# Subagents Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/subagents
> Markdown URL: https://aitinkerers.org/technologies/subagents.md
> Technology record last updated: 2026-02-28T17:13:14Z
> Generated: 2026-09-21T04:43:23Z

Subagents are specialized, autonomous AI assistants: they execute delegated, focused tasks with isolated context windows and custom tool access, maximizing efficiency.

Subagents function as a dedicated AI task force, operating within a larger orchestrator like Anthropic's Claude Code (or Spring AI). Each subagent is a separate instance, running in its own context window with a unique system prompt and specific tool permissions: for example, a 'Read-only' agent for code exploration versus a 'Write' agent for implementation. This architecture delivers three key advantages: context preservation, specialization, and parallel processing. By delegating tasks—like a security audit or a unit test run—to a focused subagent, you keep the main agent's context clean and leverage specialized expertise, often routing to cost-effective models like Haiku for simple lookups. This setup enables high-throughput workflows, allowing multiple agents to run concurrently for rapid refactoring or complex, multi-step development projects.

- Official technology site: https://docs.anthropic.com/en/docs/claude-code/sub-agents
- Public AI Tinkerers demos and talks: 2
- Result page: 1 of 1

## Recent Public Talks and Demos

### [Jules: Uma Raposa Que Toca o Meu Negócio Inteiro no Claude Code](https://saopaulo.aitinkerers.org/talks/rsvp_dm9xlYot2e4)

AI collaborator system on top of Claude Code that helps run my whole business and a live recording studio, Jules Live Studio, that turns it into a real two-way conversation with my AI cohost Jules (a fox). The system has two layers worth showing builders: The skills layer. - Custom workflow skills — /think, /dialogue, /write, /research, and a routing rule that picks between them — force the model through structural gates instead of letting it default to the two failure modes every LLM has: enthusiastically agreeing with whatever I just said, and anchoring on its first answer instead of asking what I actually think. Each skill is a small workflow with explicit phases — elicit, challenge, adversarial review, only-then-recommend — that the model cannot skip. The studio layer. - Jules Live Studio is a local recording app that wires the skills layer into a live voice and video loop: speech-to-text in with Whisper, Claude reasoning, ElevenLabs voice out, fox avatar on screen, and OBS recording. The audience watches a real conversation, not a terminal scrollback. For an AI Tinkerers audience, the demo focuses on the parts that matter to builders: - how the routing layer sends different kinds of requests to different workflows - how a skill structurally forces the model to push back instead of agree - how memory and feedback loops keep it useful across sessions - how the live voice/video loop is wired together with off-the-shelf parts - how far you can get without piling on heavyweight agent frameworks

- Event context: AI Tinkerers SP - Meetup de Maio - Kiro & AWS — 2026-05-28 — São Paulo
- Public talk page: https://saopaulo.aitinkerers.org/talks/rsvp_dm9xlYot2e4

### [Four Design Patterns for Claude Code](https://raleigh.aitinkerers.org/talks/rsvp_6kQ75AJRcmQ)

I'd like to share my Claude Code Design Patterns project with people. It has four handy patterns people can use to create more effective and powerful Claude Code setups. Each pattern has documentation, a demo, and even a teach-claude.md to bring it into your setup. https://github.com/adamrdrew/claude-code-patterns

- Event context: AI Tinkerers Raleigh Meetup — February 11, 2026 — 2026-02-11 — Raleigh
- Public talk page: https://raleigh.aitinkerers.org/talks/rsvp_6kQ75AJRcmQ

## Related Technologies

- [Claude Code](https://aitinkerers.org/technologies/claude-code) ([Markdown](https://aitinkerers.org/technologies/claude-code.md)) — 214 public demos
- [Hooks](https://aitinkerers.org/technologies/hooks) ([Markdown](https://aitinkerers.org/technologies/hooks.md)) — 2 public demos
- [CLI](https://aitinkerers.org/technologies/cli) ([Markdown](https://aitinkerers.org/technologies/cli.md)) — 9 public demos
- [Codex realtime-audio-2](https://aitinkerers.org/technologies/codex-realtime-audio-2) ([Markdown](https://aitinkerers.org/technologies/codex-realtime-audio-2.md)) — 1 public demo
- [GitHub](https://aitinkerers.org/technologies/github) ([Markdown](https://aitinkerers.org/technologies/github.md)) — 74 public demos
- [Markdown](https://aitinkerers.org/technologies/markdown) ([Markdown](https://aitinkerers.org/technologies/markdown.md)) — 24 public demos
- [Skills](https://aitinkerers.org/technologies/skills) ([Markdown](https://aitinkerers.org/technologies/skills.md)) — 10 public demos
- [TTS](https://aitinkerers.org/technologies/tts) ([Markdown](https://aitinkerers.org/technologies/tts.md)) — 5 public demos