Don't Prompt It. It's Already Working: The Case for Google Jitro and the Age of Intentional AI

The End of the Prompt Era

There is a quiet but seismic shift happening in how software gets built, and it has nothing to do with a new model, a faster chip, or a cleaner interface. It has to do with a question that most AI tools have never had to answer: what should I work on next?

Until now, every AI coding assistant has waited for you. You spotted the bug, wrote the prompt, watched the output, and decided what came next. The human was always in the execution loop. Most AI coding agents work the same way: a developer spots a problem, writes a prompt, and watches the agent execute. It is fast, it is useful, but it still puts the developer in the driver's seat for every single decision. (Source: IMR Press)

Google is now building something that breaks this contract entirely.

What Jitro Actually Is

The company is reportedly building the next generation of Jules, its autonomous coding agent, under an internal project name: Jitro, a parallel effort focused on a completely new version that moves beyond the prompt-and-execute model that defines most coding agents today. (Source: IMR Press)

Project Jitro marks the evolution of the Jules coding agent from a task-executor to an outcome-driven collaborator. By shifting from manual prompting to KPI-driven development, Jitro autonomously identifies and executes codebase improvements, allowing developers to manage high-level goals rather than individual lines of code. (Source: arXiv)

The distinction is more profound than it sounds. The proposed interaction is closer to delegation: you define an outcome, and the system determines the path, intermediate steps, and execution plan. Put simply, it marks a shift from AI as a tool to AI as a self-driving system, with the human role moving from operator to supervisor. (Source: MDPI)

In practice, Jitro is structured around a persistent workspace model. Instead of telling AI what step to take next, you define the result you want and the system plans how to get there automatically. The agent remembers what success looks like instead of just responding to isolated prompts. (Source: PubMed Central)

The Architecture of Intentional AI

To understand why this matters architecturally, it helps to trace the lineage. Jules, Jitro's predecessor, was already a meaningful step beyond its peers. Jules is a Google Labs-born asynchronous agent that integrates with GitHub, clones codebases into Google Cloud virtual machines, and uses AI to fix or update code while developers focus on other tasks. Upon completion, it presents its plan, reasoning, and a diff of the changes made. Birow

By running asynchronously in a virtual machine, Jules stands apart from top AI coding tools like Cursor, Windsurf, and Lovable, which all operate synchronously and require users to watch the output after each prompt. (Source) In the words of Google Labs' Director of Product, Kathy Korevec: "Jules operates like an extra set of hands. You can basically kick off tasks to it, and then you could close your computer and walk away." (Source).The market validated this. During the beta, thousands of developers tackled tens of thousands of tasks, resulting in over 140,000 code improvements shared publicly. (Source)

But Jules still required a prompt for every task. Jitro removes that constraint entirely. Rather than asking developers to manually instruct an agent on what to build or fix, Jitro is designed around high-level goal-setting, KPI-driven development where the agent autonomously identifies what needs to change in a codebase to move a metric in the right direction. Instead of telling the agent what to do, a developer would define the desired outcome, better test coverage, lower error rates, improved accessibility compliance, and the agent figures out the path to get there. (Source: ScienceDirect)

Early signals point to a workspace where developers can list goals, track insights, and configure tool integrations, a layer of continuity that current coding agents do not offer. This marks a departure from the task-level paradigm used by competitors like GitHub Copilot, Cursor, and even OpenAI's Codex agent, all of which still rely on developers defining specific work items. (Source: Hub)

The Numbers Behind the Shift

The timing of Jitro's emergence is not coincidental. The broader AI agent market has hit an inflection point that makes goal-driven architecture both commercially viable and strategically necessary. While 88% of companies now apply AI in at least one area, only 23% run fully autonomous agent systems. (Source) That 65-point gap between "using AI" and "trusting AI to act" is the market Jitro is positioning itself to unlock.

Agentic AI traffic grew 7,851% year over year in 2025, as autonomous systems began navigating the web, completing checkouts, managing accounts, and executing transactions without direct human intervention. (Source) The infrastructure for agentic behaviour already exists at scale. What has been missing is the goal layer, and that is precisely what Jitro introduces.

