Augment Agent: The first AI Agent built for professional software engineers and large codebases

--- type: transcript date: 2025-04-03 youtube_id: YPY8zQ8bYAY --- # Transcript: Augment Agent: The first AI Agent built for professional software engineers and large codebases in this video I'm going to be showing you augment agent which is one of the most advanced coding agents that is built for real world engineering problems and not just toy examples first I want to go over some highlights from their blog post and then I'm going to dive in and show you some examples one of the key differentiators that I notice with augment when compared to other tools like cursor or wind surf is its understanding of context especially for particularly large repos and what they built is what they call a context engine so this allows it to deliver the right context to every AI interaction so whether you're chatting with the interface or if you're using autocomplete or now if you're using the agent it's going to be able to reliably pull the different pieces of your code base at the relevant time the way that they do this is they have a proprietary method where they're effectively chunking the relevant pieces of code to one another there is that understanding as well as that retrieval that happens very quick with each interaction that you have with an augment and leading up to the release of this agent is they released an open source agent with the S bench verified Benchmark and by combining claw 3.7 as well as 01 they were able to rank the top verified score that's just to give you an idea on some of the expertise that's within the augment team paired with the agent release they're also introducing memories which automatically update your work with the agent and will persist across the conversations to constantly improve the code generated solve task faster and match the code style and pattern that you have within your project in addition to the agent they've also fully embraced red mCP you're going to be able to have access to the wide range of tools and systems that exist and are increasingly cropping up within the mCP ecosystem and also built in they have native tools for GitHub linear notion Confluence as well as Jura the really cool way that you can think about this is now within your editor you're going to be able to create branches commit codes open PRS generate issues but also within tools like linear or jira or whatever platform you might be using is you can read issues and also even implement the solutions and update the issue status what you'll be able to do is ask for the agent to actually move those different tasks that you have within backlog and go through the workflow starting to close the gap of the software development life cycle where you have the backlog of items all the way through to actually pushing this code potentially into production and even managing some of your infrastructure on things like Cloud flare or verel now in terms of the agent there's a 200,000 token context window this will be able to pass in a ton of different contacts or all of those related pieces of code that it could be sending in as a part of its request in terms of some important features there's code checkpoints if at any point you do want to go to a previous step you'll be able to just reference up within the conversation to roll back those changes there's also native support for multimodality you can pass in things like screenshots or figma files to be able to work through bugs within the UI or actually build out new UI elements and then there's also terminal commands instead of just file search and encode the agent can actually run the commands within the terminal so it can install things start your server and interact with Git directly in addition to this there is also an auto mode if you don't want to accept every change that the agent is taking whether it's within the terminal or the different file edits you can go ahead and turn on auto mode where it will be able to agentically go through all of the different steps that it needs to take now just to quickly touch on some of the best practices that they mentioned for using AI coding agents there are some really interesting tips here in terms of how you interact with the model coding agents are a little bit different than something like say chat gbt where you're used to just iterating on a response back and forth especially if you're using it within something like auto mode some really good patterns in terms of how you can actually interact with the agent just some at a high level if you ask for it to find the relevant context but also mention to the agent particular keywords this could be file names function names variable names within your codebase and then give the agent more information about what you're trying to achieve the other thing to note is to provide the agent with positive feedback if the agent has provided the correct answer or the right context for whatever you're asking for providing that feedback and then providing additional instructions can be a helpful pattern and here is just a really quick example of that within practice one of the biggest things with the announcement is right now at time of recording their developer tier for the agent mode there's unlimited requests instead of within cursor or wind serve where you only have a set number of requests each month if you subscribe to that tier you have unlimited requests then need addition to that they do also have a community plan where you'll be able to access and try out the agent with up to 50 requests per month now without further Ado let's dive into some examples to get started with augment as well as their new agent it's super straightforward right now for the agent in particular it's supported in vs code as well as jet brains so you can either download or search for the extension within the marketplace once it's installed you'll have this chat panel that you can access and in addition to that you'll also see the augment next edit suggestions in the bottom that you can select if you'd like once it's updated you can go ahead and try out the new agent mode now if you're using this in a new project what augment will do at first is it will go and index your repository as a first step and that's an essential part for how augment is able to give really good answers as soon as you click Start agent it's effectively an introduction to the agents you see meet Auggie I'm an AI coding assistant I excel at understanding large complex code bases but I am happy to chip in on code bases of all sizes to demonstrate how it all ties together is it's able to get the context passing into the conversation we can see the name that it got as a result of that terminal command we see since I am an llm I don't have real memories I'll be using augment memories and then from there it gives you some instructions in terms of how you can commit things to memory for example you can say Commit This to Memory and then finally there's an example on how you can set up the Integrations within settings if I click over to settings here we see that I have unlimited agent requests and then for the tools we have the first class tools of GitHub linear notion Confluence as well as from here you can add in your mCP servers similar to how you would within something like wind serf or cursor if you've used either of those finally within the settings we have the context listed out for this project I have 42 different files we can see the structure and we can see all of the different pieces that it either did or didn't index we can see it's excluding certain things like node modules and some log files as well as environment variables some things that you might otherwise expect now within this update you can go and click over to the augment memories within the memories we have memories help me remember useful details for further interactions and this is going to be where it puts all of the different memories that we have for this particular projects we can go and reference particular things maybe conventions that we have or certain patterns that we follow now with augment you can select in three different modes you have chat agent as well as Auto agent which I'll demonstrate in just a moment you can also add an attachment or alternatively if you want to reference different files or folders you can do all of that within here I'm just going to run through a handful of examples within the auto agent mode and show you how this works now I'll just demonstrate a few quick features first I'm going to say I want to add in a navigation within my application have the title read developers digest and add in a few menu items make sure that everything's responsive then once that's complete let's go ahead and start our server what we'll see is similar to the examples that I showed earlier within the video is we'll see augments context engine going and finding the different pieces of our application as how it's structured here we can see it's reasoning through the different steps and if there are further files that it will have to read we'll see that it will list all of those out here here we're going to see that it's going to check if there are any UI components that we can use for the navigation menu and then finally we see that it's developed a plan for implementation So within here we see we're going to update the na. TSX we're going to change the title to developers digest we're going to add responsive menu items and then we're going to add in the necessary icons here we can see it went through it made that edit and then finally it ran the command to start our server here's the navigation it added in those navigation items it also has the correct context for the different pages that I have within here and then what you can do within here is you can go and you can keep all of the different changes that you made what I can do within here is I can bun create next app I'll go ahead and start a new project now when it is a new project or a repo that you're pulling into augmen you just have to make sure that you index that codebase as a first step as soon as it's done indexing we'll have a breakdown about the Rebo we have some suggestions of questions that we can ask but in this case what I'm going to say is I want to create a beautiful landing page for my brand developers digest I want to create a rich nav and footer and add in some content we go ahead and I'll kick that off it's going to gather the relevant information about our project without us having to reference and add different files it's able to do all of that for us once it has all of those different pieces of context that it needs going to go and just generate a detailed plan to create what we have asked for the other nice thing with this is you can see all of the edits within the chat panel as it goes through if I just expand this here we can see within the layout it went and made that edit within the global. CSS we can see all of those edits as well here's what it generated for us we have developers digest and overall we have a really nice structure for a starting point for a potential website we have this really rich footer that it generated for us as well as all of these relevant details from there we can begin to iterate on this I encourage you to try episode if you've tried augment code let me know your thoughts in the comments below otherwise if you found this video useful please comment share and subscribe otherwise until the next one