Which steps need an agent and which should stay fixed? A booking breakdown
AlarmingPepper9193 · reddit · 2026-11-15
Using a fictional sign-installation booking ("move the shop sign before Friday, same site, but the entrance changed"), the author argues that calling the whole process "an agent" hides distinct decisions that should be separated before choosing architecture:
- Interpret the message — model proposes a structured summary, keeping original wording beside it; fluent text doesn't make a valid booking.
- Apply established rules — field checks, authorized-record lookups and price-list math belong in code; a model output shouldn't decide permissions.
- Investigate variable questions — where an agent may earn its place: choosing which permitted records to inspect next, with bounded records, tools and stopping points. If the sequence is always fixed, a workflow with a model call suffices.
- Make the business commitment — a human coordinator resolves ambiguity; generating a plausible date isn't authority to promise it.
Useful per-step note: input → allowed decision → evidence needed → next owner when unresolved. The author stresses this is a hypothesis to validate by comparing fixed workflow vs agent on identical enquiries. Affiliated with Codoki Studio.
More from coding & agent
- The T.R.U.S.T. check: deciding which jobs you can safely hand to your AI bots — bfrench · 2026-11-15
- Read-only MCP server lets Claude Code write regression tests for agents — tom_mathews · 2026-11-15
- Bot audit experiment: triage plus deep-read cuts input 64% while staying stable — bfrench · 2026-11-15
- How do you track all the AI agents running inside your company? — CupRare814 · 2026-11-15
- Developer builds VR island adventure game with Claude Code writing the C# — Illustrious-Lynx3673 · 2026-11-15
- Attest: record agent runs once, replay in CI with zero model calls — tom_mathews · 2026-11-15