AI Workflow Designer

Overview

The AI Workflow Designer is a conversational interface that creates complete, production-ready workflows from a natural language description. Instead of manually configuring steps, variables, and loops, you describe what you want the workflow to do and the designer builds it for you.

The designer understands workflow architecture — step isolation, variable wiring, JSON validation patterns, loops, tool usage, and benchmarking — and applies best practices automatically. The result is a fully configured workflow ready to import and run.

How It Works

The designer operates in a two-phase conversational flow:

Phase 1: Clarifying Questions

After you describe your workflow, the designer may ask clarifying questions to ensure the best result. Questions can be:

• Multiple choice — Select from provided options (e.g., output format, number of items to process)
• Open-ended — Provide free-form answers (e.g., specific data points to extract)

Answer all questions and click Send Answers to proceed. If your initial description is detailed enough, the designer may skip directly to Phase 2.

Phase 2: Workflow Generation

The designer generates the complete workflow, including:

• All steps with fully written prompts
• Variable definitions with descriptions and default values
• Loop configurations for iterating over collections
• JSON validation steps for reliable structured output
• AI model assignments where specific models are needed
• Benchmark configuration for quality evaluation
• Final output step selection

During generation, you will see progress status messages:

• Analyzing workflow... — The designer has finished generating and is reviewing the result
• Validating variables... — Variable issues were detected (unused or undefined variables)
• Refining workflow... — The designer is correcting variable issues automatically

Once complete, click View Workflow to import it directly into your workflow editor.

Writing Effective Descriptions

The quality of the generated workflow depends heavily on how you describe it. The designer works best when you are specific about three things: inputs, outputs, and processing.

Be Specific About Inputs

Tell the designer exactly what data the workflow will receive:

• File inputs — "Input: an uploaded contract PDF" or "Input: meeting transcript file"
• Text inputs — "Inputs: company name, industry, and job title"
• Lists — "Input: a JSON array of competitor names" or "a list of employee names"
• Site data — "Read employee folders from a receipts site"

Be Specific About Outputs

Describe what the workflow should produce:

• Structured data — "Extract key terms as JSON with effectiveDate, termMonths, and liabilityCaps"
• Reports — "Output a comprehensive markdown report with executive summary and recommendations"
• Actionable content — "Draft follow-up emails for each action item" or "Draft redline language for each risk"

Be Specific About Processing

Explain how items should be handled:

• Iteration — "For each competitor, use web search to research their products and pricing"
• Tool usage — "Use list_content to discover employee folders, then retrieve each receipt file"
• Web research — "Use web search to find current salary benchmarks"
• Multi-step processing — "Extract data as JSON, then analyze trends, then generate recommendations"

Example Descriptions

Good — Specific inputs, outputs, and processing:

Build a competitor intelligence workflow. Inputs: company name, industry, and a JSON array of competitor names. For each competitor, use web search to research their products, pricing, and market position. Then compare all competitors against our company and output strategic recommendations in markdown.

Good — File-based with actionable output:

Create a contract risk review workflow. Input: an uploaded contract PDF. Extract key terms (effective date, term length, liability caps, termination clauses) as JSON. Then identify risk areas and draft specific redline language with negotiation rationale for each flagged clause.

Good — Tool-driven with hierarchical data:

Design an expense report workflow that reads from a receipts site. List employee folders, then for each employee find the matching month/year subfolder, list receipt files, and extract vendor, amount, date, and category from each receipt. Aggregate into a company-wide summary with per-employee breakdowns.

Too vague — Missing inputs, outputs, and processing details:

Make a workflow that analyzes documents.

What the Designer Builds

Context Preambles

Every step includes a three-part context preamble because workflow steps execute in complete isolation with no shared memory. The preamble ensures the AI model understands:

1. Workflow context — What overall task is being accomplished
2. Step purpose — What this specific step contributes
3. Data context — What the incoming variables represent

JSON Validation Steps

Whenever a step produces JSON output, the designer automatically adds a validation step immediately after it. This two-step pattern ensures reliable structured data:

1. Generation step — Produces JSON, outputs to a variable with a Raw suffix (e.g., termsRaw)
2. Validation step — Cleans and validates the JSON, outputs to the final variable name (e.g., terms)

Subsequent steps reference the validated variable, not the raw output.

