Already implemented in claude code and cursor. I think this should definitely be added to antigravity, along with specific skills.
12 Likes
Hello @Damien_Menigaux,
Welcome to AI Forum!
Antigravity is architected as an agent-first platform, meaning the Agent Manager (Mission Control) automatically delegates specific tasks to specialized Sub-agents to optimize performance and context usage.
For example, when a task requires web research or documentation lookup, the system invokes the Browser Subagent; for command-line operations, it uses the Terminal Subagent. This allows the main model (e.g., Gemini 3 Pro) to focus on high-level reasoning while specialized models handle the execution details in parallel.
1 Like
Dude, that is not what sub agents is about. Sub agents is about being able to create them yourself, them all having different models in the background, having specific skills
7 Likes
Think classical functions. We want to be able to write âAI functionsâ, where subagent context is analogous to the stack in function calls. Of course not pure functions, since often we want side effects.
1 Like
Thatâs an excellent distinction. While the current âMission Controlâ handles background task delegation, the ability to architect custom sub-agents with specialized skillsâessentially treating them as programmable AI modulesâwould be a powerful addition to the Antigravity workflow.
We will share these insights with our product team. Thank you.
5 Likes
Absolutely agree, another case is calling subagents in a SKILL, multiple SKILLs can share an agent
2 Likes
Iâll upgrade to the top tier if you implement this custom SubAgents feature btw. Itâs definitely an absolute must.
Furthermore, you should also look into âAgent teamsâ by Claude code. Orchestrate teams of Claude Code sessions - Claude Code Docs
3 Likes
I am still using Claude Code because Antigravity is missing several features: subagents, agent teams, and hooks. Once Antigravity includes these, I will compare them again and may reconsider my decision.
1 Like
Could not agree more.
Have been an Antigravity convert since adopting it in December and I couldnât praise the annotations-based review + artifacts (Plans and Walkthroughs) more.
HOWEVER: you need to be able to specialize different parts of the development workflow into sub-agents/identities, so that you can spec out initiatives into planning documents (persisted ones not the transient ones which the Plan workflow very usefully does) - thatâs a Product agent for example, design/architect them into system components (Architect agent), implement them (Builder agent), integrate/refactor them properly into the codebase and iterate back on any design/architecture deviations discovered during implementation (review workflow with the Architect and Product agents), and finally wrap up the feature doc with any final polishes, notes for future considerations, learnings and insights etc.
Youâre just never going to do this all within a single agent chat / context window - you will pollute it to heck on anything but the most trivial features.
Please, please make sub-agents and agent teams the next major feature to bring into the otherwise amazing Antigravity system. Those separate agent chats in Mission Control need to be aware of one another, have different colors/icons/avatars, be able to be grouped together so that you can see exactly which agent team is working on what, and share a communication/collaboration space (ie: shared chat channel?) where you can see which agent requested what from whom, possibly with different communication patterns set up such as centralized (main orchestrator / hub & spoke), human-in-the-middle (always ask you for permission to collaborate across agents), or free will (collaborate automatically until a given objective is met or guardrail is hit ie: ambiguous requirements, architecture decision needed, testing requested etc.)
4 Likes
If Antigravity Does not bring custom subagents like claude code , I would leave it within a month !
That feature is a must , Imagine the amount of context we could save , and structurize the workflow
4 Likes
I want to second what the others say and put emphasis on the need for Antigravity to offer multi-agent / sub-agents and orchestration capabilities to stay competitive.
I used to code back in the days. I was never good enough for my own patience so I quickly figured out that my forte was to find problems, architect solutions, form talented teams and deliver high impact products and services. So I did. But today, everything that I did before, that relied on the last mile being in the hands of others, on a lengthy human cycle of consuming info, chopping it up, implementing step by step, building tests blah blah is now gone. Except, models drift, models forget, and the best setups actually do what we did for LLMs to become reality (with neural networks) we mimic reality. That means; we need to have a virtual organization based on agents. For the sake of this thread, it does not need to cover the full organization, just the product development part, but it should and could cover all.
We need an AI team, from architect to every type of developer, test and qa, security, deployment and analytics and then back to the top again for product and commercial to evaluate results and pipe new features in at the top. The other guys are building it, and there are several good open source frameworks to do this. Antigravity team, please get this done.
The current agent manager, and its terrible separate box, does not give you the domain separation needed not to make code crashes and git fâŚ. ups.
Hello, is there any plan to support sub agent in Antigravity? Since a lot of popular skills in Claude Code are powered by sub agents, adding this support will make the adaptation more smoothly.
We appreciate the communityâs interest in custom sub-agents. This feature request has been escalated to the relevant internal teams for review. We are committed to sharing further information as our roadmap evolves. Thank you for your continued feedback, which is vital as we strive to improve the Antigravity platform.
5 Likes