Video22 min watch

NEW Aider Architect & Editor Updates Are INSANE!🤖(Beast Cursor?!?) Best AI Coding Agent?! OpenAI o1

In this video

  1. 00:00Introduction: Ader AI Pair Programmer Version Updates
  2. 00:30Claw Dev’s Fast Development Cycle & Reflexive Coding Loops
  3. 00:57Ader AI Update 0.5.8: Architect & Editor Features
  4. 01:33Comparing AI Model Performance: Gemini and Quinn 2.5
  5. 02:17Ader Writing 53% of Code in Version 0.5.8 Release
  6. 03:42Architect vs. Composer: New Model Configurations
  7. 04:25Quality-of-Life Improvements and New Features
  8. 05:04Model Performance Breakdown: GPT-4, O1, and Quinn 2.5
  9. 06:00Ader's Increasing Code Contribution and Release History
  10. 06:57Getting Started with Ader's Architect Mode: Installation & Setup
  11. 07:31Installing Ader in VS Code & Development Mode
  12. 08:08Architect Mode Overview: Requesting and Editing Files
  13. 09:29Setting Up API Keys for Ader
  14. 09:58Running Architect Mode with Different Models
  15. 10:47Example: Creating a Calendar App with Architect Mode
  16. 12:17Editing and Iterating on the Generated Code
  17. 13:04Calendar App Performance Review: Dark Mode & Bug Handling
  18. 14:48Ader vs Cursor: Dev Experience and Efficiency
  19. 15:29New Gemini Models: Improved Performance and Cost
  20. 15:54Gemini 1.5 Pro Features: Rate Limits and Latency Reduction
  21. 16:40Final Thoughts: Ader's Progress & Future Potential
  22. 17:23Outro: Join the Community and Subscribe for Daily AI Content

Links from the video

Transcript

3,274 words

What's going on everybody Josh ful here and in today's video we're going to be taking a look at some crazy new updates from ader the AI pair programmer last week they launched some really cool stuff with version 0.57 and then just a week later today they launched version 0.58 which has their new architect and editor feature I'm going to show you how you can use it today let's Dive Right [Music] In all right guys so recently I did a few videos

I did a video on Claude Deb's recent updates and I mentioned in there that the founder has been rolling out updates really really quickly and it's very impressive to see but also I mentioned in the video that damn he's probably writing claw Dev or building claw Dev and then using claw Dev to build claw Dev it's like a reflexive cycle and the input feedback loop of him being able to code iterate and improve is getting shorter and shorter and tightening up

But we can see the same thing with AER all right that's what's happening with these AI pair programmers so uh last week I was actually going to do a video on this update 0.57 and before I could even get to it they actually just dropped an even more impressive version which is 0.58 12 hours ago we're going to dive into this in just a second but before I do I want to go over to Twitter and take a look at some

Of the tweets here so uh new quen 2.5 models are on the AER leaderboard so I did a video on Quin 2.5 yesterday or no two days ago and um we'll take a look at how it Compares on their ater leaderboards all right so on September 21st you could see the version 0.57 dropped we'll go over the change log in just a second and then and then we can see he says that the new Gemini models perform basically the same as

The previous versions on AER code editing benchmarks differences seen within the margin of error and we can see the graph right here and then 12 hours will he tweets out saying AER version 0.58 is out with the new architect and editor coding configuration support for Gemini 002 better support for quen 2.5 and many Q features and Bug fixes the crazy part to see is that he says AER wrote 53% of the code in this release really really cool really impressive to

See and this is like I said this is what we're going to be seeing more and more happening this percentage increasing as these models get better and um yeah it's really uh requiring less you know manual human code being written in these releases and more just AI I mean but this is also pretty common we're seeing if you look at you know the activity at any single repo on GitHub and any single project over the last 2 3 years all right

So here's version 0.57 we'll not really go over this one too too much cuz it's kind of a lot of it is referenced in version um 58 but we'll go over some so you know support for the new 01 models and then you can now recursively add directories with read only numerous bug fixes and and I'd say this one's a smaller release opposed to something obviously like the 58 with artifact and editor but we can see here AER wrote 70% of

The code in this release right so it's just it's getting to the point where it's really just self-improving and uh well I mean it is but you know until it's just completely self-improving um anyways so we can see with the 58 update we got uh use the new architect and editor models for improve your coding use a strong reasoning model like 01 preview as your architect and use a cheaper faster model like GPT 40 as your editor all right my first

Thoughts on this um before even really trying it out is that this is like kind of similar in a way to composer but it's not at the same time it's a little bit different but I mean it has a similar type name a lot of these models are getting kind of going in that route where there's like an architect or composer or I think Claude dev has a name for one I forget but or they will if they don't um and