Loop Steps for Collections

When your workflow needs to process multiple items (competitors, receipts, action items), the designer creates loop steps rather than asking the AI to process everything in a single prompt. This ensures:

• Each item gets the AI's full attention
• Output structure is consistent across all items
• One failed item doesn't affect others
• Results are automatically aggregated as a JSON array

Variable Wiring

The designer ensures every variable reference is properly connected:

• Every workflow input variable appears as {{variableName}} in the appropriate step content
• Every step that uses data from a previous step references that step's outputVariable
• Every validation step references the corresponding Raw variable
• Every loop source references the correct array variable

Server-Side Validation

After generating the workflow, the designer automatically validates variable wiring on the server before delivering the result. If it detects unused variables (defined but never referenced) or undefined variables (referenced but never defined), it sends the workflow back to the AI for correction. This happens transparently — you see status messages during the process — and ensures the delivered workflow has clean variable connections.

AI Model Selection

The designer assigns AI models strategically:

• Intelligent (auto-selection) as the workflow default for most steps
• Pro-tier models (e.g., GPT-5.2 Pro) for complex reasoning and synthesis steps
• Image generation models when visual output is needed

Benchmarking

By default, the designer enables the default benchmark with runOnCompletion: true. This provides interactive quality visualizations evaluating readability, accuracy, clarity, tone, engagement, and vocabulary after each workflow run.

After Generation

Once the workflow is generated:

1. Click View Workflow to import it into your workflow editor
2. The workflow is created in your account and opens in the editor
3. Review the steps, variables, and configuration
4. Adjust prompts, add or remove steps, and modify variables as needed
5. Create a job to test the workflow with real data

Tips for Best Results

Start with the Outcome

Think about what the final output should look like and work backward. If you want a competitive intelligence report, describe the report sections you need (executive summary, competitor comparison matrix, strategic recommendations) and the designer will decompose that into the right steps.

Mention Tools When Relevant

If your workflow should read data from a site rather than receiving it as input, say so explicitly:

• "Use list_content to discover folders in the site"
• "Use retrieve_file_content to extract data from each receipt"
• "Use web_search to research current market data"

The designer will include the correct tool references in the prompts and set allowWebAccess when needed.

Request Value-Adding Steps

Don't stop at extraction and analysis. Ask the designer to include steps that produce actionable content:

• "Draft follow-up emails for each action item"
• "Draft redline contract language for each risk"
• "Draft personalized outreach messages for each prospect"
• "Generate strategic recommendations based on the analysis"

These steps transform passive reports into action-ready tools.

Specify JSON Structure When It Matters

If specific field names matter for your downstream use, mention them:

Extract receipt data as JSON with vendor, amount, date, category, and paymentMethod fields.

The designer will use those exact camelCase property names in the JSON schema, ensuring consistent dot-notation access in subsequent steps.

Use JSON Arrays for Lists

When providing a list of items to process, describe them as a JSON array:

Input: a JSON array of competitor names like "Slack", "Microsoft Teams", "Discord"

This is more reliable than comma-separated strings and handles items containing commas or special characters.

Iterate on the Result

The generated workflow is a strong starting point. After importing:

• Review the prompts and adjust phrasing for your specific domain
• Add or remove steps as needed
• Adjust variable default values
• Test with a manual job and refine based on output quality
• Enable benchmarks to track quality over time

Comparison: Designer vs. Builder

Clear Ideas offers two ways to create workflows automatically:

Use the Designer when you know the process you want to automate and can describe the inputs, steps, and outputs.

Use the Builder when you have a reference document and want to create a workflow that generates similar documents from your organization's data.

Common Use Cases

Document Analysis and Extraction

• Contract review — Extract terms, identify risks, draft redline language
• Financial report analysis — Extract metrics, calculate trends, generate executive summaries
• Policy compliance review — Extract criteria, evaluate documents against policies

Research and Intelligence

• Competitor intelligence — Research competitors via web search, compare positioning, recommend strategy
• Market research — Gather data on market size, trends, and competitive landscape
• Prospect research — Build prospect profiles and draft personalized outreach

Data Processing from Sites

• Expense report processing — Discover employee folders, extract receipt data, aggregate summaries
• Invoice processing — List vendor folders, extract line items, flag discrepancies
• Content audit — Discover documents in a site, evaluate against guidelines, produce remediation reports

Content Creation

• Social media calendars — Generate platform-specific posts across themes and weeks
• Meeting summaries — Extract decisions, action items, and draft follow-up communications
• Proposal generation — Extract RFP requirements, draft tailored responses, compile compliance matrix

Reporting and Dashboards

• KPI reporting — Aggregate metrics, calculate variance, generate leadership summaries
• Project status — Collect workstream data, synthesize into executive dashboard
• Talent review — Extract performance data, generate calibrated team summaries