Google I/O: The presentation that left us speechless
Really, the Google I/O presentation has been spectacular. It was the bomb, in a nutshell. They have put a lot of impressive features on the table, which if later at the moment of truth are as good as they show, would undoubtedly put Google in the top 1 of the best artificial intelligence providers.
But today what urges us here is Jules, that code assistant. You can enter its website yourself, which is super pretty, I mean it’s not irony, I really love it. You have it at jules.google. They have a very nice octopus that edits the code for us. Honestly, the pixel art is cool.
What exactly is Jules?
At a glance you can see what they want to indicate to us: Jules takes care of doing the tasks you don’t want to do. That’s it, as simple as that.
Like resolving bugs, upgrading dependencies, doing tests (because nobody likes doing tests, really), even fixing your colleague’s wrong code. In the end, their motto is that they want to give you more time to solve and do the code that you are really interested in doing. And the rest of the minutiae that you are not interested in doing would be handled by Jules.
The truth is that the setup they seem to have mounted here is quite big, but well, for being Google it’s normal, right? I expected no less from them.
Integration with GitHub: Everything connected
Once you enter and log in you have direct access with GitHub. You basically have to connect your account so that it can access the projects and can create branches, pull requests, issues… everything, come on. Everything that can be done on GitHub it can do.
Well, I’m not a Google programmer, but surely they have easily done this integration with GitHub with the GitHub MCP. That if you don’t know what an MCP is, it’s your lucky day because I have a very interesting post on the blog where I explain what an MCP is and I really explain it without beating around the bush, without technical words or weird things, I explain it so that it is clear to you and I give you examples so that you become a better programmer and much more productive.
How Jules works inside
Returning to the subject of Jules, we see that once you select the repository where you want to work, it clones it in a virtual machine and there it performs checks and makes the changes. As if that were not enough, once the changes are made, Jules will execute the tests you have in the application to verify that nothing has been broken. And if you don’t have tests, as they indicate here, it will create new ones for you to verify that functionality.
And of course, Jules shows you what it is doing. It will show you the execution plan it has by tasks, which this, well, is something that Juni from JetBrains already does. It will reason about what you have asked it and will make a diff of the changes. Come on, like when you go to review a colleague’s pull request, you see the changes there, the differences they have touched, well exactly the same. Only this time, instead of being your colleague, it is a Gemini model.
Surprising, huh?
Best of all: It’s free (for now)
And all this paraphernalia that I have told you, for the moment, is free. Totally free. You can enter the page and sign up for the beta. I did it two days ago and I already have access, so I don’t think it will take you long to get access. Or well, maybe I’m totally privileged, who knows.
Jules vs Firebase Studio: Do not confuse them
One thing I want to highlight is that not long ago Google also released Firebase Studio, which is an online code editor, it is not downloaded, which does something similar, but it is not the same. In truth, it has nothing to do with it.
I say similar because it does have artificial intelligence and it is Gemini’s, but it has nothing to do with it. So that you don’t get confused, Firebase Studio is more of a more visual code editor, which focuses on creating applications, especially in the technology stack that Google prefers, in the most visual way possible. It is a bit more high-level, because you tell it the orders and it creates it for you, although you do also have the code part. And above all it is very focused on the Firebase theme, because well, this Firebase Studio, well some backend it would have to suggest to you, of course, the Firebase one.
But on the other hand, Jules does not opt for any technology, nor does it suggest using Firebase, nothing in particular. In the end, you can use Jules in any type of repository you have on GitHub, that’s the beauty. And although it is not so much a visual code editor that you ask and it does things, like Cursor, it is something more autonomous and as its name suggests, asynchronous. That is, I ask for it and we will see later, it does it when it wants, when it can.
Jules is thought more at a low level, in fact they are complementary. You can create your application with Firebase Studio and then you can ask Jules for tasks, separately, well later because you are iterating the application or improving things.
The competition: Jules vs OpenAI Codex
A comparison that would be fair with Jules is OpenAI Codex. Which this, well, was announced very recently. A week, I don’t think not even two weeks ago. Codex in the end is the same. Another code assistant that allows you to connect your repository, blah blah blah blah blah blah. It is the same, exactly the same.
But there is only one difference. Codex, currently, as of the date I record the video, is only available in the most expensive ChatGPT plan and no, I am not referring to the 20 dollar plan, I wish. It is available in the 200 dollars monthly plan, monthly yes.
So, as you can understand, I cannot show you Codex, nor can I compare it because the channel budget is not enough to do so. But the advantage of Jules for the moment, because we already know that these things are changing, is that it is free. So it is a very good time if you want to try it and I understand that you will be much more interested because in this channel we love free things.
The power of Gemini 2.5 Pro behind Jules
Another thing I wanted to comment on about Jules is that it indicates that it will always use the most updated Gemini 2.5 Pro model. Because you already know that they are releasing new versions. Therefore we would be using the model, well, practically the best model on the market and most expensive.
OpenAI currently has O3, which is comparable, and it is rumored that soon it will release a new model that will surely be the best on the market. You already know that here, in this world, every week a model comes out that is the best on the market. But well, the fact is that Google is letting us use its most recent and expensive model totally free. I indicate it again in case it was not clear.
Ah, and well a detail, if by one of these you want to try more thoroughly and in a Gemini 2.5 chat, Google has another tool apart, aistudio.google. Here I leave it to you, where you can try these models totally free. Here the trick is that in the end the chats do not have memory and it is a bit more laboratory if you want to tinker and you don’t have an API key either, if you want it you have to pay, but well to get out of the way, try things because you also have the image generation that came out a month or two ago, you have many things that you can try here for free, I leave it as a tip, it has nothing to do with Jules, but I thought it was interesting to mention it.