Then support for the Gemini models better support for quen 2.5 models many configuration questions have been can be skipped for the rest of the session with don't ask again responses and then autocomplete for readon support the enti supports the entire file system new settings for completion menu colors and then new copy command to copy the last llm response to your clipboard I mean that's pretty useful renamed clipboard to paste and then we'll now follow HTTP requests when scraping URLs so that's

Pretty impressive and then new voice format switch to send voice audio as wave MP3 or webm and then model settings takes extra prams dict to specify any extras to pass to light llm completion and then support for cursor shapes when in Vim mode numerous bug fixes and then you can see a graph here basically um around their architect model and we can see some of the top um models right here GPT 40 mini 01 mini um GPT 40 Claud Sonet 3.5

And then 01 preview and we can see this is with the editor or architect model and then the editor model and the edit uh format so we can see Baseline uh with GPT 4L mini gp24 mini hole and then 01 mini Baseline deep seek diff DPT 40 diff deep seek hole um Baseline deep C cool with gbg4 diff uh and then uh gbg4 diff and then Baseline with clonet deep sea coold deep seek div Claud Sonet div here and then same

Thing with 01 preview we got Baseline bt40 diff deep seek diff Claude Sonet diff and then deep seek cool so deep seek the new 2.5 which I did a video on and basically was saying in that video that you know it's one of the best if not the best currently right now for open source coding we can see here it being paired with some of these other models basically gives um within their test a lot of the time gives the best

Results and we'll dive into that in just a second and then also too I just want to highlight here's the release history here so percent of new code written by AER by uh by release we can see see that this not this has substantially increased as we can see here like 50% it's consistently started doing like 50 to 60 to 70% like 70% was the peak which was this last recent release at 57 uh so the the release before 58 and

Uh it's yeah it's been increasing and then lines of new code written by AER by release we can see here it's been increasing pretty much as well so pretty impressive to see all right so the way to start actually using architect is you're going to want to go to the AER sit and links for everything I showed in this video will be in the description down below want to go on installation go to optional steps and you will see there's like

Optional steps right we can see enable playright we can see enable voice coding right here the different commands for that add AER to your editor Neo vs code so they have a vs code extension too if you want to check that out but I think I don't know if how well that works um but here we're going to want to install the development version of Aer all right so you're going to run this Command right here all right so I'm in

Vs code in a terminal and I'm running python DM pip install D- upgrade get plus HTTP GitHub etc etc so we're installing building dependencies all right as that's installing we'll see here architect mode and editor mode when you are in architect mode AER sends your request to two models first it sends your request to the main main active model that main model is configured with for SL model-- model or the shortcut switches like D- son it after the main model replies

AER will offer to edit the files based on the response to edit the files AER sends a second llm request asking for specific editing instructions this request goes to the editor modeler has built in defaults to select an editor model based on your main model or you can choose an editor model yourself with the D- editor- model and then model command the editor diff and editor whole edit formats are recommended for use by the editor model architect mode produces better results

Than code mode but uses two llm requests this probably makes it slower and more expensive than using Code mode architect mode is especially useful with open ai's 01 models which are strong at reasoning but less capable at editing files pairing an 01 architect with an editor model like GPT 40 or Sonet will give you the best result results but architect mode is also quite helpful when you use GPT 40 or Sonet at both the architect and the editor allowing the model

To requests to solve the problem and edit the files usually provides a better result and I'll leave a link to that as well as this article going over the architect and editing mode and you can see here it says try it the development mode of Aer has built in default to support architect and editor coding with o1 preview o1 mini gbt 40 son it run the dash architect or get started quickly like this all right so you can start it in

A few couple different ways like shown here like this all right and then we can see the benchmarks for the different models with diff whole Baseline Etc here all right so if you haven't already exported or set your API Keys you can do so so you can either run export open router API key equals and then your key here if you're on Mac or Linux or set X open router API key uh and then your open AI key here if you're

On Windows now this goes for pretty much any mod you're using whether it's olama deep seek open AI anthropic um Gemini Etc reason I'm just saying open router is because if you want to use open AI 01 models which it says it is one of the best for like 01 preview for reasoning Etc then you're probably going to want to use open router unless you do have tier five unlocked on the open AI API console all right so to run the

Architect you can do AER d model and then your model right here and then D- architect flag and then D- editor- model if you want to select the editor model all right by default anthropic has some that it selects uh depending on your base model but uh then you can either use for me I'm going to use claa 3.5 so you can either use it through anthropic you could also use it through open router um they suggest sometimes just because anthropic

