AI Coding Tool Breakdown: AI Copilots vs AI Coding Assistants vs AI Software Engineers

ai co-pilots ai coding assistants and AI software Engineers what's the difference what are the capabilities and what tools should you the engineer be focusing on to write high quality code at insane rates we are in the age of generative Ai and if you don't want to get left behind while your peers supercharge their productivity with the current and Next Generation AI coding tools you need to be paying attention to these three categories of of tools this is a full breakdown let's get into it so the first thing we need to do is differentiate these tools at each level we increase the complexity and the AI autonomy of the tool so at the bottom level we have our AI co-pilots also known as AI autocompletes at the second level we have ai coding assistants AI coding assistants are a superet of the capabilities of AI co-pilots so everything that you can do with an AI copilot your AI coding assistant can do or doesn't need to do because it has has Superior capabilities when we go one level higher to the top we have ai software Engineers you've heard of these but this is how you can break down the relationship between all these tools at the bottom we have co-pilots then we have ai coding assistants and then we have ai software Engineers so let's look at some specific tool examples right you've likely heard or use one or many of these tools right so these are AI coding tools get up co-pilot we have super Maven and then we have cursors copilot Plus+ at the next level we have our AI coding assistants these are tools like cursor AER and continue and then at the highest level we have our AI software Engineers Devon here we have open Devon and we have co-pilot workspace we need to be moving our technology stack into the future by using tools powered by llm technology here's one more way to look at this at each level we increase AI autonomy and we decrease the amount of work and effort we need to do as the engineers to generate value to generate value with software so at the bottom level we have our tier one these are our AI co-pilots they boost our productivity by a ton not as much as the next tier right tier two we have our AI coding assistance the capabilities in our AI coding assistance supercedes the co-pilots by a massive multiplier and we'll look at why in just a second by digging into the key capabilities and features of AI coding assistance but then at the highest level we of course have our AI Engineers when these really get to Market and get to scale and are usable and affordable they will essentially replace all of the capabilities of coding assistants and co-pilots effec ly making them obsolete as they improve I think the right way to utilize these tools is when you need to dive deeper you remove the highest layer right so there's going to be a lot of boilerplate work that your AI engineer can do for you but when you need to dive in and get some more detailed coding work done you'll use your coding assistant and same thing right now right if you're using coding assistant a lot right now it's not like you're not using tab AI autocompletion with your co-pilot it's just that you're likely doing it less and you're not relying on it so much because the AI coding assistant can do so much of what the co-pilot can do I want to be super clear here there are levels and kind of sub tiers in between these tools this is just a great way to break down and differentiate co-pilots versus coding assistants versus full-on AI Engineers right there are levels in between I'm sure you can point out examples that don't quite fit these categories this is just a good way to break it down to add Clarity to your tooling so that you can know what you need to grab and so that you can know what type of technology is emerging in each one of these categories so these are our three tiers let's go ahead and dive into specific examples of what these look like so let's break down some of these features right so let's start with ai co-pilots ai co-pilots have two key features really one key feature they have ai autocomplete let's dive back into the code base we were working on in the previous video feel free to check that out we went into more details on AI code prompting you can think of this video as a part two we're going to talk more about the highlevel capabilities of each one of these tools so let's start with the AI autocomplete most of us started out and get up co-pilot and then we migr it to something a little more advanced I am using cursors co-pilot Plus+ let's go ahead and create an add many resource function okay so I'm just going to start typing add many resources and as you can see it's autocom completing the rest of the capabilities for me comment autocomplete this is very similar but what we do is we trigger the functionality we want to write by writing what is effectively a comment prompt so go ahead and type another prompt here I'll say modify resource hit enter here and I'll just type e copilot will come in auto complete the functionality for us and I'll just hit tab that's exactly what we want we're passing in an ID and a new link and this will modify an existing resource by matching it by ID That's The Power of AI co-pilots also known as AI autocompletes right so really powerful stuff there let's go ahead and check that off and move on to the most important AI coding tool right now the AI coding assistant in this category we have cursor we have AER we have continue definitely check out the previous video we went into more in depth into examples of how to prompt we basically build this entire code base with a few prompts let's start with the selection prompt so we're going to highlight a block of code we're going to command K here add IDs as second param going to hit enter and it's going to infer the change I want there so instead of just passing in our links which is just a list of strings we're also going to pass in our IDs and make sure that the ID doesn't already exist