Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

FAQ

Frequently Asked Questions


General

Q: What is Agentic Engineering?

A: Agentic Engineering is a software development methodology where AI Agents play a central role in the entire engineering lifecycle - from design to deployment to operations.

Q: Why 1000 Agents?

A: 1000 is a sweet spot that provides:

  • Visual impact and clear communication
  • Actual productive capacity
  • Manageable complexity
  • Cost-effective scale

Q: Can I run this locally?

A: Yes! The MVP can run on a single machine with Docker. Full 1000-Agent scale requires a Kubernetes cluster.


Technical

Q: What LLM models are supported?

A: The platform is model-agnostic. Currently optimized for:

  • Alibaba Cloud: Qwen3.5-Plus, Qwen3-Max
  • OpenAI: GPT-4, GPT-4-Turbo
  • Google: Claude, Gemini

Q: How do Agents communicate?

A: Agents communicate through:

  • Task queues (for work assignment)
  • Shared state (in file system)
  • Direct messaging (via Orchestrator)

Q: What happens when an Agent fails?

A: The Cage health monitor detects the failure and:

  1. Saves current state to persistent storage
  2. Attempts automatic recovery (restart)
  3. If recovery fails, escalates to human operator
  4. Reassigns pending tasks to other Agents

Cost

Q: How much does it cost to run 1000 Agents?

A: Approximately $464,000/month, broken down as:

  • Compute: $285,000
  • Tokens: $144,000
  • Storage: $15,000
  • Overhead: $20,000

Q: Can I reduce costs?

A: Yes! Strategies include:

  • Use spot instances (60-70% savings on compute)
  • Reserved capacity (30-40% savings)
  • Token budgeting (20-30% savings)
  • Idle detection and scale-down (15-25% savings)

Security

Q: How is data isolated between Agents?

A: Each Cage has:

  • Dedicated Kubernetes namespace
  • Isolated persistent storage
  • Network policies restricting cross-Cage access
  • Separate service accounts and credentials

Q: Can Agents access external systems?

A: Yes, but with strict controls:

  • Whitelisted external APIs only
  • Rate limiting and quotas
  • Audit logging of all external calls
  • Human approval for sensitive operations

Migration

Q: How long does mono-repo migration take?

A: Based on 10-repo experiment:

  • Analysis: ~2 hours per repo (parallel)
  • Planning: ~1 day for 400 repos
  • Execution: ~2-4 weeks (phased approach)

Q: What’s the risk of migration?

A: Key risks and mitigations:

  • Build complexity: Gradual migration, parallel testing
  • Agent coordination overhead: Layered scheduling, batch processing
  • State consistency: File locks + transaction logs
  • Human acceptance: Gradual automation,保留 approval points