Sometimes gets rate limited but we're going to go ahead and run this okay so we can see here um our main model right here uh o preview editor model week model get repo etc etc right and I'll leave this uh command in the description down below and you can switch it out for whatever model you desire just so you know the right Flags to get started using this um this uh architect all right so I'm saying please build me a very

Modern calendar application using HTML CSS JS make it professional with some cool useful features we're going to go ahead and send that and I'm going to guess this is going to take a little while so sit back grab a coffee and let's wait okay so we can see here we got this output from the architect so to build a very modern um calendar app you know it's going through the different steps so step one project setup step two HTML structure styling

Uh CSS styling for step three JavaScript functionality um Advanced features optional testing deployment maintenance Etc so this is from 01 preview of course it's going to be pretty in-depth and descriptive and we can see that we have tokens 196 Cent tokens received 3.2k cost uh 20 cents per message 20 cents session so uh edit the files we can say yes or no if we want to continue so I'm going to say yes now I'm curious how much of that it's actually

Going to do that's quite a lot that it gave us um so okay so it's created wow okay that that was really quick obviously it was clots on it so we got the HTML the CSS here and the JS and we are going to say yes create those files yes create that file yes create this file so similar to something like how claw Dev Works how cursor works and now add JS to chat I mean sure okay so created the files

Here looks like trying to add it to chat and it's having a bit of an error let's go ahead and run this okay so here we have our calendar app and it's actually a okay I'm not going to lie this is a lot better than I expected especially on a onot go right usually when I do a calendar app first of all like usually when you do any app even if it's a very basic app you're going to have to go

Back and forth and just like maybe do some minor bug fixes which of course we would want to improve upon this if we were making this an actual calendar app but one thing to note just off the bat is that um most times when I do a calendar app it completely screws up in terms of the dates like it won't get the dates right so as you can see it's December here and it's you know Monday uh the 30th it's got

The dates correct right which should be something that's fairly simple for you know an advanced llm but believe me a lot of them get this wrong so out of the box we can see all the dates are working and it I've actually haven't really seen this too much work the first try especially with a basic prompt like that let's go ahead and to wow okay so toggling dark mode on even though this is a very basic like feature and setup like

Usually lots of times when I will toggle dark mode on it may have some issues with like The Styling maybe the the text is like black as well and it doesn't go right so right out of the box as much as this is just a very simp whole app with not M much uh functionality or features everything that it did add is correct and is working really really well so I'm actually very impressed um this is a oneshot prompt right so

And it actually didn't take as long as I expected I actually thought it was going to take a lot longer now in terms of like iterating upon this using it back and forth and using this on a day-to-day with the dev experience I think it uh it may be a little bit longer it may not be as nice maybe as something like cursor where it's a little bit more you know has a nice nicer Dev experience in my opinion personally but

Let me know what your thoughts are compared with this to Cur I definitely think ader is on the right track and I think they are um really releasing some cool features and if they continue increasing that percentage from 50% 60% 70% of new releases written by AER and uh I think it's going to be pretty impressive in the next few weeks next few months I mean even just what we're seeing right here I think that you know this improving upon itself

Is going to be pretty crazy making it really really strong now also too like I mentioned at the start of this video it does now have support for the Gemini models so or well the new Gemini models so um you can see I'm not going to cover this in depth here but Gemini 1.5 Pro um essentially long like pretty good long context uh uh you can see the input cost per million tokens and output has been slashed in half a which

Is pretty good um well effective October 1st so tomorrow we got increased rate limits two times faster output and three times less latency so you can check out these Gemini models as well um they're definitely good if you want it for basically very very cheap all right so let's say if we started up uh AER normal just with like a open um open AI 01 preview model alone and then we did for SL architect right here and we can say for/

Architect and then you can say a prompt here such as make aask app that is very modern and professional using HTML JS and CSS all right guys so all in all I think this release is pretty big for AER I think this is the next step in terms of the architect and editor now I don't think this is like the end all be all and um personally I don't think that I I'm still going to be using cursor composer and I

Still really like cursor but I think they're on the right track like I said and very excited to see what they have in store for the future but let me know what your thoughts are in the comments Down Below in terms of Aer versus maybe cursor in terms of using AER with the architect and editor do you think it's useful are you going to be using it have you been using it yet and what's your favorite AI pair programmer are you

On team AER claw Dev cursor let me know in the comments Down Below guys if you're new to this channel we upload videos every single day on AI automation business growth AI coding so if you like that type of content you got some value here make sure to smash that like button and subscribe to stay up to date with the daily uploads than that guys if you haven't already joined our free Discord and free Facebook Community go to stri community.com link

Will be in the description down below as well as all the different resources we went over in today's video other than that guys I will see you in tomorrow's video keep hustling keep grinding and of course guys accelerate your stride take care

Latest from the blog

View all