# How do you handle routing with little context? — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/25a0c805d6d2af0c

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-09-30 15:13 ingested 2026-10-01 02:37

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1wu7km2/how_do_you_handle_routing_with_little_context/)

## retained text



```
Hey all,

I'm working on an agent built with graph that has a few subagents under it. One of them is a triage agent whose only job is to send each request to the right place. If someone asks about metrics, KPIs or dashboards, it goes to the analytics agent. If it's about raw events data, it goes to the raw data agent.

That works fine when the user is explicit. The problem is that a lot of real prompts are vague.

I've thought about two approaches, and I'm not happy with either:

Writing a markdown doc for each context and giving them all to the triage agent. I'm worried this gets bloated fast, burns tokens, and makes routing slower and noisier as I add more contexts.

Just asking the user which context they mean. It works, but it kills automation. I'd like people to be able to run tasks without having to babysit them.

So I'm curious how others handle this.

Would really appreciate hearing what's actually worked for you in production, not just in theory. 

Thanks!
```



document fetched 2026-10-01 00:20 · canonical https://reddit.com/r/AI_Agents/comments/1wu7km2/how_do_you_handle_routing_with_little_context
