# At what point did we decide that adding a fifth supervisor agent was better than writing three deterministic if statements? — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/58b37a8a622ccba4

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-09-23 11:42 ingested 2026-09-24 02:36

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1wo3n2r/at_what_point_did_we_decide_that_adding_a_fifth/)

## retained text



```
Spent the last three weeks refactoring a multi agent customer support and triage system that a client paid forty grand to an agency to build. 

The original architecture had a master router agent, a sentiment classifier agent, a context extractor agent, and a tool selector agent, all orchestrated through a massive graph. Every single inbound user query was burning twelve thousand tokens and taking nine seconds of latency before anyone even touched a database query. 

Worse, the router would randomly hallucinate edge cases twice a day, routing simple billing refund tickets to technical tier 3 engineering because a customer used the word "broken." 

We deleted three of the middle agents, replaced the routing logic with a fast regex pass, an embedding cosine similarity check, and forty lines of plain Python conditional logic. Latency dropped from nine seconds to 800 milliseconds, token cost dropped by 75%, and routing accuracy jumped to near 99%. 

It feels like the entire agent space has caught a disease where people try to solve deterministic software engineering problems by throwing nondeterministic LLM loops at them. If a router can be written as a switch statement or a pydantic validator, turning it into an agent is just burning compute to look sophisticated. 

Where are people drawing the line between actual autonomous agentic behavior and basic procedural code that never needed an LLM in the first place?
```



document fetched 2026-09-24 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1wo3n2r/at_what_point_did_we_decide_that_adding_a_fifth
