I test a lot of AI coding tools, and this stunning new OpenAI release just saved me days of work

Last week, OpenAI quietly dropped a programming bombshell post on X/Twitter. It turns out you can now connect GitHub repos to Deep Research in ChatGPT. What makes this particularly interesting is that you can put ChatGPT Deep Research to work scanning that repo for all sorts of yummy nuggets of information.

GitHub is an online resource owned by Microsoft that holds an enormous number of programming projects, both open source and proprietary. It’s used by teams to coordinate and track development. A GitHub repo is a repository for a given project. What Deep Research can now do is dig through the full source code for an entire software project and provide value.

Also: The best AI for coding in 2025 (including two new top picks – and what not to use)

Look, I have to come clean before we go on any further. I’ve tried this thing out. It can effectively produce an internal code review. This is very cool. I am seriously chuffed by what we can get this beastie to do for us. AIs creep me out, so why do I want to hug this one?

My editor says, “That’s all part of their plan.”

Seriously, this is some powerful mojo. Most of my biggest programming projects have been based on code acquisitions from other programmers. Digging through their code and deciphering what the living heck they were thinking throughout the code can be an excruciatingly time-consuming process.

It’s not just about a line-for-line what-does-this-code-do sort of thing. It’s about how the project is architected. What are the various modules, and how do they interact? Where are their strengths, and where are their potent gotchas? What makes the UI function? What needs to be done to add functionality, remove functionality, or replace functionality?

Also: How I used this AI tool to build an app with just one prompt – and you can too

Deep Research can now do this with GitHub repos. It is almost totally sweet. Almost is because of the restrictions. Let’s cover those first, before I show you the hoops I made this thing jump through.

According to ChatGPT 4o, here are the limits for each of the subscription tiers.

The Standard Deep Research queries are powered by the o3 model. I pay $20 per month for the Plus tier, which gets all of 10 Deep Research queries of the sort we’re going to talk about per month. That’s not a lot to be able to fully dig in and have ChatGPT deconstruct an entire project.

Also: How to turn ChatGPT into your AI coding power tool – and double your output

I used three Standard queries just testing for this article, and another four Lightweight queries just convincing the thing to work for me.

So, it’s awesome. But full usage may require you to upgrade to the $200-per-month Pro version, at least until you’re done with active examination of a new project. That seems like a lot, but when I think of the number of hours I put into each of the projects I acquired, two hundred bucks to short-circuit weeks and weeks of work seems like it might be well worth it.

I can certainly see myself boosting my plan to Pro for a month, doing some major Deep Research digging for the month, archiving all the answers, and then dropping back down to the $20 tier.

Testing against my own code

While I think that one of the biggest benefits of this capability is analyzing someone else’s code in order to work on it, I decided to have it look at one of my projects. I felt I’d have a much better chance of seeing how it performed if I were already familiar with the code I was asking it to forensically analyze.

My Private Site is a freemium security plugin for WordPress that promises to protect an entire site with one click. I acquired it from another developer ten years ago and have been updating and expanding it ever since. It’s running on more than 20,000 active sites. Three purchasable add-ons support my expenses maintaining the core code, which is free and open source.

My first step was to change the language model to o3. And yes, GPT-o3 is more advanced than GPT-4o. The AI is amazing, but the product management in terms of naming conventions is not.

Next, I cut and pasted the prompt directly from the X post that announced the feature. With only 10 queries available to me, I didn’t want to waste any trying to reinvent how to unlock this capability.

I made sure to click the Deep Research button and then the tiny drop-down menu on the button. This gave me access to the new GitHub capability. I clicked on the GitHub drop-down and typed in the full name of my GitHub repository.

Prior to this step, I had to give ChatGPT access to my GitHub repos. I’m working on a second article showing how to move a codebase into GitHub and make it available for ChatGPT. For this article, though, I want to focus on what this new Deep Research feature can do.

Also: What is AI vibe coding? It’s all the rage but it’s not for everyone – here’s why

Once the repo is selected, you’ll see it at the bottom of the prompt field. This narrows the scope of the Deep Research queries to this named structure.

Once I fed o3 Deep Research the original X prompt, it asked for some clarification. Since I originally specified the repository for the entire My Private Site project, I wanted it to look at that entire corpus of code.

This was a new experience for me. In the past, I’ve managed to get ChatGPT to examine one file or one block of code. But this repo contains 231 files, about 12,000 lines of PHP code, and a lot of supporting code. Telling it to examine the entire repo meant telling ChatGPT to look at all of it.

