---
title: "Comment by Hayrullah Kar on \"Why 95% of AI Agent Prototypes Never Reach Production — and How to Build One That Does\" | Hashnode"
description: "The deterministic backbone with one model call for scoring is a pattern I can point people to. A small mismatch in the policy JSON: allowed_tools uses names lik"
url: "https://hashnode.com/posts/why-95-of-ai-agent-prototypes-never-reach-production-and-how-to-build-one-that-does/6ae9a58925e12c86443b59ce/comment/6aeb19cdeba87a600a642a24"
---

[Discussion](https://hashnode.com/posts/why-95-of-ai-agent-prototypes-never-reach-production-and-how-to-build-one-that-does/6ae9a58925e12c86443b59ce)

Comment [Why 95% of AI Agent Prototypes Never Reach Production — and How to Build One That Does](https://hashnode.com/posts/why-95-of-ai-agent-prototypes-never-reach-production-and-how-to-build-one-that-does/6ae9a58925e12c86443b59ce)

[Hayrullah Kar](https://hashnode.com/@magesheet)

[5d ago](https://hashnode.com/posts/why-95-of-ai-agent-prototypes-never-reach-production-and-how-to-build-one-that-does/6ae9a58925e12c86443b59ce/comment/6aeb19cdeba87a600a642a24)

The deterministic backbone with one model call for scoring is a pattern I can point people to. A small mismatch in the policy JSON: allowed\_tools uses names like crm.create\_deal, but requires\_human\_approval lists create\_deal. If the runtime matches tool names literally, the approval rule never fires for the one tool it was written for. Using the same identifier in both lists, plus a test that fails when an approval-listed action matches no tool, would close that gap. Also, the MIT figure in the body is about pilots with no measurable P&L impact, which is a different claim from the title's "never reach production".