This OpenAI o1 pro coding workflow is insane...

[@realmckaywrigley]

One thing I want to emphasize is that this is a workflow demonstration video. The task I chose is helpful for a tutorial video as it keeps the focus on the workflow. o1 pro can handle much more complex requests. I also want to emphasize that I use both o1 pro and Sonnet 3.5 in my daily workflow. Sonnet is a faster model that is still great at codegen tasks. For less complex tasks, Sonnet is a great model to use, and if you run into something complex, you should probably use o1 pro.

[@realmckaywrigley]

@@DanielStitzel-jp3wy In the video description you can find the link for my open source starter template - it's in there! I should probably do a Takeoff module on using that template.

[@annonymbruger]

@@realmckaywrigley it’s quite new so I can’t say I use it on a regular basis, but I tested it for 20 hours with C#, Python and Rust for doing complex code. Also tested it for implementing different use case following SOLID design principles and CLEAN architecture, starting out with most used .net project templates and equivalent for the other languages. And finally for elaborating on specification, both architecture and pure business processes. My conclusion is that the many more black box iterations pro does in most cases isn’t very helpful, when you have more then 3 iterations over a request. I have been using sonnet, mistral, o1 preview and variants of llama for a long time, and can’t imagine working without them. If I had to I would find another work place ;) But the pro o1 is just hype if you ask me. They really have to step up, otherwise I will also be cancelling my 20$ subscription I have had for years. Lately just in hope of gpt-5 or at least something big - so far it just looks like they are eager to onboard investors without actually having anything new to show.

[@ChrisCrooks703]

What is the average efficiency boost with o1 Pro?

[@upadhyayulahk]

@DanielStitzel-jp3wy Do you tend to reach usage limitations (hitting prompt limits) in Sonnet 3.5 for codegen tasks quite frequently?

[@JohnKruse]

This is tremendous. Thanks! I'd tried using o1 with 16x Prompt, but without great success. I'm anxious to try Repo Prompt and the XML Parser. I'd run into someone on Reddit who also espoused using highly detailed context in prompts, but I really didn't grok it until I saw this process flow knitting it all together.

[@FactsNoCare]

This is crazy useful and thank you so much for making these videos. The XML Parser is something I'm gonna be using a ton

[@tdoge]

Would love to see the o1 XML parser/applicator turned into a Cursor/VSCode extension! o1 pro might even be able to one shot that haha

[@pvncher]

Parsing is a ton of work to do well with how many mistakes LLMs can make when following instructions. It’s gonna be built into repo prompt though and even support generating and applying diffs directly.

[@tradingwithwill7214]

Hopefully OpenAI will interface directly with Cursor, Windsurf to read/write all the files or Cursor/Windsurf will use o1 and o1 pro to avoid this ridiculousness. :)

[@realmckaywrigley]

Completely agree!

[@drunktrump5209]

@@realmckaywrigley no way, chatgpt is not priced for coders, rather coders are the golden hen that pay up the hefty API fees. openai will specifical work against bridging this chatgpt frontend layer with any backend function like serving cursor or other IDEs. basically your "workflow" is a pricing hack for $200/month. however, I really like this video and your content, and I've subscribed because of that quality.

[@jaysonp9426]

​@@drunktrump5209calling a programmer a coder is like calling a mechanic a grease monkey

[@mattmmilli8287]

You can get all this going locally now 😅 I’m glad people think cloud is where its at though

[@poisonza]

maybe pipe the web interface output to litellm and use just like the api 😂. But i dont think it is a good idea to use reasoning model for some simple coding task.

[@socrcow]

Amazing, thank you so much McKay! I'm a Takeoff subscriber and your course has been a game changer for me and allowed me to start building my own full stack AI based app, and I l got O1 Pro from your recommendation and am loving it so far. I'm interested to try the XML parser workflow and compare it to what I've been doing using the Mac OS GPT app that can see files in Cursor and prompting O1 Pro that way.

[@GeorgeFoxRules]

This is genius! I literally laughed out loud when I saw you paste 50,000 tokens into o1 pro

[@ezanchi5422]

Milking the code out of the thing 🤣

[@qwertyuuytrewq825]

That's super useful so see an entire workflow! Thank you.

[@briankgarland]

Looking forward to Repo Prompt for Windows. I actually started on the same thing a couple of months ago with a Python CLI app. Yours is much cleaner and has the prompt stuff which I never did. I only did the file tree with filters.

[@slavakurilyak]

There's a waitlist for Windows on their site

[@tommywilczek]

It would be really helpful to get your "o1 format" prompt - I can't find it in any of your public repos

[@ryanthayer8946]

It's in the readme

[@ddemaiolo]

Thanks so much for sharing! Highly educational, much appreciated.