Agentic AI usage is poised to rise sharply in the next two years, but oversight is lagging: only one in five companies has a mature model for governance of autonomous AI agents (Source). This governance gap is not a minor footnote. It is the central risk Jitro must navigate.

The Category Differentiator

To appreciate just how much of a category break this represents, consider the competitive landscape. Every major coding agent on the market today, Cursor, Windsurf, Copilot, Codex, operates on the same fundamental assumption: the human defines the work unit. The AI executes it. The human reviews. Repeat. Google Jitro is the signal that AI coding agents are moving from task helpers into persistent collaborators that improve repositories continuously instead of waiting for instructions. Persistent workspace agents reduce friction by maintaining awareness across modules instead of reacting only to commands, allowing optimisation strategies to compound naturally across development cycles rather than restarting repeatedly. (Source: arXiv)

This compounds in ways that task-by-task agents simply cannot. An agent that retains context across sessions, tracks its own progress against defined outcomes, and identifies the next highest-leverage improvement autonomously is not just faster. It changes the fundamental role of the developer. Teams begin managing targets instead of managing tasks. Creators begin directing strategy instead of editing outputs. Operators begin supervising systems instead of building checklists manually. (Source: PubMed Central)

Google has been steadily expanding its AI developer tooling through Gemini integrations in Android Studio, Firebase, and Cloud, and a goal-oriented coding agent fits neatly into that strategy, particularly for enterprise teams that care more about outcomes than individual pull requests. The launch is expected to roll out under a waitlist, suggesting Google is taking a measured approach rather than a broad release. (Source: IMR Press)

The Risks Are Real, and Google Knows It

Here is where the story becomes more complex. The very properties that make Jitro a category differentiator, autonomy, persistence, self-directed planning, are also what make it dangerous if deployed without adequate guardrails. The risk is that autonomous goal-pursuing agents introduce unpredictable changes, and trust will be the key barrier to adoption. Hub

This is not a theoretical concern. In 2025, AI agents expanded what individuals and organisations could do, but they also amplified existing vulnerabilities. Systems that were once isolated text generators became interconnected, tool-using actors operating with little human oversight. Anthropic disclosed how its Claude Code agent had been misused to automate parts of a cyberattack, illustrating a broader concern: by automating repetitive, technical work, AI agents can also lower the barrier for malicious activity. The scale of this risk is now quantifiable. Autonomous agents outnumber humans by an 82:1 ratio in enterprise environments, and a single forged command can start an automated disaster. Oracle In November 2025, Anthropic disclosed that a Chinese state-sponsored cyberattack had leveraged AI agents to execute 80 to 90 percent of the operation independently, at speeds no human hackers could match. Wollenlabs

The regulatory environment is hardening in response. The EU AI Act went into enforcement in late 2025, transforming AI governance from an abstract concept to a concrete procurement requirement, introducing a four-tier risk classification system. Key compliance deadlines are August 2, 2026 for core requirements, with penalties up to 35 million euros, or 7 percent of global turnover. ScienceDirect The US introduced 59 AI-related regulations in 2024, double the number from 2023. ScienceDirect

Why Approval Checkpoints Are Not Optional

Google appears to have internalised these risks structurally, rather than treating oversight as an afterthought.

• The early indication is that Jitro will not be a fire-and-forget system. It looks more like a structured workflow: set a goal, review the agent's approach, approve the direction, which puts enough guardrails around autonomy to make it practical for real teams. (Source: IMR Press)

• Jules already surfaces its plan and reasoning before making changes, and developers can steer the work mid-execution. If Jitro inherits that transparency layer and extends it into goal-level visibility, it could give teams more confidence in what the agent is doing and why. (Source: ScienceDirect)

• The clearest framing of what genuine oversight requires comes from the clinical research community. The practical issue is whether checkpoints deliver real inspectability, clear plans, test evidence, and an intelligible mapping from goal to code edits, rather than a single yes/no gate at the end. Without legible rationales and meaningful validation, "human-in-the-loop" can become performative, particularly in large codebases where no one can realistically scrutinise everything. (Source: MDPI)

• True governance makes oversight everyone's role, embedding it into performance rubrics so that as AI handles more tasks, humans take on active oversight rather than passive approval. (Source)

This is the design challenge Jitro must solve. Approval checkpoints that are too frequent recreate the prompt fatigue they were designed to eliminate. Checkpoints that are too sparse reduce humans to rubber stamps. The architecture of trust is as important as the architecture of execution.

