AI Studio · Fundamentals

Agent Development

How agents are designed, configured and built in AI Studio — capability-first, not code-first.

Agent development in AI Studio is configuration-first. An agent is not a script — it is a configuration that binds together a prompt, a set of capabilities, knowledge sources and output contracts. Engineers assemble agents from building blocks, and the platform handles orchestration, retrieval and execution.

Agent Anatomy

Every agent in BIOS™ has five components: a prompt (what it should do), capabilities (what tools it can use), knowledge (what information it can access), an output contract (what format its response must take) and confidence thresholds (when it can act autonomously vs when it needs human review).

sergenta ~ bios
// Agent configuration in AI Studio
const agent = {
  name: "lead-scoring-agent",
  prompt: { ref: "lead_scoring", version: "3.2.0" },
  capabilities: ["web_search", "crm_read", "enrich"],
  knowledge: ["knowledge-hub/default", "memory/leads"],
  output_schema: {
    type: "object",
    properties: {
      score: { type: "number" },
      reasoning: { type: "string" }
    }
  },
  confidence: { auto_approve: 0.85, review: 0.65 }
};

The Development Lifecycle

  • Design — define capabilities, knowledge sources and output contracts in the Agent Builder.
  • Write — author the prompt as a versioned template in the Prompt Registry.
  • Test — run the agent against a test dataset in the Evaluation Harness.
  • Review — submit for review with diffs, evaluation results and reasoning.
  • Deploy — blue-green deployment with instant rollback.
  • Monitor — track performance, quality and cost in the Command Center.

FAQ

Frequently Asked Questions

For most agents, no. The Agent Builder uses a configuration interface — you select capabilities, bind knowledge sources and write prompts. Code is needed only for custom integrations or specialised logic that the platform's built-in capabilities don't cover.

Yes. Agent configurations, prompts and workflows are versioned artifacts. Changes are submitted as versions, reviewed and merged — similar to a code review workflow. Multiple engineers can work on different versions in parallel.

Privacy & Cookies

We use cookies to optimise your experience, analyse performance, and enhance AI-powered features. You can manage your preferences at any time.

Privacy PolicyVersion 2.0