We Gave the AI Supervisor Structured Tools So It Couldn’t Hallucinate. It Still Made the Wrong Call.
Last Updated on May 27, 2026 by Editorial Team
Author(s): Tarun Agarwal
Originally published on Towards AI.
Where prompts stop working and code has to take over.
The routing problem felt solved.
The author explains that while a tool-based “AI supervisor” with structured routing fixed hallucinated agent names, unvalidated logic, and poor traceability, it still failed because the model’s reasoning about when and why to route wasn’t reliable. They walk through five production failure modes: an infinite retry loop when an agent fails (solved with hard circuit breakers), ignoring stored state and re-routing to already-successful agents (solved with code-level routing gates), a loop created by those gates (solved with gate escalation and forced terminal routing), prematurely calling completion when downstream work never actually ran (solved with runtime generation of strict per-intent enforcement rules from the capability manifest), and silently dropping part of a multi-intent query (solved by supporting compound workflow IDs and executing unioned requirements). The throughline is that LLMs can handle fuzzy decisions, but invariants must be enforced deterministically in code; structured tools clarified what the supervisor decided, and subsequent code guardrails ensured contracts couldn’t bend. The post closes by saying the next iteration will be plan-based—having the model commit to an execution plan before firing agents—and points readers to a GitHub repo for the series.
Read the full blog for free on Medium.
Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.
Published via Towards AI
Towards AI Academy
We Build Enterprise-Grade AI. We'll Teach You to Master It Too.
15 engineers. 100,000+ students. Towards AI Academy teaches what actually survives production.
Start free — no commitment:
→ 6-Day Agentic AI Engineering Email Guide — one practical lesson per day
→ Agents Architecture Cheatsheet — 3 years of architecture decisions in 6 pages
Our courses:
→ AI Engineering Certification — 90+ lessons from project selection to deployed product. The most comprehensive practical LLM course out there.
→ Agent Engineering Course — Hands on with production agent architectures, memory, routing, and eval frameworks — built from real enterprise engagements.
→ AI for Work — Understand, evaluate, and apply AI for complex work tasks.
Note: Article content contains the views of the contributing authors and not Towards AI.
Related posts
Recent Posts