we'll push our resources and then we'll go ahead and write them so this is one of the key features of AI coding assistence it gives you a lot more control and power versus just autoc comp completing with the surrounding contexts that AI co-pilots have right so this is really powerful let's look at the selection prompt with context so this is really cool let's actually go ahead and modify our readme and what I'm going to do here is we have our CLI application again we built this out in the previous video it has four four five CLI commands I want to create some usage documentation in the readme so I'm going to prompt this I'm not going to do this myself cuz why would I waste that time I'll start a new selection prompt on line three I'll hit at and then I'll reference my index.ts file so now I have the context of this file inside this prompt and then I'll explicitly say I'm going to ask for some markdown I want a how to use and that's it I'm just going to hit enter I'm giving it a header and a context file and it's going to know that it's going to read me and it probably wants usage doc mation you can see it's autoc completing all of the existing functions and from the index.ts file um I'm going to go ahead and create a followup instruction we are using bun so I just want to say use bun as a follow-up and it will now go in and use bun syntax and reference the TTS files perfect this is great right now we have documentation we just ran one prompt and we gave it some context really powerful feature of a coding assistance let's go ahead and check this one off and let's look at the terminal prompt and what we're actually going to do here is we'll look at the terminal prompt with context right we'll just continue building off the work that we just did let's open up the terminal inside the terminal we'll hit command K and now we have a prompt inside the terminal and what I'll do is I'll say how do I add a resource and then I'll reference our readme file which now has our documentation on how exactly to um you know add modify and remove based on our index ATS file right so just hit enter here and it will generate the command for me automatically right so it has this written perfectly I'll just hit run and you can see there it ran that code it added the resource with id5 if we go ahead and crank open our AI resources you can see that file has been added so this is awesome right we were able to create documentation using AI coding we were able to hop in the terminal and just ask how do we do something right how do we do this it can help you remember arbitrary bash commands or G commands we also have two chat prompts I'm going to go ahead and add with context here looks like that was just missing so we looked at file prompting and multi file prompting in our previous video I'm going to go ahead and defer you to that I'm going to link that in the description where we ran an entire file and multi prompt using ater I'm going to say incursor in this video and just kind of run through the high level of these key features so the file prompt is effectively a selection prompt of the entire file and then of course the multi-file prompt would be a selection prompt of multiple files right in cursor you can't do that in ad you can let's talk about the chat prompt this is really important so in cursor and in continue and I think even GitHub co-pilot has this where you can with a model with context of your code base right so let's just go ahead and ask we have the read me open right now I'll just say how to delete resource right you can see here it gave us a clear examples you replace it with exactly what you we're looking for I'll just go ahead and highlight this and run this all delete item one you can see Resource One deleted that was the correct command and if we look at our AI resources we have no id1 anymore right so that worked great let's go ahead and run something a little more intense I'll create a new chat with some context right and what I'll do here is I'll use this really incredible feature you can hold command and hit enter and it'll use the context of your entire code base so I'm just going to ask at a high level how does this work if you're an engineer who works on three or more codebases on a weekly basis having the ability to use codebase QA question answer uh is especially useful when you're jumping back into code bases and you just need a quick refresher on how things work all I'm going to type here is how does this codebase work where should I start right just something like that right and I'll hit command enter this time and you can see here it's reading multiple files and you can see exactly what it's what what what it does right so it's going to give us a nice breakdown it's telling us the main kind of highle components the main functionality index crud crud test telling us the starting point and then it's probably going to give us a command here let's go ahead and see if we can get something yep nice so we're going to get some commands which is awesome to see uh looks like the formatting got a little weird here but this still looks great right again if you have clients that you work with and you're just hopping around between multiple projects this is so useful for just redownloading the codebase back into your personal Ram your mental memory so that's the chat prompt and that's the codebase question answer the chat prompt with context is basically the same thing as code-based question answer you can you know refer to a specific file so I can say do we have test coverage for every method in crud dots and then I'll go ahead and add the context of cr. TS and I'll also have the context of crud test right so I'm just going to go ahead and ask that question there and looks like like we're missing a couple items add many modify and view don't have test coverage right so this is really cool you can add context and then ask questions against your context so that's chat prompt with specific contexts last thing I want to show off you can do web search