> ## Documentation Index > Fetch the complete documentation index at: https://explore.airia.com/llms.txt > Use this file to discover all available pages before exploring further. # Agent Development Through Natural Language # Agent Development Through Natural Language > Use natural language to go from idea → working agent → deployed experience in minutes. Airia’s **Build Agents with AI** capability lets you describe what you want in plain language while an AI-powered builder plans, assembles, and iterates your agent directly in **Agent Studio**. At a high level, the flow looks like this: 1. **Describe & Plan** – Tell the AI what you’re trying to achieve in plain language. 2. **Build & Iterate** – Watch the AI assemble the workflow on the visual canvas and refine it with you. 3. **Test & Deploy** – Try it, fix it, and ship it—using the same AI to improve what you already have. *** ## How it Works ### 1. Describe & Plan Thumbnail Start by describing your use case in natural language: * Your **goal** (e.g., “Qualify inbound leads and create tickets in Jira.”) * Any **systems** it should connect to (e.g., Jira, Slack, your internal APIs) * Relevant **constraints** (e.g., handoff rules, approval steps, routing logic) The Airia Agent will: * Ask clarifying questions when needed * Provide **feedback** on your idea (feasibility, gaps, edge cases) * Generate a **structured plan** for the agent workflow tailored to your use case Example prompts: * “I want an agent that triages customer support requests and routes them by priority.” * “Help me design an agent that collects product feedback and creates Jira tickets.” *** ### 2. Build & Iterate Thumbnail Once you approve (or adjust) the plan, the Airia Agent: * Autonomously creates and configures steps on the **visual canvas** in real time * Adds **AI model blocks**, **tool actions**, **routing logic**, and **memory** * Can incorporate **Python steps** to support more advanced behaviors Typical capabilities it can add for you: * AI model integration (OpenAI / Anthropic models preconfigured for your project) * Tool actions (e.g., Jira, Slack, HTTP calls, internal APIs) * Branching and routing logic * Custom Python steps * Memory and context configuration You can then iterate using the same natural language chat: * “Add a fallback branch if Jira is unavailable.” * “Log every error into this data source.” * “Split high-priority issues into a separate path.” > ℹ️ **Note**\ > While the **Build with AI** chat is building or editing your workflow, the canvas is temporarily locked for manual edits to avoid conflicting changes. *** ### 3. Test & Deploy Once the agent is built: 1. **Test** the agent directly from Agent Studio using your own example inputs. 2. **Ask the AI to review** and suggest improvements: * “Suggest areas of improvement for this agent design.” * “How can I improve my existing prompts for better response quality?” 3. **Deploy** the agent once you’re satisfied with how it behaves. You can also use the Airia Agent to **improve existing agents**: * Review routing and error handling * Suggest prompt improvements * Propose additional steps or guardrails * Troubleshoot failed executions and offer fixes *** ## Why This Matters for Enterprise Teams Enterprise teams still struggle to turn business goals into reliable AI agents: * Business objectives are **hard to translate** into agent workflows. * Many agents end up being **poorly structured** and difficult to maintain. * Teams depend heavily on **specialized AI engineers** for even small changes. * Non-technical users are blocked or forced to work through long handoff cycles. Airia’s natural-language agent development is designed to solve these problems by: * **Removing the steep learning curve** of new AI platforms * **Automatically structuring reliable workflows** that follow Airia best practices * **Reducing dependency on specialized AI developers** for day-to-day work * **Empowering business users** to create and adjust agents without deep technical expertise *** ## Availability & Access ### Feature Name * Tenant setting: **Build Agents with AI** * In Agent Studio: **Build with AI** chat ### Current Access Model * The feature is **disabled by default** at the tenant level. * Only the following roles can **enable or disable** the tenant setting: * **Platform Admin** * **Admin** * Once enabled, **any user who can access Agent Studio** can use **Build with AI** in draft mode (subject to their existing Agent Studio permissions). Thumbnail > ⚠️ **Important** > The **Build Agents with AI** tenant toggle can only be changed by **Platform Admin** and **Admin** users. *** ## Enabling Build Agents with AI for a Tenant > 🔐 **Permissions** > Only **Platform Admin** and **Admin** users can enable or disable **Build Agents with AI** for the tenant. To enable the feature: 1. Go to **Settings** in the Airia platform. 