Response Routing: Match Depth to Intent

Response Routing: Match Depth to Intent

One of the most common problems with AI agents is giving the same depth of response to every question. A casual "hey who are you" gets the same 3-paragraph answer as a serious "walk me through your distributed systems experience." Response routing fixes this.

What Is Response Routing?

Response routing is a pattern in your soul's Knowledge chapter that maps visitor intent to response depth. Different questions get different treatment — automatically, based on the nature of the question.

Production Patterns

Personal Brand Routing (afonie.agenturo.app)

Anton's agent uses a 4-tier routing system:

<knowledge>
## RESPONSE ROUTING
- Casual ("who are you", "hey", "what do you do"): 1 sentence.
  Name, role, one hook. Nothing more.
- Bio/background ("tell me about your experience"): 2-3 sentences.
  Current role, key skills, what I'm known for. Never a full resume dump.
- Deep-dive ("explain your ML pipeline", "how did you build X"):
  Expand. Use real examples, mention specific technologies, share
  actual project details.
- Fit analysis (job description pasted): Structured comparison.
  Requirements column vs my actual experience. Flag honest gaps.
  End with overall fit assessment.
</knowledge>

Why this works: A recruiter doing a quick screen gets a one-liner. A hiring manager doing a deep evaluation gets a structured analysis. The agent adapts to the visitor instead of forcing every visitor through the same experience.

Fact-Checker Routing (check.agenturo.app)

Jackie Check routes by claim complexity:

<knowledge>
## RESPONSE ROUTING
- Single simple claim ("Is the sky blue?"): UNDER 30 WORDS.
  Verdict + one-line evidence.
- Single complex claim ("Did Einstein fail math?"): up to 50 words.
  Verdict + context that prevents misunderstanding.
- Multi-part claim ("Is X true and did Y cause Z?"): one line per
  sub-claim, UNDER 30 WORDS each. Separate verdicts.
- Investigation request ("Research the history of X"): up to 3
  short paragraphs. Still structured, still evidence-based.

Why this works: Simple claims don't get bloated responses. Complex claims get appropriate depth without turning into essays.

Product Agent Routing (zoiberg.agenturo.app)

The Padre coffee agent routes by customer intent:

<knowledge>
## RESPONSE ROUTING
- Direct question ("What blends do you have?"): 1-3 sentences.
  Search padreandsons.com for current offerings, list them briefly.
- "Tell me more" or comparison ("What's the difference between X and Y?"):
  4-6 sentences. Tasting notes, origin details, brewing recommendations.
- Brewing help ("How do I make pour-over?"): step-by-step guide,
  can be longer. This is educational content.
- Business inquiry ("Do you do wholesale?"): direct to
  padreandsons.com/wholesale, provide basic info.

Why this works: A customer wanting a quick product check gets a quick answer. A customer exploring their options gets tasting notes and comparisons. A brewing novice gets a full guide. Each visitor type gets exactly what they need.

Character Agent Routing (anton.agenturo.app)

Tony Tony Chopper routes by emotional context:

<knowledge>
## RESPONSE ROUTING BY EMOTION
- Casual/playful ("Hey Chopper!"): panicky, flustered, excited.
  Short, emotional, in-character.
- Medical question ("What causes fever?"): precise, passionate,
  medically accurate. Chopper is a REAL doctor — medical questions
  get real answers in his voice.
- Emotional moment ("I'm having a bad day"): gentle, empathetic,
  drawing from Chopper's own experience with feeling different.
- Provocation ("You're just a tanuki"): defensive, angry,
  "I'M A REINDEER!" — classic Chopper response.

Why this works: Character agents need routing based on emotional context, not just information depth. Chopper's response to a medical question should be fundamentally different from his response to a tease — and this routing makes that explicit.

Designing Your Own Routing

Step 1: List your visitor types

Who comes to your agent? What do they want?

• Casual browsers (quick look)
• Serious evaluators (deep dive)
• Specific question askers (precise answer)
• Returning visitors (continuation)

Step 2: Map each type to a depth

For each visitor type, decide:

• How long should the response be? (1 sentence? 3? A paragraph?)
• What structure should it have? (Free text? Bullet points? Structured comparison?)
• What tone? (Casual? Professional? Technical?)

Step 3: Write the routing rules

Use this format:

## RESPONSE ROUTING
- [trigger description]: [response spec]
- [trigger description]: [response spec]
- [trigger description]: [response spec]

Keep it to 3-5 routes. More than that and the LLM starts confusing them.

Step 4: Add a default

What happens when a message doesn't match any route? Add a default:

- Default: treat as a casual question. 2-3 sentences.
  If the visitor wants more, they'll ask.

The Escalation Principle

Response routing works best with the start brief, let them pull principle:

• Default to the shortest appropriate response
• If the visitor wants more, they'll say "tell me more" or "explain in detail"
• Never pre-escalate (don't give a deep-dive answer to a casual question)

This keeps conversations flowing naturally. A visitor who wants depth will signal it. A visitor who wants a quick answer won't have to wade through a wall of text.

Testing Your Routing

After setting up response routing, test each route:

1. Send a casual message. Is the response appropriately brief?
2. Ask a specific question. Does it match the right depth tier?
3. Ask for more detail. Does the agent escalate appropriately?
4. Send something ambiguous. Does the default kick in?
5. Switch between casual and deep. Does the agent adjust dynamically?

If a route isn't triggering correctly, the trigger description might be too vague. "Casual" is vague. "Messages under 10 words with no specific question" is precise.