what's something that we might want to do let's look at something like using the Shell command with this cool new dollar sign syntax we'll ask the cursor chat to fill in some information for us right so let's just ask some questions against this I'm going to open up a new chat here and then I'm just going to type um I'm just going to paste this URL in now it can read this documentation and I'll just say how do I run bash commands from a bun script and we'll just let that fire you can see there is using the currently open file but it's also going to load in this web page and you can see it's breaking down this dollar sign command and you can see just like that uh this is probably coming right from the docs as well right yeah exactly we have this hello world that's perfect and then it's just walk us through how to await results from bash um you know huge shout out to the bun developers this is so cool I've been really enjoying using bun it's making JavaScript and typescript and the entire ecosystem so much easier to work with so anyway shout out bun this is an example of AI coding assistants web search so that's a lot of the key features a lot of the key value propositions from AI coding assistants you can see there's a lot more going on here than there is with the um AI Auto completes where essentially it's just really really smart autocomplete again it's not like we don't need that it's not like it's not useful it is just completely superseded by the functionality from AI coding assistants um there's not really a lot to talk about with AI software Engineers yet this technology is still super gated and it's really immature I have tried open Devon there is potential there but it is way too early to recommend uh AI software Engineers to anyone right now there's just too many costs associated it's very unreliable it takes too much time so we're going to hold off on covering a software Engineers but really the promise is just one thing right it's end to end task prompting you give it a problem you give it code you give it the tools and it does everything for you this is really important we're going to be covering on this channel subscribe if you're interested in seeing coverage on this but um it's just too soon it's just too soon to really cover these tools this has been a high level coverage of the key features specifically with AI coding assistance that that's where all the value is right now and let me just go ahead and say a couple things about you know recommendations you know my recommendation is you want to have a AI Cod pilot running at all times but then you want to be doing most of your work using an AI coding assistant most of coding is writing boiler plate setting up types setting up endpoints it's a bunch of you know wrapper logic a lot of surrounding code and then you have the like 10% the 5% of code where you need to really put your engineering hat on you know whip out the Whiteboard really get into the technicals of it that's incredible that's awesome at the end of the day though your AI coding assistant is going to be able to hop in and complete even the very very complex code that you have that you're writing that you know 10 5% as long as you write about it clearly as long as you prompt it clearly you're going to be able to get all that value out so I recommend that you always have an AI co-pilot running and you're always using an AI coding assistant there's just no argument against these tools being useful anymore maybe a year ago you could have argued that I disagree with that even still but that that's irrelevant right with gbg 40 Cloud 3 Opus there's just no excuse to not be using an AI coding assistant anymore um you are just burning your own time you're burning your own productivity you don't want to fall behind the AI coding train so on the channel we focus a lot on using the best tool for the job like I mentioned in the previous video AI coding assistants are the best tool for writing code right now so I recommend you jump in start getting used to any one of these tool tools and just building on top of it because what comes next will be mind shattering if you cannot handle the transition to Ai co-pilots and AI coding assistants when AI software Engineers hit it will just break your brain it will make you feel like um it's over you've heard the whole Narrative of coding jobs are going to disappear there is some truth to that frankly but it's not going to disappear in the way you think it's going to disappear it's Engineers like myself it's a lot of the engineers that watch the channel we are using these tools these AI coding tools won't replace you we will right we are using these tools we are getting supercharged if you're not using these tools we will replace you because we'll be two 3 four five 10 times more productive than you because we'll be using these tools so you know join the journey hit the Subscribe button hit the like button um we've got a lot more AI coding content coming as well as agentic workflow a agent and a lot of llm powered automation tools tricks and guides coming like I mentioned I'm working on a really really cool resource that we can use to track a lot of the interesting things a lot of the interesting topics that we cover on the channel in a concrete website I'm really excited for that when I'm not filming videos planning doing research or doing my Consulting work I am spending all my free time working on this incredible resource I'm building for you so that we can evolve as engineers and so that we can become agentic engineers thanks so much for watching hit the like hit the sub and I will see you in the next one

Channel: IndyDevDan

Views: 3,632

Keywords: aider, ai copilot, ai autocomplete, ai coding assistant, ai software engineer, ai engineer, continue, github copilot, copilot workspace, devin, opendevin, supermaven, coding with ai, ai coding, cursor ai, aider ai, gpt-4o, llm coding, cursor, gpt4o, indydevdan

Id: 2j_fgMPJGM0

Length: 16min 1sec (961 seconds)

Published: Mon Jun 10 2024