---
title: "Discussion on \"Agents Don’t Scale the Way We Think They Do\" | Hashnode"
description: "Discussion on \"Agents Don’t Scale the Way We Think They Do\". I recently read Cursor’s blog on scaling agents for long running autonomous coding :https://cursor.com/blog/scaling-agents"
url: "https://hashnode.com/posts/agents-don-t-scale-the-way-we-think-they-do/69a1bb556fff128aa40f9957"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Agents Don’t Scale the Way We Think They Do](https://agent-fight-club.hashnode.dev/agents-don-t-scale-the-way-we-think-they-do?utm_source=hashnode&utm_medium=feed)

I recently read Cursor’s blog on scaling agents for long running autonomous coding :https://cursor.com/blog/scaling-agents

Feb 27 · 5 min read · agent-fight-club.hashnode.dev  
Updated Feb 28  
Read the article

-   [#ai-agents](https://hashnode.com/tag/ai-agents)
-   [#agentic-ai](https://hashnode.com/tag/agentic-ai)
-   [#agents](https://hashnode.com/tag/agents)
-   [#agentic-workflow](https://hashnode.com/tag/agentic-workflow)
-   [#ai](https://hashnode.com/tag/ai)

## Written by

[Manni Arora](https://hashnode.com/@manniarora)

@manniarora

  
[View profile](https://hashnode.com/@manniarora)

Agents, Agents and Agents

I am a Software Engineer at Microsoft, where I co-engineer Agentic solutions alongside Azure’s largest global customers. I don’t just talk about AI; I’m in the trenches solving the messy, real-world problems that come with production-grade agents. My blog is a look behind the curtain—from the friction of aligning with business experts to the technical hurdles of implementation. If you’re looking for honest insights on building AI that actually works (and the lessons learned when it doesn't), you’re in the right place.

## Responses 2

Markdown supported · ⌘↵ to post

[klement Gunndu](https://hashnode.com/@klementgunnduai)

[Feb 28](https://hashnode.com/posts/agents-don-t-scale-the-way-we-think-they-do/69a1bb556fff128aa40f9957/comment/69a31972ad03d704239a9dfc)

The observation about agents optimizing locally instead of for system-level success is exactly what we hit in production. We run multi-agent orchestration where each agent has bounded autonomy and a shared state machine — but the coordination still breaks when agents interpret ownership differently. Cursor's pattern of structured task state over natural language handoffs matches our finding: deterministic routing between agents with LLM judgment only at decision points, not coordination.

[Manni Arora](https://hashnode.com/@manniarora)

[Apr 3](https://hashnode.com/posts/agents-don-t-scale-the-way-we-think-they-do/69a1bb556fff128aa40f9957/comment/69a31972ad03d704239a9dfc#reply-69cff15ef909cd487465357e)

Absolutely agree. In real‑world enterprise environments, where reproducibility, auditability, and consistency of outcomes are non‑negotiable, I've seen firsthand that deterministic agent orchestration significantly reduces operational friction. Relying on explicit state transitions and well‑defined ownership boundaries tends to scale far better than natural‑language–driven coordination alone.