---
type: Note
id: c6a61c6e24d4
title: Agent design is still hard
description: High-level SDKs break under real tool use. Caching, reinforcement, failure
  isolation and shared state decide whether an agent is reliable.
tags:
- ai-native-sdlc
generated:
  by: seed-import/1.0
  at: '2026-10-17T19:25:45+00:00'
  body_hash: sha256:58abb5f7203397c70a14e046dc5780bc09d09d771c8f09bd4de298ca7d5a8d0b
sources:
- id: s1
  resource: /raw/20261017-192545-agent-design-is-still-hard.md
  title: Agent design is still hard
kind: Essay
status: stable
stale_after: '2027-10-17T00:00:00Z'
classification: public
owner: human:muthu
canonical_essay: https://muthub.org/writing/agent-design-is-still-hard
verified:
  by: human:muthu
  at: '2026-10-18T02:01:22Z'
---

# Overview

High-level SDKs break under real tool use. Caching, reinforcement, failure isolation and shared state decide whether an agent is reliable.[^s1]

# Read the original

The full note is on muthub.org: [Agent design is still hard](https://muthub.org/writing/agent-design-is-still-hard).

[^s1]: Agent design is still hard, muthub.org, 2025-11-25.