Khaberni - A Google engineer revealed an astonishing experiment with an AI-powered coding tool that sparked a wide debate about the future of software development and the capabilities of modern technologies in this field.
Jana Dugan, the lead engineer at Google working on the Gemini API, shared her experience with the "Cloud Code" tool from Anthropic Company.
Dugan explained that she gave the tool a description of a complex programming issue, and within one hour, it produced a result that closely matched what her team at Google had been working on throughout the past year.
The task involved building distributed agent coordinators, systems that coordinate the work of multiple AI agents, and Jana Dugan noted that the Google team had tried multiple approaches to this problem without reaching a final design that was agreed upon.
Details of the experiment and results
Dugan wrote on her X platform post: "I'm not kidding and it’s not funny, we have been trying to build distributed agent coordinators at Google since last year, there are diverse options, and not everyone agrees on one direction".
She added that she gave "Claude Code" a description of the problem, and the tool built what the team did last year in just one hour.
The engineer clarified that the instructions she provided were not very detailed, but were limited to just 3 paragraphs because she could not use Google's internal information, she prepared a simplified version of the problem using already existing ideas to test "Cloud" performance.
Realistic evaluation of the results
Dugan acknowledged that the result was not perfect and still needed improvements, yet she encouraged skeptics of AI coding agents’ capabilities to try it in areas where they already have deep expertise.
Dugan responded to a query about Google's use of "Cloud", affirming that its use was permitted only in open-source projects and not for internal work. When another user asked when "Gemini" would reach the same level, she replied: "We are now working hard on models and tools".
A visionary outlook on the future of AI
Dugan emphasized that developing AI is not a zero-sum game, and she expressed the logic of acknowledging strong work from competitors, saying: "Cloud Code is indeed impressive, I feel excited and empowered to push us all forward".
The engineer described the astonishing speed of AI-supported programming evolution over the years, saying: "In 2022, systems were completing individual lines of code, in 2023, they were able to handle entire sections, by 2024, they could work across multiple files and build simple apps, by 2025, they were capable of creating and restructuring complete codebases".
She added that in 2022 she did not believe in the possibility of reaching the accomplishments of 2024 as a global product for developers, in 2023, she thought today's capabilities were still 5 years away, and wrote: "The gains in quality and efficiency in this field exceed what anyone could have imagined".
Moreover, the post garnered more than 7 million views and a wave of reactions, where one user commented: "It's amazing how AI surpasses the bureaucracy that slows us down in large institutions, which could spark a renaissance in individual creativity".
While another wrote: "Cloud Code is really impressive and more developers will recognize this. The speed of code generation and feature development will increase at a different pace".
