How to make the Google Dev tool manager make artificial intelligence coding

As the Director of the Google Project for Developers, the Ryan Salva has a seat in the front row of the methods of the AI ​​tools to change coding. Previously from GitHub and Microsoft, he is now responsible for tools like Gemini Cli and Help the Gemini symbolPaying the developers to a new world of the agent.

Release New third party research On Tuesday, it shows how developers already use artificial intelligence tools – how much is the remaining progress. I sat with a predecessor to talk about the report and his personal experience with artificial intelligence coding tools.

This interview was released for length and clarity.

Every year, Google performs a survey of developers trends – but this year’s report really focuses on artificial intelligence tools, specifically how developers are ready to obtain their programming approach. Was there anything in the search surprised you?

One of the really interesting results was the average date when developers began to use artificial intelligence tools. They found that it was April 2024, which is accurately corresponding to Claude 3 Exodus and Yimoen 2.5 coming out. This is in fact the detonation of examples of thinking or thinking, and about the same time, we got much better in calling tools.

For coding tasks, you really need to be able to take advantage of external information in order to solve problems, so you may need Grep, you may need to assemble the code. If the code collects it, he may want to run this unit test, and this integration test. I think calling tools is really the important piece that gave models the ability to automatically correct while moving.

How to use artificial intelligence coding tools personally?

Most of my distinction these days are intended for hobby projects, and I spend most of my time using the orders line. So that Gemini includes Cli. Then there is a little Claude icon, a little manuscript there. You never use the peripheral tool in itself, so I am really heterogeneous about the IDES I use. I use Zed. I use Vs Code. I use the indicator. I use Windsurf, all of them, because I am only interested in seeing how the world works and how the industry develops.

On the professional side, product managers tend to live in documents, so the first thing is to use artificial intelligence to help write specifications and requirements.

I am curious about how it works. You use Gemini Cli to build Gemini Cli, but I imagine that it is not just running.

The development task usually begins as an issue, perhaps the GitHub issue is dropped by someone. Often, if you are truly sincere, it is a somewhat unlimited issue. So I will use Gemini Cli to create more powerful requirements in Markdown. This usually creates about 100 lines of technical specifications to some extent, but also the specifications that depend on the results. Then I will use Gemini Cli to write the code based on this general specifications and preferences in team documents.

Through the engineering team, we have some different layers of the rules and discount documents that the model consumes, just put our way of work: here is how to test, here is how to manage the consequences, and so on. So when the symbol produces, it also works from these documents.

While Gemini Cli passes the exploring and repairing errors, I will update my requirements, “I have fixed this step. Now I am on the next step,” and so on. Each of these creates a request for commitment and withdrawal in the warehouse, so that I can always weight or retreat.

I would like to say perhaps from 70 % to 80 % of my work is to work at the station in the natural language, try to use Gemini Cli to formulate the requirements, then allow Gemini Cli to write most of the code for me, which I will then review and read with any IDE that I use. But I mostly use the IDE as a place to read the code, instead of writing the code.

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

For three decades, IDE was where we went to do everything in software development. You had an IDE, you had the browser, and you had a peripheral window.

I think this is still the case greatly, but I think that over time, we will finally finish spending more time working with the requirements, and the amount of time you spend in IDE will gradually shrink. And I think this change may actually happen within a very long horizon.

There is a lot of concern about what this means to develop software as progress. If 10 years from now, we no longer look at code, so what does this mean for developers? Will there be a job for them?

I think your job as a developer will look more like an architect. It will be related to big and complex problems and breaking them into smaller tasks. You will need to think about the biggest image about what you are trying to produce, instead of the intermediate language to express this in the device’s symbol.