[@MinaEllis-XXAI]

Awesome, love your content, every video of yours is so valuable!!!

[@rlimb]

This is ridiculously cool. Thanks for sharing. One thing I am curious about is how o1 pro would do using this workflow with a less mainstream stack such as Rails 8.

[@yugshende3]

4:15 seems you’ve a typo there where you’ve mentioned it to XLM rather than XML it seems that your workflow still works!

[@nikhilmaddirala]

Is the complexity of switching back and forth between the IDE and ChatGPT (with the complex two way formatting rules) worth it vs using models like Claude 3.5 directly inside your IDE?

[@realmckaywrigley]

You should do both. Use o1 pro to start, and then finish up with Sonnet in Cursor. That's how I currently work!

[@PiyushSharma95]

Very informative & helpful. Will definitely give them a try

[@ademczuk]

could you use repoprompt to write itself as a windows compatible executable instead of mac?

[@pvncher]

Porting whole programs is complex. There’s a windows waitlist on the repo prompt website!

[@Rohan-ni2xw]

Now this is AWESOME. the amount of friction this eliminates really helps my adhd brain get things done!!!

[@RyanBrosnahan]

I am confused about the referenced instructions. I see there are using instructions called "o1 pro base" "o1 format" but I cant find these files. I also don't see them in the git repo. Are these files available? Also, thanks for sharing your workflow. I had not heard of Repo Prompt before it looks interesting. Very good stuff on your channel!

[@FriezaAura123]

Can you share your cursorules please 🙏🏾

[@MehmetBOZKURTEvil]

which editor is that?

[@codewithstephen6576]

how does it handle complex tasks? i.e. not a todo app and is this actually tested? looks like a very basic app, but i suspect it would not even pass testing.

[@agsvk-com]

Hi thanks for the tutorial. Do you share your prompts and rules in the course?

[@finnpoitier]

Since you have been using Sonnet a lot, can you describe, based on your A/B test experiences, how O1 Pro is superior to Sonnet?

[@patruff]

It's not

[@realmckaywrigley]

@@patruff The complexity of task it can handle is significantly higher than Sonnet (which is still a great model - still use this model within Cursor).

[@yurijmikhassiak7342]

I feel o1 models shine when larger context is needed

[@vuhoang5903]

More context window for o1 pro. The context is a real pain working with sonnet

[@RhettReisman]

Do you think it’s meaningfully smarter than the $20 tier models?

[@realmckaywrigley]

Yes. I have A/B tested is fairly comprehensively at this point. It's a lot better (but probably only worth it if you're a professional dev, hobbyists may find it to be overkill).

[@user-ze9tj9yj4t]

I’ve been testing out o1 standard a lot recently, and it makes me really want o1 pro… o1 really is limited, it feels even worse than o1 preview at points and gives quite short responses, much shorter than o1 mini even.

[@elawchess]

@@user-ze9tj9yj4t o1-preview also gave much shorter responses than o1-mini. But o1-mini wasn't as smart. It was more quantity over quality.

[@BillDirkes-o1g]

How do I start the o1-xml-parser? I've followed instructions for installing it with npm, no issues. But it looks like you already have it started in Chrome / localhost for this video.

[@jjdorig9712]

Why do we actually want to use the xml parser again? Is it the same as using repo prompt?

[@slavakurilyak]

Repo Prompt introduced Apply feature which is similar to the o1 XML Parser

[@alphamindset9634]

A video on repo prompt would be good going over everything you have

[@realmckaywrigley]

What would be helpful? My usage us purely what you see here in the video (unless you think I missed something!). Fairly simple.

[@ryanangel3355]

Repo prompt only works with Mac at this time. Is there anything else you know of like that can work with windows?

[@ryanangel3355]

@realmckaywrigley

[@slavakurilyak]

There's a waitlist for Windows on their official website

[@benjaminkarlog8551]

Any windows alternative to repo prompt? It looks sick.

[@pvncher]

There’s going to be a windows port in the new year. There’s a wait list on the site.

[@dudebroholla]

What's the benefit of having O1 output everything in an XML formatted response?

[@jordanmartinez8652]

I really want to try out repoflow for windows. Any idea on a timeline for beta release?

[@unknown_blr_]

hey, it seems it generates those "use server" cause you have it in your cursorrules at line 388

[@friendofzeus]

When i click apply it says that it applied the changes successfully, however git status shows that nothing's changed in the target project. I made sure that the path is correct. Any leads?