And then the process began.

And then it stopped. I got this as an error message.

Of course, there was going to be an error message. ChatGPT thought the problem might have been one of three access issues: whether I was the owner, something about how it was connected, and whether the repo was public.

Once I confirmed I was the owner and that the repo was public, that was enough to convince o3 to get back to work.

Some serious chugging and thinking went on behind the scenes. It took ChatGPT about 20 minutes before the AI was ready to tell me things. First was the repository review. You can see from the green arrow pointing at the scroll bar just how big this review document is — big.

The review includes five major sections:

1. Project purpose & architecture
2. Key modules & components
3. Tech stack & primary technologies
4. Notable open issues and upcoming changes
5. Suggestions to improve code quality

I was heartened at the AI’s conclusion that, “Overall, My Private Site is a well-architected plugin that does its job effectively.” As an editor, it pains me a bit that some of the section heads used an ampersand while one used the word “and,” but I won’t hold it against the quality of analysis.

Also: How to use ChatGPT: A beginner’s guide to the most popular AI chatbot

The analysis was good. Really, really good. In about 20 minutes, it completely “got” My Private Site. It did a deep dive and explained the internal workings clearly and accurately. It’s a full-on definite wow.

If you want to read the full analysis, as well as some early setup discussion and some follow-on questions, click here. I’ve shared the entire Deep Research session, so you should be able to follow along if you’d like.

Deeper deep research

As I said, I found the analysis to be impressive. But I wanted to see how much it could help me if I wanted to really work on the code. There are two major subsystems that I’ve had to carefully document in my lab notes to be able to keep track of.

The first is how the UI works. I completely rebuilt the dashboard for My Private Site from how it worked when I acquired the plugin back in the day. It is fairly complex and often a bit confusing. So I asked ChatGPT to dig into it.

Once again, Deep Research did a great job. It produced an analysis of that entire subsystem, stayed on track, and provided clear, comprehensive, and useful information. You can find ChatGPT’s full analysis in the session share link I provided above.

Second, I wanted to see if Deep Research could understand and explain how access blocking works. Access blocking is the main feature of My Private Site. It’s also extensible, allowing add-ons to modify the functionality of the access-blocking capability, often with entirely new protections and access options.

Also: I tested ChatGPT’s Deep Research against Gemini, Perplexity, and Grok AI to see which is best

Could Deep Research explain the access-blocking capability, and then explain how add-ons can tap into it? It could. Here’s a sample of that interaction.

Once again, Deep Research provided a spot-on analysis. You can also find ChatGPT’s full analysis for this capability in the session share link I provided above.

Bottom line

What does it all mean? This is an extremely useful tool. I have acquired, updated, and transformed a lot of code into viable products over the years. I then sold off most of those properties once I felt their lifecycle with me had run their course. Because all of my acquisitions involved bringing in other peoples’ code, a tool like this would have proven to be an enormous time saver.

I am concerned that the 10-query limit is too limiting. While $200 per month is arguably worth the cost for this capability, it’s still a tough reach, especially after an acquisition. It usually takes a year or more before a new acquisition starts paying for itself. But I do expect that the current pricing structure will change over time.

Also: Is ChatGPT Plus worth your $20? Here’s how it compares to Free and Pro plans

I’m also curious how it will do as a coding helper. For example, My Private Site runs inside of WordPress, itself an open-source project with a mirror on GitHub.

I can envision a time when, trying to figure out how to tap into some capability (or get around some restriction) in WordPress, I might use Deep Research. It could look at the entire WordPress codebase, ask for clues about how components work, and provide advice about where I might look to get my code’s interface with it working the way I want.

Overall, this is an amazing tool. But use judicious caution when incorporating it into your workflow. Like all AI, it’s likely to get some things wrong. It may not be able to help with everything. And keep in mind that this is not a coding tool. It’s an analysis tool. Limit its use to when you want a bigger picture, which you can then use to fuel your understanding before and during coding.

Also: How to use ChatGPT to write code – and my favorite trick to debug what it generates

Have you had a chance to explore ChatGPT’s Deep Research capabilities for GitHub repositories? What was your experience like? Do you see this as a valuable tool for analyzing unfamiliar codebases or speeding up onboarding? How do you feel about the current usage limits and pricing tiers? Could you see yourself using this regularly in your development workflow, or do the limitations hold you back? Let us know in the comments below.

---

You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.

Get the morning’s top stories in your inbox each day with our Tech Today newsletter.