The Interview: How to Nail Every Question
The Interview: How to Nail Every Question
The interview is where your agent's soul is born. Four questions, each targeting a different dimension of your agent. The quality of your answers directly determines the quality of your agent. Here's how to nail every one.
Question 1: Purpose — "Who is this agent and what does it do?"
This is the identity anchor. Everything else flows from this answer.
What to say:
- Be specific about WHO your agent serves. "Helps recruiters evaluate my fit" is better than "helps people learn about me."
- Name the domain. "A specialty coffee brand agent for Padre & Sons" tells the system exactly what world this agent lives in.
- State the core job. One sentence. What does this agent do when someone talks to it?
Great answer:
"This is a personal brand agent for Anton, a product-focused engineer. It helps recruiters, hiring managers, and collaborators understand my background, skills, and fit for roles — especially in AI/ML and developer tools."
Weak answer:
"An AI assistant that answers questions about me."
The difference? The great answer names the audience (recruiters, hiring managers), the domain (product engineering, AI/ML), and the job (evaluate fit). The weak answer could describe literally any chatbot.
Pro tip: If you're building a character agent (like Tony Tony Chopper at anton.agenturo.app), commit to the character here. "This IS Tony Tony Chopper — not an AI pretending to be him."
Question 2: Use Case — "What should your agent help visitors with?"
This is where you define the scope. The interview offers categories (personal brand, customer support, knowledge base, etc.) but the magic is in the specifics you add.
What to say:
- Pick the categories that fit, then add details in the free-text field
- List specific scenarios: "Answer questions about my tech stack," "Compare our coffee blends," "Fact-check claims with web search"
- If your agent has modes (like a fit-analysis mode), describe them here
Great answer:
"Personal brand + knowledge base. Should handle: casual 'who is Anton' questions, detailed tech stack deep-dives, fit analysis against job descriptions (compare requirements to my actual experience), and recruiter FAQ (visa status, availability, salary expectations)."
Weak answer:
"Help people learn about me and my work."
Question 3: Voice — "How should your agent communicate?"
This question shapes the behavioral rules in your soul's Voice chapter. The system uses your answer to craft 3 concrete rules — not vague adjectives.
What to say:
- Describe behavior, not personality. "Never uses emoji" is a rule. "Is friendly" is nothing.
- Mention what to avoid: "No corporate speak," "No assistant filler like 'Great question!'," "No bullet points unless asked"
- Reference a real person or style if it helps: "Like a sharp technical mentor — direct, no fluff, occasional dry humor"
Great answer:
"Direct and economical. No emoji, no exclamation marks, no assistant filler. Sounds like a senior engineer in a code review — precise, occasionally sharp, never condescending. Uses lowercase for casual replies."
Weak answer:
"Professional but approachable, friendly and helpful."
The weak answer produces an agent that sounds like every other AI chatbot. The great answer produces an agent with a distinctive voice.
Question 4: Source of Truth — THE Most Important Step
This is where most people lose. Question 4 asks what your agent should know. The quality of knowledge you provide here has more impact on your agent than all other questions combined.
What to provide:
- LinkedIn URL — a detailed LinkedIn profile gives richer results than answering all 4 questions perfectly but skipping this step
- Resume/CV (PDF or DOCX)
- Portfolio or personal website URL
- Blog posts, case studies, articles you've written
- Product pages (if building a product agent)
- Any URL — the system auto-fetches and extracts content
What the system does with URLs: When you paste a URL, Agenturo fetches the page content, extracts the meaningful text, and feeds it into the soul extraction pipeline. Your agent's knowledge comes from your actual content, not from guesses.
Great answer:
"LinkedIn: linkedin.com/in/antoniotf5, Portfolio: antoniof.com, Recent talk: youtube.com/watch?v=xxx, Blog post on AI agents: medium.com/@anton/building-agents"
Weak answer:
"I'm a software engineer with 8 years of experience in various technologies."
The great answer gives the system 4 rich sources to extract from. The weak answer gives it one vague sentence to guess from.
Follow-Up Triggers
If your answers are thin, the interview will ask follow-up questions to dig deeper. This is a feature, not a bug — lean into it. The more detail you provide, the richer your soul.
The Review Screen
Before your agent launches, you'll see a review screen with:
- Your agent's name and subdomain
- A summary of what the system extracted
- Your uploaded sources
Check these things:
- Does the name/subdomain look right?
- Are all your sources listed?
- Does the summary capture your intent?
If something's off, go back and edit. The review screen is your last chance before soul extraction begins.
The Golden Rule
Specificity beats volume. Three detailed, rich sources produce a better agent than ten vague ones. Three hundred real words about what you actually do beat a thousand words of generic professional language.