This year is not like last year. Yes, I know. Anyone who follows the news or buys eggs knows that this year is different. But I'm not talking globally. Instead, I'm talking about Google and its Gemini Code Assist announcements at the Google Next get-together.

Last year, I discussed Google Next 2024 announcement of its Google Code Assist AI programming helper. It offered a lot of interesting integration and productivity possibilities for programmers. But it was dogged by one irredeemable limitation: the Gemini implementation couldn't code its way out of a paper bag at the time.

When I ran my coding tests last year, Gemini Advanced, based on Gemini 1.5, failed three out of four tests. So, in 2024, when I talked about Google's powerful new programming tool, its programming ability was dogged by the fact that it was pretty much nonexistent.

Also: Gemini Pro 2.5 is a stunningly capable coding assistant - and a big threat to ChatGPT

This year, concurrent with Google Next 2025, I'm going to discuss new features for Google Code Assist. The difference is that Google now has Gemini 2.5, andthatversion didn't fail my coding tests. In fact, it aced them with extra credit.

That makes our discussion of a tool that helps to program much more notable because, this time, the code it produces (or helps you produce) will probably work.

That takes Google Code Assist from a pleasant little curiosity of a programming utility to something that could well become the powerhouse core of your programming efforts.

Got your attention now, don't I? So let's dig in.

Gemini Code Assist Individual news

Gemini Code Assist comes in three variants: individual, standard, and enterprise. The individual variant is free and offers a rather substantial number of features given its lack of price.

Upgrading to standard for$22.80/user/month or$228/user/year gets you intellectual property indemnification, Gemini in Firebase (Google's backend platform for developers), and Gemini in databases.

Also: The best AI for coding in 2025 (and what not to use)

Punching it up to enterprise for$54/user/month or$540/user/year adds on customized code suggestions from your GitHub, GitLab and Bitbucket codebases, Gemini in BigQuery and Apigee, and some automation flow features for application integration.

The big news for the free version of Gemini Code Assist is that it will now come with Gemini 2.5, which I think is good news.

Here's a bit of inside baseball when it comes to Google announcements. A little before announcement time, Google sends a giant press release to columnists like me. This time, it was a 76-page release covering a wide range of announcements.

Now, you'd think with a 76-page press release, Google would indicate whether the Gemini 2.5 it is announcing the availability of for Gemini Code Assist individual is the same as the Gemini 2.5 Pro version that has been available for testing. You'd think.

Also: The top 20 AI tools of 2025 - and the#1 thing to remember when you use them

I did reach out to Google's press relations team and I was told that Gemini 2.5 and Gemini 2.5 Pro are two separate beasts. I was told that the Pro version is a thinking model. Here's a blog post that explains the difference.

As to whether developers can benefit from the Gemini 2.5 Pro performance results I got, the answer is yes. According to the PR guy, "Gemini 2.5 Pro is now available for developers in our Gemini Code Assist for individuals offering [the free tier]."

Now, that's just cool.

Gemini Code Assist agent news - and my questions

There are a lot of things you'd think Google would specify in all those pages of announcements. For example, Google is announcing agent capabilities for Gemini Code Assist.

Nowhere on the product version comparison table or in the press release is there any clarification of whether these agent features are available to individual, standard, or enterprise, or only one or two of the variants.

Also: Microsoft is offering free AI skills training for everyone - how to sign up

But my friend from Google PR sort of helped clarify this one. He told me, "The new agents are now available in private preview that anyone can sign up for. Tier availability to be shared post-private preview."

There's a pretty big leap from writing code on your computer to shipping a product to a user. End-to-end, we call this the software lifecycle. Google's Gemini Code Assist agents are being introduced to help with tasks across the lifecycle.

The company specifically announced agents in the following seven areas.

1. Generate new software

The company says Google Code Assist will create applications from product specifications written in Google Docs. This one could be problematic in so many ways.

Unless Google is also documenting a specification writing language, the AI is likely to miss many features and capabilities not specified. This sort of feature will give non-programmers the idea that they can generate working code. However, describing what you want is just a small piece of the overall project.