2. Open the **Build with AI** page. 3. Turn on the toggle **Build Agents with AI**. Once enabled, this will: * Activate the **Build and Edit with AI** chat in **Agent Studio** for your tenant. * Allow any user who can access **Agent Studio** to use **Build with AI** in **draft mode**, subject to their existing Agent Studio permissions. *** ## Using Build Agents with AI in Agent Studio ### Draft-Only Editing The **Build with AI** chat is available **only in draft mode**: * ✅ You **can** build and edit **draft** versions of agents. * ❌ You **cannot** modify **published** versions via the AI chat. To edit an existing published agent: 1. Open the agent in Agent Studio. 2. Switch or navigate to its **draft** version. 3. Use the **Build with AI** chat from there. *** ### Starting the Build with AI Chat 1. Navigate to **Agent Studio**. 2. Open the agent you want to build or improve (draft). 3. Use the **Build with AI** chat, available from the **left navigation controls** on the canvas. From there, you can: * Start with a **new idea** and ask the AI to design the workflow. * Ask questions about your **existing configuration**. * Request **changes or improvements** to the current design. *** ### What the Airia Agent Understands The Airia Agent is **domain- and platform-aware**. It understands: * Your **project configuration** * Existing **components and steps** in the agent * Which **step types** Agent Studio currently supports * How to map your requirements to **best-practice patterns** in Airia This makes it a strong assistant for: * Choosing the right step types for your use case * Suggesting routing and branching patterns * Proposing guardrails, validation, and error-handling flows Example prompts: * “Walk me through how this agent currently routes conversations.” * “Explain what each step does and where data is stored.” * “Recommend a better structure for this workflow given we want stricter approval.” *** ### Troubleshooting & Debugging The Airia Agent is particularly good at **troubleshooting failing agent executions**: * Analyze failed runs, logs, or error messages * Suggest **concrete changes** to steps or configuration * Propose additional logging, retries, or fallback paths Example prompts: * “This execution failed when calling the Jira tool. What should I change?” * “Help me debug why this branch is never triggered.” * “Suggest improvements to make this workflow more resilient.” *** ### Python Code Assistance The Airia Agent can also **write and troubleshoot Python code** used in your workflows: * Generate new Python steps based on your description * Propose refactors or performance improvements * Help you debug failed Python executions Example prompts: * “Write a Python step that normalizes user input and scores it from 1–5.” * “My Python step is failing. Here’s the error—what should I fix?” * “Optimize this Python logic for readability and reliability.” *** ## Billing & Model Usage As of today: * All executions in the **Build with AI** chat in Agent Studio are **charged to the tenant**. * The system is optimized for: * **Maximum quality** of the generated workflows and guidance * **Best possible experience** for complex agent design and troubleshooting tasks ### Models Used We currently use a combination of: * **Sonnet 4.5** * **GPT 5.1** As we expand the skills of the Airia Agent, we may: * Route specific tasks to **smaller, specialized models** optimized for those tasks * Continue to tune the underlying models and instructions for **quality and reliability** > 🔐 **Important**\ > Today, users **cannot change or override** the models used by **Build with AI**.\ > The feature is tightly optimized and tested with a specific setup to ensure consistent behavior across tenants. *** ## Known Behaviors & Limitations * **Canvas locking during AI edits** * While the **Build with AI** chat is actively building or modifying your workflow, the canvas is **temporarily blocked** from manual edits to prevent conflicts. * **Draft-only operations** * You can only use Build with AI on **draft** versions. To edit a published agent, open its draft version first. * **Model configuration** * Model selection for Build with AI is **managed by Airia** and cannot be customized per tenant at this time. *** ## Best Practices * Start with **one clear objective** per agent (e.g., “qualify leads,” “triage tickets”). * Let the AI propose a **first version of the workflow**, then iterate instead of over-specifying upfront. * Use the AI not just to **build**, but also to **explain** and **review** your workflows regularly. * When something fails, **send the error/context** to the Airia Agent and let it suggest the next steps. If you’re unsure what to ask, a good starting point is: > “Explain how this agent works and suggest improvements to make it more reliable for production use.”