Most AI coding tools work the same way. A developer spots a problem, writes a prompt, and the agent executes. It's useful. But the developer still drives every single decision.
Google appears to be rethinking that model entirely.
The company's Jules agent — already available on free and paid tiers through Google AI Pro and Ultra — operates differently than most coding assistants. Rather than requiring developers to wait for output after each prompt, Jules operates asynchronously, running in the background so developers can stay focused on other tasks. During its beta, developers tackled tens of thousands of tasks through Jules, resulting in more than 140,000 publicly shared code improvements.
That asynchronous model already puts Jules ahead of synchronous tools like Cursor and Windsurf. But Google isn't stopping there.
Enter Jitro
The Jules team is working on something new internally. The project, referred to as Jitro, appears to be the next generation of Jules — and it represents a different way of thinking about what a coding agent actually does.
Instead of asking developers to instruct an agent on what to build or fix, Jitro is reportedly designed around high-level goal-setting. Think KPI-driven development: the agent autonomously identifies what needs to change in a codebase to move a specific metric in the right direction.
Rather than telling the agent what to do, a developer would define the desired outcome — lower error rates, better test coverage, improved accessibility compliance — and the agent figures out the path to get there. A dedicated persistent workspace suggests Google envisions Jitro as an ongoing collaborator rather than a one-shot tool.
Why this matters for enterprise teams
This shift has the most immediate value for developers managing large, mature codebases. On a greenfield project, the ability to set goals and step back may feel unnecessary — you already know what you're building. But for enterprise engineering organizations seeking to measure software quality at scale, a goal-setting agent with a persistent workspace changes how they plan.
Incremental improvements in performance, test coverage, and security posture compound over time. Giving an agent a target and letting it work — across a large codebase, over time — is a fundamentally different workflow than the current prompt-and-review cycle most teams are running.
The deeper shift
The move from prompts to goals isn't just a UI change. It reflects a different model of what developers do.
Today, even with powerful agents in the loop, the developer is still functioning as a project manager — writing instructions, reviewing outputs, deciding what comes next. Jitro-style goal-setting changes that dynamic. The developer becomes more like an engineering director: setting strategy, defining success criteria, and reviewing results.
That's a harder transition than it sounds. Writing a good prompt requires precision. But defining a meaningful outcome requires judgment — an understanding of the codebase, the business, and what "better" actually looks like. That's not something an agent can define for you. It also means the developers who understand their systems deeply will get more out of these tools than those who don't.
Timing and context
Google I/O 2026 kicks off May 19, and Jitro is exactly the kind of showcase-ready capability Google would want to announce alongside its broader Gemini ecosystem updates. Whether it ships as described remains to be seen. But the direction is clear.
The current generation of coding agents has made developers faster. The next generation is setting its sights on making them more strategic.
And that's a meaningful distinction.
