What You'll Learn
How to use Antigravity's chat-based agents and Manager View to break a complex project into parallel tasks, delegate each to an autonomous AI agent, and review their work through verifiable artifacts — turning hours of manual effort into minutes of supervised automation.
Why This Matters
A single AI prompt can only take you so far. When a task has many distinct steps — research, then analysis, then writing, then review — you either have to manually copy outputs from one prompt to the next, or compress everything into one unwieldy instruction that the AI struggles to follow well.
Multi-agent workflows solve this by splitting the work. In Antigravity, you describe what you want in a chat panel, and the AI agent autonomously plans the work, writes code, runs commands, and verifies results. The Manager View takes this further — you can spin up multiple agents working on different tasks simultaneously, like having a team of specialists each handling their part of a project in parallel.
Step-by-Step Guide
Step 1: Define your project as a set of parallel tasks
Before opening Antigravity, write out the distinct pieces of work your project needs. Think about which tasks can run independently and which depend on others. Tasks that don't depend on each other can run in parallel using the Manager View.
Example project — Competitive Research Report:
1. Research the top 5 competitors of [company name] (independent)
2. Build a data comparison spreadsheet of their features and pricing (depends on 1)
3. Write a 500-word executive briefing summarising findings (depends on 2)
Or, as parallel tasks:
- Agent A: Research competitors and extract pricing/features/audience data
- Agent B: Write a report template with sections for findings, comparisons, and recommendations
- Then combine: Feed Agent A's research into the template Agent B created
Step 2: Open Antigravity and start a project
Download Antigravity from antigravity.google if you haven't already. Open it — it looks like VS Code with a chat panel on the side. Sign in with your Google account and open a project folder (or create a new one).
Step 3: Start with a single agent on the first task
Type your first request into the chat panel. Use Planning mode (the default) so the agent shows you its plan before executing. Be specific about the format you want.
Chat prompt:
Research the top 5 competitors of [company name]. For each competitor, create a
markdown file with these sections:
- Company name
- Pricing (free / freemium / paid, with specific prices if available)
- Top 3 features
- Primary target audience (one sentence)
Save each competitor as a separate file in a /research folder.
The agent will create a plan, show it to you for review, then execute — creating files, running searches, and delivering structured output. Review the artifacts (plans, files, screenshots) before moving on.
Step 4: Use the Manager View for parallel tasks
Click the Manager View icon to open the multi-agent dashboard. Here you can spawn additional agents, each working in their own workspace on a different task simultaneously.
Agent 1 (already running): Researching competitors
Agent 2 (new): "Create a professional report template in report-template.md with
sections for: Executive Summary, Competitor Comparison Table, Key Differentiators,
Pricing Gaps, and Strategic Recommendations. Leave placeholder markers like
{{COMPETITOR_DATA}} where research findings will be inserted."
Both agents work independently and in parallel. You can monitor each agent's progress from the Manager View dashboard.
Step 5: Combine the results
Once both agents finish, use the chat panel to bring their outputs together:
Chat prompt:
Read the competitor research files in /research and the report template in
report-template.md. Fill in the template with the research data:
- Build a comparison table from the competitor files
- Identify common features, unique differentiators, and pricing gaps
- Write 3 specific strategic recommendations based on the analysis
- Fill in the executive summary last, summarising the whole report
Save the final report as competitive-analysis.md.
Step 6: Review, refine, and iterate
Review the agent's output by reading the generated files. If something needs improvement, give feedback directly in the chat — the agent has full context of everything it produced and can make targeted adjustments.
Chat prompt:
The recommendations section is too generic. Make them more specific — each
recommendation should reference a specific competitor's weakness and suggest
a concrete action our team could take within the next quarter.
Tips for Better Results
- Use Planning mode for complex tasks. It shows you the agent's plan before execution, so you can catch misunderstandings early. Switch to Fast mode only for simple, well-defined tasks.
- Define Rules for consistency. Set project-level rules like "always output in markdown" or "use British English" — the agent will follow them across all tasks without you repeating yourself.
- Keep parallel agents independent. The Manager View works best when each agent's task doesn't depend on another agent's output. Plan your task breakdown so parallel agents don't block each other.
- Review artifacts, not just files. Agents produce task lists, implementation plans, and screenshots as artifacts. Checking these helps you verify the agent's reasoning, not just its output.
- Use Google AI Studio for prompt refinement. If an agent isn't producing the right output, paste the prompt into Google AI Studio to iterate quickly on the wording before bringing it back to Antigravity.
Tools That Work Best for This
- Antigravity — The primary environment for delegating tasks to autonomous AI agents. The chat panel handles single-agent tasks; the Manager View enables parallel multi-agent workflows.
- Google AI Studio — Useful for refining individual prompts in a lightweight playground before using them in Antigravity.