The Bigger Shift This Signals

Jitro is not just a developer tool. It is an early signal of what professional work looks like when AI systems hold context, pursue goals, and report back rather than waiting to be instructed. In a prompt-based workflow, the human breaks the work into steps and continuously steers. In a goal-driven workflow, the system decomposes the work on its own, and you assess the plan, the edits, and the evidence that the goal has been met. The difference is not merely speed. It is a change in who performs task decomposition. MDPI

That is a meaningful transfer of cognitive labour. And as with every such transfer in the history of technology, from assembly lines to spreadsheets to cloud infrastructure, the leverage goes to those who learn to operate at the new level of abstraction fastest. The most constructive stance is neither dismissal nor enthusiasm, but disciplined curiosity: if goal-driven agents are becoming engineering teammates, we need supervision science to match. That includes studying which checkpoint designs actually reduce error, how to quantify drift in agent-modified pipelines, and how to preserve interpretability when plans are generated by systems optimised for throughput. MDPI

Google I/O 2026 kicks off May 19, and Jitro is exactly the kind of showcase-ready feature Google would want to unveil alongside its broader Gemini ecosystem updates. Whether it ships on that timeline or not, the direction is set. The era of prompting AI is ending. The era of supervising it has begun.

----------------------------------------------------------------------------------------------

References:

1. Google's Next Coding Agent Could Change How Developers Think About Their Work — DevOps.com, April 2026 https://devops.com/googles-next-coding-agent-could-change-how-developers-think-about-their-work/

2. Google Jitro AI Agent Is Google's Biggest Agent Move Since Jules Launched — Julian Goldie, April 2026 https://juliangoldie.com/google-jitro-ai-agent/

3. Google Jitro Changes How Agencies Build Software With AI — Goldie Agency, April 2026 https://goldie.agency/google-jitro/

4. Google Tests Jules V2 Agent Capable of Taking Bigger Tasks — TestingCatalog, April 2026 https://www.testingcatalog.com/google-prepares-jules-v2-agent-capable-of-taking-bigger-tasks/

5. Google's AI Coding Agent Jules Is Now Out of Beta — TechCrunch, August 2025 https://techcrunch.com/2025/08/06/googles-ai-coding-agent-jules-is-now-out-of-beta/

6. Revolutionizing Coding: Google's Next-Gen Autonomous Coding Agent — TechCloudUp, April 2026 https://techcloudup.com/2026/04/news-revolutionizing-coding-googles-next-gen-autonomous-coding-agent/

7. Google's JITRO and the Clinical Logic of Goal-Driven AI — Dr. Rania Kassir, April 2026 https://happybraintraining.com/googles-jitro-and-the-clinical-logic-of-goal-driven-ai-when-systems-stop-waiting-to-be-prompted/

8. AI Agents Arrived in 2025: Here's What Happened and the Challenges Ahead in 2026 — The Conversation, January 2026 https://theconversation.com/ai-agents-arrived-in-2025-heres-what-happened-and-the-challenges-ahead-in-2026-272325

9. The Future of AI Agents: Trends and Predictions — MindStudio, January 2026 https://www.mindstudio.ai/blog/future-of-ai-agents

10. The 2026 State of AI Traffic and Cyberthreat Benchmark Report — HUMAN Security, March 2026 https://www.humansecurity.com/learn/resources/2026-state-of-ai-traffic-cyberthreat-benchmarks/

11. The State of AI in the Enterprise 2026 — Deloitte US https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html

12. 2026 Predictions for Autonomous AI — Palo Alto Networks, November 2025 https://www.paloaltonetworks.com/blog/2025/11/2026-predictions-for-autonomous-ai/

13. How 2026 Could Decide the Future of Artificial Intelligence — Council on Foreign Relations, January 2026 https://www.cfr.org/articles/how-2026-could-decide-future-artificial-intelligence