AI Playground guide

Build useful AI, one clear step at a time

Start with Quick Build, improve your prompt in the unified Studio, then explore workflows and agents when you are ready. You do not need to configure everything at once.

Open AI Playground

Your easiest path: Quick Build → Studio

You only need two main screens to get started. Start with a working result, then improve the instructions in one workspace.

STEP 01

Choose & run

Choose Customer Feedback Analyzer or another template. Use sample input, then run.

STEP 02

Make it yours

Replace the sample with your own request and compare the result.

STEP 03

Customize & test

Open Customize in Studio. Edit prompts, variables and run settings, then test in the same workspace.

Try Quick Build

Understand the unified Prompt Studio

Edit instructions and test your prompt side by side, without navigating between multiple editor pages.

Prompt name & description

Give your prompt a clear purpose so you can find and reuse it.

User prompt

Describe the task, required context and desired output format.

System prompt

Set the role, rules, boundaries and tone.

Variables

Define reusable placeholders, descriptions, defaults and required inputs.

Provider & model

Choose the supported AI model and access method for the run.

Temperature

Use lower values for consistency and higher values for variation, depending on model support.

Maximum output tokens

Control the upper limit of response length.

Thinking mode

Where supported, enable additional reasoning for complex tasks; it can affect latency and usage.

Version history

Save improvements as versions and compare iterations.

Test & response

Enter sample values, inspect the rendered prompt, run and review the result.

Prompt improvement in three passes

V1 · Simple

Analyze this customer feedback.

V2 · Structured

Identify sentiment, themes, pain points, urgency and recommended actions. Return clearly labeled sections.

V3 · Reliable

Analyze only the supplied feedback. Do not invent details. Explain urgency using evidence; distinguish facts from assumptions and return prioritized, actionable recommendations.

Common Build questions

The recommended journey

Step 1

Create a project

A project keeps related prompts, workflows, agents and execution history together.

Step 2

Build prompts

Write and test reusable AI instructions. Variables make each prompt reusable.

Step 3

Connect a workflow

Arrange tested prompts into ordered steps and pass one step’s output to the next.

Step 4

Create an agent

Give a tested workflow a clear role, goal, instructions and reusable identity.

Practical examples · Knowledge & RAG

Knowledge Bases & RAG: Three Practical Examples

Try a simple FAQ assistant, a policy evidence assistant, and an advanced support knowledge assistant. Includes copy-ready prompts, sample knowledge, test questions, and evaluation checks.

Explore Knowledge & RAG Examples

End-to-end examples

Build a complete AI solution

Follow a real use case from Prompt to Knowledge, Workflow, Agent and MCP. Each guide includes copy-ready prompts, mappings, test data and success checks.

Career use case

CV & Job Match Career Coach

Compare a candidate profile with a real job, retrieve evidence from the CV, create an improvement plan and optionally enrich the analysis with live company research.

Open case study

Product operations use case

Customer Feedback Action Planner

Turn customer feedback into grounded themes, priority decisions, owners and a safe response while checking related live support signals.

Open case study

Use sample input whenever you are unsure

Sample input is an example, not fixed production data. Replace it with your own input before a real run.

Prompt

Use Quick Build sample input first; in Studio, variables and default values make your prompt reusable.

Workflow

Define a sample value for each starting input, then use “Use sample workflow input”.

Agent

The agent starts with workflow examples, which you can customize before saving the agent.

How to build and run

Create a prompt

Start in Quick Build and select Customize in Studio, or open an existing project and select New Prompt.
Give the prompt a clear name and purpose.
Write the system and user instructions.
Add every variable referenced as {{variable_name}}.
Add a helpful default/example value and test the prompt.

Create a workflow

Create the workflow starting input.
Add a realistic sample value.
Select the prompt for step one.
Add later prompts and pass earlier output forward.
Create and test the workflow before creating an agent. Current V1 workflows execute steps in order.

Create an agent

Give the agent a name, role and goal.
Add concise operating instructions.
Select a tested workflow and version.
Review or customize its sample input.
Create the agent and run it from Agent Runner.

Review a run

Read the final response.
Inspect every step in the execution trace.
Check the model and whether the result was cached or freshly executed.
Select an older run to restore its inputs, output and trace.
Use failures to correct missing inputs or quota issues.

Five complete examples

Example 1

Customer Feedback Action Plan

Project: Customer Experience
Prompts: Feedback Analyzer + Action Recommender
Workflow: Feedback Analyzer → Action Recommender
Agent: Customer Feedback Agent

Sample input · customer_feedback

The product is useful, but onboarding was confusing and I could not find the export option.

Expected result: Sentiment, friction points and a prioritised action plan with owners and expected outcomes.

Example 2

CV and Job Match Assistant

Project: Career Toolkit
Prompts: CV Skill Extractor + Job Match Reviewer
Workflow: CV Skill Extractor → Job Match Reviewer
Agent: Career Match Agent

Sample input · candidate_and_job

Candidate: Senior Python engineer with FastAPI, AWS and ML experience. Role: GenAI engineer requiring Python, Azure OpenAI, RAG and stakeholder communication.

Expected result: Match score, strengths, missing skills and concrete CV/interview improvements.

Example 3

Research to LinkedIn Post

Project: Content Studio
Prompts: Topic Researcher + LinkedIn Writer
Workflow: Topic Researcher → LinkedIn Writer
Agent: Thought Leadership Agent

Sample input · topic

How small businesses can use AI agents safely without losing human oversight.

Expected result: A structured research summary followed by a concise, audience-friendly LinkedIn draft.

Example 4

Support Ticket Triage

Project: Support Operations
Prompts: Ticket Classifier + Response Planner
Workflow: Ticket Classifier → Response Planner
Agent: Support Triage Agent

Sample input · support_ticket

Customers cannot export invoices after today’s release. Three enterprise accounts are affected and receive a 500 error.

Expected result: Category, urgency, impact, recommended owner and a safe first-response plan.

Example 5

Meeting Notes to Actions

Project: Team Productivity
Prompts: Meeting Summarizer + Action Item Extractor
Workflow: Meeting Summarizer → Action Item Extractor
Agent: Meeting Follow-up Agent

Sample input · meeting_notes

The team agreed to complete API testing by Friday. Maya will prepare test data. Jonas will review security findings. The release decision is Monday.

Expected result: A concise summary plus action items, owners, deadlines and unresolved decisions.

Current V1 boundaries

A workflow runs steps in order; it does not branch or loop yet.
An agent runs its pinned workflow version. It does not independently use external tools yet.
Changing prompt content creates a new prompt version; existing workflows continue using their pinned version.
AI output may be incorrect. Do not enter passwords, secrets or highly sensitive personal data, and review important results before acting.