[@Doloufoks]

Same here while I use the XML parser to modify script on my local machine

[@itskittyme]

What's fascinating is how your workflow is so vastly different from mine, it's amazing to see in how many different ways AI can be used for coding. The main question I have though is why you are going through ChatGPT instead of through the API directly. My best guess is that it's to cut down on costs? Major bottleneck though.

[@_MrGameplay_]

API access for o1 pro is not yet released

[@itskittyme]

@@_MrGameplay_ ah okay, i'm really confusing all these different model names it appears

[@mattvr]

One thing unclear - why use your custom XML format vs. unified diff format (i.e. what git uses)? Could save an extra tool and prompt

[@kdawgwilk]

Have you tried having the model just format things as a git diff that you can apply rather than that custom format? I would think the model understands that out of the box better than a custom xml format.

[@realmckaywrigley]

XML is significantly better. If you have time check out Anthropic's prompting guide. Everything they do recommends XML, and I agree.

[@terrestrialorigin529]

Is Repo Prompt (or an alternative) available for Linux or Windows?

[@slavakurilyak]

There's a waitlist for both on their website

[@nicomelvin3595]

nice work. To close your iteration loop do you just save the your ide folder, delete your repo prompt original folder and load your updated folder? then start the process again.

[@pvncher]

Repo prompt automatically fetches the latest versions of files and also supports applying edits directly now fyi. It’s similar to the xml flow here but built in, and it also supports search replace.

[@nicomelvin3595]

@@pvncher saw that, though you cant test. the refresh is cool.

[@sunkwolf]

I really liked the video! I do something similar to what you do with your program, but in a more manual way. Having Repo Prompt would be a great help to speed up my workflow. However, the download link is not directly available. I signed up on a waitlist form, but I haven’t received any email to download the application. Still, your video is great content. I enjoyed seeing that you use a workflow very similar to mine. It makes me feel like I’m not alone in the world and that my method might not be so wrong after all.

[@pvncher]

The windows build is not ready yet. You’ll have to hold on a bit as it gets worked out. Thanks for the interest!

[@Robin-ky8vr]

do you have recommendations on how to use this in science for helping structure and write papers, include loads of papers, weigh arguments ect. im curious how this would perform, given enough context, to use context specific words correct, with gpt-4 some definitions are seriously wrong.

[@pvncher]

Hey I’m the dev for repo prompt! Would love to talk about this use case. I’m not sure if you’re on a Mac but please do give it a try! It supports all text files with most file formats. PDFs might be wonky though There’s a discord link on the website if you’re interested in sharing.

[@benjaminkarlog8551]

@@pvncher What is the state of a windows version? 🙂

[@pvncher]

@@benjaminkarlog8551 WIP! Im only one person so I have to manage my time carefully, but it is being worked on, and Im hoping to potentially get help shipping it.

[@jjdorig9712]

how does this compares to cline and sonnet 3.5?

[@markkwong9512]

What did you mean was unreliable? Have you tried structured outputs instead of xml outputs for o1?

[@augmentos]

Are you seeing the code results so much significantly better than even sonnet and 01? Like specific to code do you definitely say it’s worth it?

[@Human_Evolution-]

So o1 seems noticably better than Sonnet 3.5?

[@Whyoakdbi]

What about the security concerns around repo prompt?

[@realmckaywrigley]

Everything you seem me do here with Repo Prompt is completely local on my machine.

[@klammer75]

Very cool and awesome share! Quick question...does anyone find sonnet 3.5 in the anthropic web UI to be a better coding assistant than sonnet 3.5 in Cursor? It's a noticeable difference IMHO...anyone else have a similar experience?

[@toxtethogrady4610]

yes

[@yurijmikhassiak7342]

It's not model. Cursor give too large prompt with his own rules, your rules, vector search, files, sometimes it's too much. 300 lines of code is already too much.. so if you are in the loop. Ask cursor to extract relevant code and only then ask to analyze it without writing code. Or go to web model with relevant code.

[@joshua_sx]

Where can I find the 01 format prompt? In your github?

[@realmckaywrigley]

Yep in the repo!

[@user-kq2yj6ov5x]

@@realmckaywrigley what about the 01 pro base prompt? cant seem to find it and you didnt explain it in the video. is this it? You are an expert software engineer. You are tasked with following my instructions. Use the included project instructions as a general guide. thanks for the great tutorial.

[@annie_the_great]

Can I use this on Windows?

[@vuhoang5903]

All fun and play until I find out that Repo Prompt is only for macOs

[@slavakurilyak]

Windows and Linux support coming soon

[@jmstockholm]

Excellent!

