Mission
AI / Automation — codename NOVA — makes the studio compound. The team's job is to build the agents, prompts, pipelines, and automated workflows that let a small team operate like a much larger one.
How they work together
Four lanes: ai-engineering (LLM apps, RAG, agent frameworks), ml-engineering (model training, MLOps, production ML), prompts (prompt design and optimization), and automation (workflow orchestration and process automation). The pattern: ai-workflow-designer scopes the system, ai-engineer or ml-engineer builds the core, prompt-engineer or prompt-engineer-advanced tunes the language layer, and automation-architect wires it into the broader stack.
For pure prompt work, prompt-engineer handles iteration and prompt-engineer-advanced handles production-grade systems with safety, evals, and conversation design.
Best practices
- Start with the workflow, not the model. ai-workflow-designer maps the human-AI handoffs before any model gets selected.
- MLOps from day one. mlops-engineer sets up tracking, versioning, and rollback — even on the first model.
- Prompts are versioned artifacts. prompt-engineer treats prompts like code: tested, diffed, rolled back when they regress.
- Cost is a design constraint. Token budgets, model selection, and caching get decided before the system ships.
Do's
- Use ai-engineer for any LLM-app work — RAG, agents, function calling, integrations.
- Pull in ml-engineer for traditional ML (classification, embeddings, recommendation).
- Hand production deployment of models to mlops-engineer.
- Let automation-architect design the integration boundary between agents and the rest of the stack.
Don'ts
- Don't build a chatbot when an automation works. Most "AI features" are just workflows.
- Don't ship a prompt without evals. prompt-engineer-advanced can build the eval harness.
- Don't fine-tune until prompting hits its ceiling. The order is prompt → RAG → fine-tune.
Common tools
Read, Write, Edit, Grep, Glob, Bash — for code and config. WebSearch / WebFetch for model research and API documentation. Task is heavily used by orchestrator agents that delegate to other specialists.
When to call on AI / Automation
LLM features, agent systems, prompt design, ML pipelines, or any process that should run without a human in the loop. If the question is "can we automate this?", start here.