My real experience: Testing Jules
Well, getting down to work I connected, synced with GitHub as it asks, very simple, and connected a repository I had around there, private, an Instagram API, which I am doing, in my spare time.
The first problem: Saturated servers
And I say, well, they say it themselves. It is seen that there have been so many people who have tried to enter, and in the end it is a free service, that they go a bit slow and they are already warning you.
The first point against that I would put on it. I understand that it is free, but of course, you cannot launch a product and people can barely use it. I would have preferred to pay a small fee and that there were fewer people and it was more fluid. But well, let’s wait and give it time.
Testing with a real error
If you enter their documentation they tell you a little what else it can do, more in detail, how to make the prompts, but well, in the end how to make the prompts is like in any artificial intelligence. You have to be clear, concise and not be ambiguous.
This time I wanted to try a prompt that any of us could do and that is that this same API I am dockerizing it and it gives me an error. So I copied the error that the container is giving me and I told it, hey, I have this error that is happening, help me solve it. And I left it there.
If I’m honest with you, I left it all night. I don’t know when it was resolved because it is not clear to me, Google already does this on its servers, but I am sure it took quite a while.
The disappointment: Mediocre results
And well, I like to be honest in this channel, I don’t like to get carried away by fads and I don’t want to sell smoke or say things that are not. So I’m going to be honest with you. I launched the prompt and after receiving the result, well, I was disappointed, honestly.
First of all, before commenting on the result, is that the message I sent appears cut off, the whole error I passed does not appear and I don’t know if it is a visibility issue, that it doesn’t show it to me whole, or that it really cut the message and said, I only accept up to here. Which would seem pretty bad to me because well Google is known for its context window because they accept many tokens and I didn’t ask for anything super long either, I passed a decent small log. So that is the first thing that surprised me.
The next thing I didn’t like is the answer. The answer it gave me was quite basic. Yes, it is true that it complies with what it says: it has created a branch for me and that branch, if I go to GitHub itself, it appears to create the pull request and that pull request already has a description and title. Maybe they are not using an MCP and are only using the API. It may have slipped.
The changes: Minimal and without solving the problem
What changes have you made? The truth is that the changes it made were minimal. And, if I’m honest, it didn’t work nor did it solve the error. Which well, I can understand because maybe it’s not a simple error, but it didn’t try at all. It made a very simple change that didn’t contribute absolutely anything to me.
Second attempt: Even worse
Since it didn’t solve the problem, I wrote to it again. I told it, hey, this hasn’t solved anything for me, etc. And here it did something I didn’t like. It started asking me to pass information from the repository itself. It asked me to copy the docker compose, but it already has access. Why are you asking me for that? I don’t understand anything.
The truth is that it is super disappointing. This is first grade agent stuff. It doesn’t find something, it looks for it in its files. Also that it cannot reference it. I cannot put an at sign and the file name as is done in all code editors. No, no, no, I had to copy and paste.
But on top of that I copied and pasted and well, I don’t even remember if it answered me or answered me nonsense because it goes very slow and the answer remained thinking, thinking and thinking and in queue, basically.
What I do like about Jules
I really like the approach. I like that the messages you have made and the tasks for each message appear here next to it. I thought it was going to be more limited.
The limits of the free plan
Why? Because it explains that it has limits. That’s right, everything free has limits. What are the limits?
- 3 concurrent tasks per day: That is, tasks that run at the same time
- 5 tasks per day
- 5 codecasts
Because well, I forgot to mention something quite cool. That is that you can ask it to generate a small podcast for you. Yes, a podcast with voice. Of the latest changes that have been made in the repository. Something that is quite good. If you arrive, you have many colleagues. You tell it, hey, generate it for me. And it generates what was done yesterday, for example. It’s very cool, but I haven’t been able to try it. Because the servers go very bad.
You can ask them to extend the limits, but you have to justify them very well, so let’s see what you tell them, what tall tale you tell them.
My final verdict: It is very green
So, my opinion, my verdict, although well, if you have read the whole post you surely already know, it is very green, it is super green.
I don’t know if it is because they have had too many unexpected customers or because it is a product that is not yet fully improved. But my experience today is bad.
Do I recommend it?
Yes, register, it doesn’t hurt to have the beta and it is possible that they improve it tomorrow, next week, next month. And you already have access to the beta and you don’t have to do this registration because then more people will want to register etc.
But today for working it is useless. No, it is of no use to me that after so many hours it makes me a PR with minimal changes. No, it is useless to me.
I tried to test another repository and told it to upload versions and add comments in a readme. Well, it hasn’t put them. Quite a few hours have passed. Quite a few.
The current reality
So that is my verdict. You will tell me, why have you made the post? Badger, what a badger you are? No man, I saw the tool and I got quite excited. But I would have been less excited if I had seen a post like mine explaining and telling the truth about the tool.
So well, to this day you don’t have to worry about AI taking your job. Because for the moment it won’t do it. We’ll see tomorrow what they surprise us with and if they can take our job.
Conclusion
Jules has an interesting approach and the potential is there. Google has the resources and technology to do something really good. But for now, the execution leaves much to be desired.
If you want to try it, go ahead, but don’t expect miracles. And if it works better for you than for me, tell me in the comments, maybe I’ve been unlucky.
For my part, this has been all. If you like these topics I recommend you watch my comparison of Cursor with Juni AI from JetBrains. We see each other in the next post and thanks for reading. Until next time!
What do you think?
Leave your opinion, question or suggestion. Comments are synced with GitHub Discussions .