle the hard part
of it right and just make sure that your
prompt is solid I think that tools like
cursor do have a slightly larger use
case for a large context and for some
type of rag based solution I hope that
uh you know both both them but also
everyone building out these uh agenc
pipelines I hope that you don't get too
lost in this sauce context windows will
go up needle in the hay stack uh
accuracy will continue to increase uh
one last thing here on cursor's blog and
we'll hop over to a Blog from ader that
I just want to highlight um they're
looking at bug detection I think this is
a really cool idea I think the question
is when do they fire this off do they
fire it off while you're writing code do
they fire it off you know during the
commit process during the pr process
during the push process lot of
interesting questions there for
automatic bug detection I think this
could be a really really killer use case
since cursor again is operating right in
an IDE they have a lot of power and
control um it's a massive Advantage
whereas something like ader is a lot
more lightweight it's in the terminal it
can read your code base it doesn't it
uses something called a REO map and I
think it's a a nice kind of lightweight
hands-off approach I like how these
tools are different I think they
actually complement each other I think
that cursor is like your Deep dive
coating and face whereas AER doesn't
want to make too many assumptions about
your code base and it kind of stays on
that like lighter wrapper layer we'll
see how this progresses let's go ahead
and just look at a quick blog from ader
AER has taken the state-ofthe-art on the
S bench light um I think this is really
really cool it beat out um I think open
Devon and uh what was the other one yeah
open Devon beat out Amazon's agent which
is really epic um and I love this you
know just like we were talking about AER
aims to be interactive not fully agentic
this is a differentiated approach that
AER is taking cursor is going a little
deeper I do think that this like serves
as a great differentiator for both
cursor and AER being agentic versus
staying lightweight just use the prompt
put it on the code base and uh cuz
things get really really complicated
once you start uh digging your hands
into you know making assumptions about
uh the best uh agentic flows and the
best agentic pipelines to set up I love
this shout out again from Paul and ader
ader does not use rag Vector search or
you know tools or anything like that
right and I do think again this is an
advantage it might seem like a
disadvantage I think over the long term
we'll see other tools fill that space
but the fact that ader is such a
lightweight wrapper that can still do so
much I think speaks a lot to its design
but also you know the power of uh large
language models and I do think a tool
like AER will benefit the most from the
continued improvements from Models
because it's not digging its hands too
deeply into the you know the agentic of
it all so that's awesome to see um and I
think this is the coolest thing so I I
want to end on this note AER has written
7% of its own code um when I saw this I
got so excited because to me this is the
future of engineering I'm working on
several code bases right now where this
number is much much higher this is more
close to 50% now I I'm I'm putting out
so much more content so much more value
as an engineer by handing off the
writing the code the individual lines to
my AA coding assistant and I think this
speaks to this idea of you want to be
upleveling yourself you want to be
telling your AI coding assistant your AI
pair programmer what to do and you let
it figure out how to do it llms are text
generation machines let them do that
that's what they know that's what they
can do code is a subdomain of text and
llms know text better than any human
being in the past ever you know on the
channel we are basically spending all of
our time to send this number to 99% it
might sound really crazy but this is
really what we're going for this is what
we're aiming for we want to take this
number and go to 99% let's just go ahead
and hop in here and just you know
filling the dream that we're aiming for
here right we want 99% the end game
really is full agentic tools
applications workflows that create value
on our behalf while we sleep I've said
it on the channel over and over and I
will continue to say it because that is
our mission that's what we're doing
that's what we're aiming for and AI cing
assistants like cursor and AER you know
big shout out to them not sponsored by
either of these guys yet uh definitely
would would love to be would love to
work with either cursor and or AER
there's a lot happening here it's
progressing and I think this is a really
important tool if you're an engineer
right now you do not want to get left
behind you want to be on the AI coding
train you want to be using these these
tools to generate more value than ever
before if you want to stay plugged in
and you want to stay close to AI coding
and agentic Engineering hit the like hit
the sub join the journey as I mentioned
I'm working really hard on a resource
that will take everything that we do on
the channel and make it more accessible
more usable with concrete guides and
some really interesting AI coding
related courses that you're going to be
able to use sign up for and really
understand how to maximize this
technology to create value for yourself
so if you're interested in that you know
what to do thanks so much for watching
I'll see you in the next one
Browse More Related Video
AWS Summit Sydney 2024: Build self-healing code with generative AI on AWS
Will AI βeat softwareβ β and whatβll happen to coders? w/ GitHub CEO Thomas Dohmke
GitHub's Devin Competitor, Sam Altman Talks GPT-5 and AGI, Amazon Q, Rabbit R1 Hacked (AI News)
Intro to GitHub Copilot in Visual Studio
7 Cursor Hacks to become the FASTEST coder
Is Coding Still Worth Learning in 2024?
5.0 / 5 (0 votes)