Best Practices: AI Agents
Last updated: April 30, 2026
Goal
Designed to help your team build accurate, reliable AI Agents that classify, qualify, and enrich companies at scale without ambiguity or misinterpretation.
Why It Matters For You
AI Agents power qualification, routing, and personalization across your plays.
Clear agent logic reduces false positives and downstream errors in Smart Snippets.
Well-structured agents create consistency across teams and use cases.
Example:
“An agent can only be as good as the question you ask it.”
What People Do Wrong
Combine multiple questions into a single agent.
Use vague or overly verbose guidance that confuses the agent.
Rely on free-text outputs when structured outputs would be more reliable.
Update agent guidance without documenting the change.
Unify’s Best Practice Recommendations
Ask one clear question per agent.
Keep agent questions concise and unambiguous.
Put all necessary context in the guidance, not the question.
Prefer True/False or Multi-Select outputs when the agent will be reused downstream.
How to Apply / Set Up / Build
Step | What to Do | Why It Works |
1 | Define the single decision the agent should make | Prevents conflicting outputs |
2 | Match the agent record type to your play trigger record type (e.g., if trigger is 'person', agent should be 'person') | Prevents agent failures and ensures reliable execution |
3 | Write a short, explicit question | Reduces misinterpretation |
4 | Add clear, structured guidance with criteria | Improves accuracy and consistency |
5 | Choose the simplest possible output type | Makes reuse easier |
6 | Document any guidance updates | Enables auditing and troubleshooting |
Example Questions
“Does this company fit our ICP?”
“Is this company B2B?”
“Does this company use Workday?”
Pro Tips
Re-read guidance from the agent’s perspective before saving.
Avoid extraneous lines that don’t affect the decision.
If the logic feels complex, the agent should probably be split into two.
Always note when guidance changes — Unify does not auto-track this.
Measuring Success / How To Know You’re Doing This Right
Higher accuracy in ICP and enrichment outputs
Fewer Smart Snippet failures tied to agent output
Reduced need for manual overrides or QA
Templates
Template Name | Guidance | Question |
ICP Qualification (True/False) | Your task is to determine whether a company fits our Ideal Customer Profile (ICP). Use reliable information from the company’s website, LinkedIn, or trusted business data sources. Return only True if the company matches the ICP criteria or False if it does not. | Does this company fit our ICP? |
Technographics (Multi-Select) | Your task is to determine whether a company is using [Insert software / tool name]. Use reliable public sources such as the company’s website, job postings, press releases, case studies, or technology lookup databases.Detection Criteria:• Mentions of the tool on the company’s website, job listings, or LinkedIn posts• References to employees working with or implementing the tool (e.g. “Workday Implementation”, “SAP Specialist”)• Evidence that the company’s HR, finance, payroll, or ERP systems are powered by the toolIf clear evidence exists → output the software name.If no evidence is found → output: none. | Which tool does this company use? |
Industry Classification (Multi-Select) | Your task is to classify the company into one of the following industries: | What industry is this company? |