[@barakben-ezer294]

Singularity is here

[@yellowboat8773]

Thanks for the vid mate, i feel that the 10x price increase doesnt justify the value though.....

[@dan110024]

yeah it absolutely doesn't lol.

[@realmckaywrigley]

It is if you are a professional developer. For hobbyists not so much - totally get that.

[@agsvk-com]

Too bad the apps are still for Mac OS and flowvoice only for Silicon. May have to wait months for the apps to be out.

[@annonymbruger]

I tried similar and ended up cancelling my pro subscription. For iterative work flow it was not worth it

[@realmckaywrigley]

You'll want to do the more iterative work with something like Sonnet 3.5. And at the end of the day the best workflow is the one that works for you :)

[@annonymbruger]

@ wake up. It’s a 200$ subscription- it should work better

[@marvpipe]

the xml structure of the prompt is not matching the xml parser in your project. the file node is missing

[@marvpipe]

ok and it creates the wrong folder. would be awesome if you can check your prompts and correct them in the readme :) thank you

[@pvncher]

There’s an xml parser built into repo prompt now fyi

[@kaiiserni]

aider --copy-context FTW!

[@Vance-kq1cx]

Why not use cursor???

[@realmckaywrigley]

The model is only currently available in ChatGPT.

[@nayemnesat]

thanks

[@bastabey2652]

is this o1 pro the $200 / month model

[@LynnXee]

I am working on a research project involving developing a framework that implements vision-aided autonomous robotic navigation under uncertainty, running this it on real robot and simulator. The context for it would include the codebase, published articles that with msthematical formulations of the algorithms, wikis of mathematical libraries used as a backbone and instructions from the researchers. I'm using chatgpt daily for coding but want to encorporate it deeper into our workflow to save development type. Any suggestions on additional resources that I can use? Do you think $20 o1 model can still handle this?

[@jacquelynhalpern9433]

I'm a newbie but I'm working on a full AI app! Your videos have been so helpful. One thing I'm trying to understand better is .cursorrules. In your other video I noticed you didn't keep them in a separate file but here I see a more robust set of rules. Do you have any suggestions for creating cursorrules for larger applications similar to what I'm seeing here? Thank you!

[@Topsquadjapan]

this is a lot of work for a basic request

[@carterjames199]

Hey I love the video workflow is GOATed

[@barakben-ezer294]

The age of the machines that build the machines has arrived. We are in very edge of singularity. I feel the gravitational pull.

[@hydrohasspoken6227]

cool down mate

[@MugiwaraNoDeji]

Nice, but I think this can be optimized a bit

[@realmckaywrigley]

More than open to suggestions :)

[@gani2an1]

That is way too much work. We need API for this so that we can connect it to things like windsurf

[@realmckaywrigley]

The API version is rumored to be here within 2 weeks!

[@la6188]

lol just use aider…

[@haploid2k]

This "RepoPrompt" thing looks pretty cool. It's basically like Aider, except far less productive, and built for five year olds who are too scared to interact with the terminal, and prefer to point and click like a chimpanzee. Good call!

[@pvncher]

Why do you assume it’s less productive than aider? It supports api calls directly and can create file edits more reliably than aider, and cheaper too by delegating edits to different models.

[@Matrix-mq7ul]

Why not use Windsurf with Claude-3.5 ?? Literally a thousand times better hahah

[@elawchess]

Except it's not is it?

[@realmckaywrigley]

I use this to begin my work and continue in Cursor with Sonnet 3.5 as shown in the video.

[@elidelia2653]

Need that API!

[@realmckaywrigley]

It can't come soon enough!!

[@wdmeister]

I wish there was a different word for making websites than "coding". It's usually trivial and LLMs can handle it quite well but as soon as you try to write something more complex LLMs spit out useless crap. After reading the title, I was hoping for something new, but I see it's again a simple webdev coding. Boring.

[@Slyndc]

Second

[@CocaNoah]

Its so sad that everyone showing off these ai tools build a to do list. Why not build something useful, so its a more realistic scenario

[@realmckaywrigley]

The video is just meant to teach people the workflow. See pinned comment.

[@CocaNoah]

@@realmckaywrigley I understand. But in my opinion the videos would have a better value if i see real applications build with these tools instead of everyone showing the same to do list demo for every tool that comes out. I mean it would only take a different prompt to generate something useful, so why deciding for a to do list?

[@yurijmikhassiak7342]

Coz that's one of the easiest types of apps. Simple CRUD. People love utopia when everything works. AI coding is 10x harder with backend and larger codebase.

[@EDashMan]

First