Google Antigravity Drops Massive Updates: Agent Manager + Skills Unlock Real Agentic Power🛠️

By PiotrMacai | Ainsider | 14 Jan 2026

Google just went all-in on agentic coding with Antigravity - their new agent-first IDE powered by Gemini 3. The latest drops? Agent Manager and Skills support. This isn’t hype - it’s the shift from typing code to orchestrating agents at scale. If you’re building AI apps, agents, or production workflows, this changes everything 🚀

What Is Antigravity?

Antigravity is Google’s agent-first development platform, launched alongside Gemini 3 in late 2025. Forget traditional coding – here you direct AI agents that plan, execute, and iterate autonomously.

It’s built for:

Solo founders shipping fast
Agencies scaling client projects
Devs who want leverage instead of grinding syntax

Runs on Gemini 3 Pro/Flash, integrates tightly with Gemini CLI, and now supports cross-model skills (Claude Code, OpenCode, etc.). Local-first privacy with cloud power when needed. Smart move by Google - they’re betting big on agentic workflows 💯

Latest Updates That Actually Matter 🚨

The big news: Agent Skills are now live + Agent Manager for multi-agent orchestration.

Agent Skills - Packaged Knowledge That Levels Up Your Agents

Skills are an open standard (check agentskills.io) delivered as simple .md files. Drop them in and your agent instantly gains new capabilities.

How it works:

Agent sees skill name + description at start
Decides if relevant → reads full instructions
Follows step-by-step guidance + uses any scripts/resources

Skill locations:

Workspace: .agent/skills/<skill-folder>/
Global: ~/.gemini/antigravity/skills/<skill-folder>/

Each skill folder needs a SKILL.md with YAML frontmatter:

YAML

--name: code-reviewdescription: Reviews code for bugs, style, performance. Use on PRs or refactors.--

Then detailed instructions, checklists, decision trees.

Best practices:

Keep focused – one skill, one job ✅
Write trigger-rich descriptions
Include helper scripts in /scripts/
Add examples and templates

Now works across Gemini CLI, Claude Code, OpenCode. Endless possibilities 🤯

Agent Manager – Orchestrate Multiple Agents Like a Pro

Agent Manager lets you:

Run parallel agents
Share context
Track artifacts
Reduce context switching across workspaces

Perfect for complex projects – one agent researches, another codes, third reviews. Real multi-agent workflows in production.

Rules – Enforce Consistency Like a Boss

Force your agents to always follow your standards. No more “it used tabs instead of spaces” surprises.

Locations:

Workspace: .agent/rules/<rule>.md
Global override: ~/.gemini/GEMINI.md

Example rule file typescript-style.md:

Always use TypeScript. Use functional components in React. Prefer modular files – one component per file. Use Prettier default settings. Never use any type.

Agents read rules at start and obey them on every task. Perfect for agencies enforcing client tech stacks or founders keeping codebases clean.

Workflows – Chain Tasks Into Real Pipelines

Define multi-step processes and let agents execute them sequentially or in parallel.

Locations:

Workspace: .agent/workflows/<workflow>.md
Global: ~/.gemini/antigravity/global_workflows/<workflow>.md

Example full-feature-workflow.md:

1. Research requirements and edge cases 2. Generate updated architecture diagram 3. Implement feature using code-review-pr skill 4. Write unit + integration tests 5. Update README and docs 6. Create PR with clear description

Trigger with one command. Agents hand off context automatically. This is where solo founders ship entire features in hours.

Summary of main core features

Feature Path Use Case Rules .agent/rules/ Code style enforcement Workflows .agent/workflows/ Task chaining Skills .agent/skills/ or ~/.gemini/antigravity/skills/ Reusable expertise Agent Manager Core UI Autonomous planning/execution