It's not clear what the back-and-forth process will be. Does new code get generated each time a Doc is updated? Or can code be written in the development environment between rounds of specifications in Docs, and can the two be merged? Do you have to write an entire app, or can you specify functions and subroutines that are later incorporated?

2. Migrate code

The idea is that agents can transform code from one language to another and possibly translate code between languages and frameworks.

This is a some yes, some no kind of feature. I have tested and seen success giving a function to an AI and asking that it be rewritten in another language. Assuming the source and destination languages have facilities for the algorithm being converted, this is a fairly effective tool.

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

But switching between frameworks could lead to a world of hurt. Frameworks often require hundreds of files and lots of little niche capabilities. Again, if we're talking about a single function, I think the migration would be possible. But if you're thinking about giving your entire folder over and trying to get the AI to make it work on a different platform, I think you'll just have a mess on your hands.

3. Implement new features from GitHub

The idea is that the AI reads "issues" from GitHub and implements that in code. Issues are bug reports or feature requests. So, if an issue listed is something like "Add 2FA to login," the idea would be that the AI would read that and then actually write two-factor authentication code into the codebase.

And yeah, for some issues, this would be a nice time saver. But the 2FA example showcases some serious limitations. For example, what method of 2FA would be used? What authenticators would be supported? What libraries should be used? Do you use free libraries or license something with additional capabilities?

It's fairly easy to implement a feature. My concern is that coders will do so and stop thinking abouthowto implement those features.

4. Perform code reviews

I think this is a great use of AI agents. Even without agents, I've been giving ChatGPT (because until recently Gemini couldn't handle it) a prompt consisting of "what's wrong with this code," followed by a block of code.

I did that to see if the AI could find fault with my implementation or surface anything that might need more work. A few times, I was lucky, and the suggestions made were minor. But most of the time, that simple prompt uncovered some fairly serious coding error or omission that needed further work.

Formalizing that into a code review process could be enormously helpful. I look forward to checking this agent out for myself.

5. Generate tests

I saw a version of this demonstrated during my Gemini 2.5 Pro programming test. I asked Gemini to fix a regular expression calculation and not only did the AI do what I requested, it also provided me with a set of test cases (both positive, where it would work, and negative, where it was supposed to fail). This allowed me to quickly and easily confirm the functionality of the code.

Using an AI to build and execute tests and report on the results is a great use of AI technology. Of course, you'll have to review the tests, explore how they're implemented, and fix the occasional test where the AI misses the point. But still, this is a great way to increase code reliability while saving time.

6. Perform AI model testing

If you want to see how an AI model behaves with certain inputs and outputs, you can use the AI to create a series of validation tests.

This can be particularly valuable if you're concerned about content safety and the performance of AI subsystems implemented within coding architectures.

7. Create documentation

Google has a wiki-builder built into one of its predefined agents. Many programmers tend to leave documentation as an afterthought. Even if the AI doesn't produce the level of documentation that a good technical writer with a coding background can produce, the AI could create a very good starting point that can be modified, improved, corrected, and tweaked by the software developers who are crafting the code.

Time to start coding

The 2.5 version of Gemini and the ever-improving Gemini Code Assist seem to be hitting their ready-for-primetime point for production and individual developers.

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

What about you? Have you tried Gemini Code Assist yet, or are you waiting to see how Gemini 2.5 performs in real-world use? What do you think of the new agent features -- do they sound helpful, or do you see potential pitfalls? Are there specific parts of the software lifecycle where you'd trust an AI agent to step in? And how important is feature availability across the free, standard, and enterprise tiers to your team or project? 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.

Featured

I never pay full price for PCs or Macs, thanks to these 7 money-saving tricks

Linus Torvalds built Git in 10 days - and never imagined it would last 20 years

Lenovo ditched tradition by redesigning the latest ThinkPad - and I'm glad it did

How to start using the new Linux terminal on your Android device

• I never pay full price for PCs or Macs, thanks to these 7 money-saving tricks
• Linus Torvalds built Git in 10 days - and never imagined it would last 20 years
• Lenovo ditched tradition by redesigning the latest ThinkPad - and I'm glad it did
• How to start using the new Linux terminal on your Android device