As Google Project Manager for developing tools, Ryan Salva has a front row seat for ways when the tools are changing the coding. Github and Microsoft’s Forminly, is now responsible for tools such as Gemini Cli e Gemini Code AssistNudding Developer in the new world of agent programming.

His released team New third -party research We were killed by showing how developers actually use artificial intelligence tools and how many progress have remained to be done. I sat with save to talk about the relationship and his personal experience with the coding tools AI.

This interview has been modified for length and clarity.

Every year, Google makes a survey on developers’ tendencies, but this year’s report really focuses on artificial and specific intelligence tools as the agents developers are willing to achieve their approach. Wasn’t there something in the search that surprised you?

One of the really interesting results was the median date in which the developers started using the AI ​​tools. They discovered that it was April 2024, which corresponds quite neatly to Claude 3 outgoing and Gemini 2.5 coming. This is really the dawn of reasoning or thought models, and arousing that time, we have improved a lot to tool football.

For coding activities, you really need to be able to exploit information to solve problems, so GREP may be necessary, it may be necessary to fill in the code. If the compile code may want to perform that unitary test and that integration test. I think Toolling is truly the important piece that has given the models the ability to be ASY self-regulating to move.

How are you personally using coding tools AI?

Most of my coding these days are for hobby projects and spend most of my time using tools based on command lines. So this included the twin cli. Then? And always give alone use a tool based on the terminal, so I am really heterogeneous surrounds the IDes I use. I use Zed. I use the code VS. I use the cursor. I use Windsurf, everyone, because I am interested in seeing only the world and how the industry is evolving.

From a professional point of view, the Product Manager tends to live in documents, so the first thing is to use to to help me write special documents and requirements.

I’m curious to know how it works. You are using CLI twins to build Gemelli CLI, but I guess you don’t just run.

A development task will begin in nature as a problem, perhaps it is a Github problem that someone has dropped with a bug. Often, if I am really honest, it is quite not very superverified. So I will use Gemini CLI to create a more robust requirement document in Markdown. This will usually create about 100 special lines, but also with outcomes drivers. So I will use Gemini Clo to write coding on this specific and general preferences in the team documents.

Throughout the engineering team, we have a couple of different levels of rules and documents of Markdown that are consumed by the model, simply by throwing our way of working: this is how we do the tests, here’s how we manage addictions and so on. So when it produces the code, it also works from those documents.

And while Gemelli Cli is going through and arousing the resolution of the problems, I will make it update my doc requirements saying: “I have solved this step. Now I am at the next step” and so on. Each of those creates their own commission and pulls the request in the re -emperial, so I can always rewind or cancel.

I would probably say from 70% to 80% of my work is I who work in the terminal with a natural language, trying to use Gemini Cle to develop the requirements and therefore allow Gemini Clo to write most of the code for me, which I will then go to review and read because whatever the IDE. But above all I am using the IDE as a place to read the code, rather than writing the code.

Do you think there is a future for the RAW computer code? Or will we move everything to the terminal windows?

For three decades, the IDE was where we had to do everything in the development of the software. You had the ide, you had the browser and had the terminal window.

I think it is still so widely, but I suspect that over time it ends up spending much more time working with the requirements and love for the time spent in the idea will gradually reduce. And I think that the May change news occurs for a long -time horizon.

There is a lot of anguish on what it means for the development of software as a progression. If in 10 years, we don’t give the code for a long time, what does it mean for developers? Will there still be a job for them?

I think your developer’s work looks a lot like an architect. It will be excessively taking big and complex problems in contact and break them down in smaller and solveable tasks. You will have to think how the wider picture of what you are trying to produce, rather than the intermediate language to express it in the machine code.