For a while, Google had too many coding tools and no clear story. Code Assist here, Gemini CLI there, Jules off to the side. In 2026 that changed: the company pulled everything under one agent-first banner called Antigravity, and the picture finally makes sense. If you write software — or you are a non-developer who wants to ship it — this is the stack worth understanding.
Let me separate the three tools people conflate, because they do genuinely different jobs.
Antigravity: the platform
Antigravity is Google's agentic development platform. The shift it represents is from autocomplete — AI suggesting the next line while you type — to agents that plan and execute multi-step work across a whole project. At I/O 2026, Google expanded it well beyond an IDE: a revamped desktop app, a new command-line interface, and an SDK so teams can build their own custom agents on top. The message was clear — Google wants Antigravity to be where agentic software gets built.
Gemini CLI becomes Antigravity CLI
If you used Gemini CLI, here is the change that matters. Google is transitioning it into the new Antigravity CLI, rebuilt in Go for speed. The migration has real dates: for consumers, Gemini CLI and the Gemini Code Assist IDE extensions stop serving requests on 18 June 2026 for Google AI Pro and Ultra and for free Code Assist users. Organisations on Gemini Code Assist Standard or Enterprise licences keep access. The good news — the features people relied on, including Agent Skills, Hooks, Subagents and Extensions, carry over, now implemented as Antigravity plugins.
Jules: the agent you delegate to
Jules is the one that changes how a workday feels. It is an asynchronous coding agent: you give it a task, and it spins up a cloud virtual machine, clones your repository, writes a plan with Gemini, runs your tests, and opens a pull request — while you do something else entirely. Antigravity is where you work interactively; Jules is where you hand off a self-contained job and come back to a reviewed PR. Used together, one developer covers a lot more ground.
Where the models fit
All of this runs on Gemini. For heavy reasoning across a large codebase, Gemini 3.1 Pro and its long context do the work; for fast, cheap iterations, the Flash models keep costs down. On the API, you pick per call — which is the discipline that keeps an agentic workflow affordable: use the cheapest model that clears the bar, escalate only when you must.
Antigravity vs the rest
Antigravity competes with the agentic-coding tools our builders already use — Claude Code, Cursor and the like. The honest take: the ecosystem you are in matters more than small feature gaps, and the workflow skills transfer between them. If your stack is Google-centric, Antigravity is the natural home; if it is not, the same agentic habits apply wherever you work. That transferability is the point — you are learning a way of building, not a single vendor's buttons.
You do not have to be a developer
The quiet story under all this is that the barrier to shipping software keeps falling. This is vibe coding: you describe what you want, the agent handles the syntax, and a non-developer can ship a working tool. What you still need is judgement — how to scope a task, review a pull request, and steer an agent that is confidently wrong. That is a learnable skill, and it is where training pays off. It pairs naturally with the design tools in Stitch, AI Studio and Opal and, for production systems, ADK and Vertex AI.
Getting your team building
Agentic coding is the highest-leverage skill on this list — it turns intent into working software. AITraining2U's hands-on AI Vibe Coding and AI Engineering programmes teach the workflow that transfers across Antigravity, Claude Code and the rest, and both are HRD Corp SBL-KHAS claimable for eligible Malaysian employers — details in the HRDC AI training guide. The full stack is mapped in Google's AI productivity stack.
