Start with the End in Mind — Define Your Output

Why This Approach Matters

Every successful Go Agent starts with one question: “What is the final output, and who needs it?”

The answer shapes the structure of your agent — from how documents are ingested, to what data is extracted, to where it goes after. Backwards planning ensures that your workflows are useful, scalable, and production-ready.

Common Output Types

Below are the three most common output patterns for Go Agents. Each has distinct downstream needs, which affect how you build your prompts, properties, and tool logic.

1. System Exports

Objective: Send clean, structured data to an external platform (CRM, CMS, internal DB, etc.)

Example:

Extract income data from an Income Statement
Push fields like “Reporting Dates,” “Operating Income,” and “Revenue Diversification” into your Investment Management Systems.

Agent Considerations:

Use text, number, or select properties (structured fields only)
Prioritize AI citations for auditability
Apply formatting logic (e.g. clean date strings, remove currency markers)

2. Report Generation

Objective: Transform unstructured documents into structured summaries or abstracts

Example:

Pull key provisions from a construction inspection report
Auto-generate a Word or PDF summary report

Agent Considerations:

Use collections for tables (e.g. repayment schedule, fee breakdown)
Add JSON properties for grouped or related outputs
Embed manual review stages if the summary is client-facing

3. Augment Human Work

Objective: Provide just-in-time insight for internal users to make faster decisions

Example:

Legal reviews a 25-page NDA to compare against our guidelines.
Agent pulls all key considerations and generates redlines, while routing to human for review or accepting those that meet all criteria.

Agent Considerations:

Focus on summary-style prompts
Use Claude or Gemini for more fluent reasoning or writing
AI citations optional — depends on internal QA requirements

From Output to Inputs

Once your output is defined, work backwards to identify:

What properties are required to get there?
What document types need to be parsed?
What tools (AI Model, Python, etc.) should be used at each step?

Key Takeaways

Always define your success criteria before you start building. Ask:

What’s the final action, output, or decision this agent is supporting?
Will there be updates to my inputs or files over time?
How will the output be reviewed or audited (if at all)?
Who uses the output? What is the most optimal output format?
Are there other files or processes that might impact/be impacted by this workflow?

With a clearly defined destination, every downstream decision — prompt writing, property types, tool choice — becomes faster and easier.