Coding Agents Still a Hassle in Real Projects
Zxen_01 · reddit · 2026-07-12
A lengthy rant about a code model/coding agent: the author used it in a CMS platform project, finding it 'limited but occasionally useful'—quick on simple requests but extremely time and token-consuming for complex changes.
Specific experiences:
- Simple yes/no questions answered within minutes.
- Modifying UI logic for a calendar component took 25 hours and 3 million tokens.
- The author notes it is more 'honest' on some tasks than another tool but still requires heavy human oversight and patching.
- Complaints about frequent popups, poor communication, insufficient support for complex projects, and potential cost escalation from monthly subscription to $25,000 per day.
Overall tone is exaggerated, but the core message: coding agents remain unreliable in large real-world projects, especially for complex UI/state modifications, and human intervention is still essential.
More from coding & agent
- Devin adds e2b sandboxes for remote agent execution — badphilosopher · 2026-07-22
- Hermes Agent Refactoring Proposal: Decoupling via Event Bus and Monorepo Slicing — Promptmethus · 2026-07-22
- ty now reads Pydantic config keywords and field metadata — charliermarsh · 2026-07-22
- Pensar Launches AI Security Agent to Autonomously Discover and Patch 0-Days — andriy_mulyar · 2026-07-22
- ty adds first-class Pydantic support, including strict and lax field handling — charliermarsh · 2026-07-22
- Google launches Gemini 3.5 Flash Cyber for CodeMender, with limited access for governments — GoogleAI · 2026-07-22