Google has just released major upgrades to its AI coding agent Jules — namely, a command-line interface (CLI) and a public API that let developers call Jules from terminals, CI/CD systems, Slack, and other tools. This move shifts Jules from being a chat-style agent to a more deeply embedded aspect of the software workflow, as Google seeks to make the tool more flexible and context-aware. In announcing this, Google emphasized that Jules is meant for well-scoped tasks (versus more open collaboration modes in Gemini CLI) and is designed to “pause and ask” when it hits unsolvable issues. Google’s own blog describes Jules Tools and the API as means to reduce context switching and let users integrate Jules into existing development pipelines. Beyond that, Jules now supports memory (tracking user interactions over time) and has introduced new UI features like diff viewers and image upload. Pricing tiers include a free plan and paid tiers for heavier use. Meanwhile, competition in AI coding agents is ramping up, with companies like Salesforce launching “vibe-coding” tools and OpenAI, Anthropic, Microsoft, and others pushing agentic coding platforms.
Sources: VentureBeat, Google Blog
Key Takeaways
– Jules is now more tightly embedded via CLI and API, enabling integration into real developer toolchains rather than just chat UI.
– Google positions Jules for well-defined, scoped tasks, distinguishing it from more collaborative or exploratory AI tools.
– The AI coding assistant market is heating up, with rivals from OpenAI, Anthropic, Microsoft, and others releasing competing agentic development tools.
In-Depth
Google’s move to expand Jules beyond just a website interface into terminals, continuous integration / continuous deployment (CI/CD) pipelines, and developer tools marks a notable shift in how AI coding agents are evolving. Previously, Jules functioned like a smart assistant that you’d call via web or GitHub, but this update gives it hooks into the very environments where developers spend their time. Google says it wants to minimize context switching: instead of bouncing between browser tabs or separate UIs, developers can let Jules live in their terminal, Slack, or in pipelines via its new API. This is no small change — it suggests that Google anticipates the future of coding will be more “agentic,” with smart assistants executing tasks under control rather than merely responding to prompts.
The tech is backed by Google’s Gemini 2.5 Pro model, which also powers the broader Gemini CLI. But Google draws a distinction: Jules is designed for narrower, specific tasks, whereas Gemini CLI is built for more interactive, iterative collaboration. In practice, that means Jules could autonomously carry out things like bug fixes, version bumps, tests, pull request edits — but if it gets stuck, it knows when to ask for help. The memory feature also raises the bar: Jules can retain user preferences, past interactions, corrections, and nudges. On the UI side, Google has added features like improved diff viewers and support for image upload, helping Jules to understand richer context.
From a business perspective, Google is positioning Jules as a platform: the API lets organizations wire it into Slack notifications, trigger tasks automatically, extend it into IDEs, or even hook it into non-Git systems in the future. While the CLI and API are available now, Google plans to build more IDE plug-ins, broaden version control support, and explore non-Git use cases.
Of course, Jules is not entering a quiet market. Rival firms are pushing their own tools. Salesforce unveiled “Agentforce Vibes,” a vibe-coding product built into its stack, offering natural language to code translation for internal developers. OpenAI is continuing to push agentic development via its Codex tools. Anthropic is releasing models like Claude Sonnet 4.5, aimed at intelligent agent and coding performance. Microsoft is embedding AI via “Agent Mode” features in Office and development tools. Analysts and industry watchers see this as a turning point in software engineering: coders may increasingly supervise AI agents rather than writing every line by hand. But challenges remain — tool reliability, code correctness, security, guardrails, and developer trust are all in the mix. Leading benchmarks such as SWE-PolyBench show that coding agents still struggle on more complex, cross-repository tasks, especially in languages and scenarios they haven’t been fine-tuned for. As Google doubles down on Jules, its success may depend less on raw model power and more on how well it integrates, how safely it acts, and how much confidence developers place in its output.