Feedback regarding fable 5 implementation

Subject: Feedback — Fable 5 on a mature production IoT system

I run a commercial IoT platform for horticulture — ESP32 firmware across several device types, a Node/Express backend, React PWA, MQTT, and PostgreSQL — that’s been in production with active customers for about six months. The system was built almost entirely with Claude, from 4.5 through 4.8, and I’ve recently brought Fable 5 into the workflow via Cursor.

The standout has been code review and judgment on a mature codebase. Fable is extremely thoughtful about picking up important details — the kind that matter in a system with deployed hardware and real customers, where a careless change has consequences. It respects existing conventions and operational constraints, and the quality of attention feels like a step up.

Wanted to pass along that this progression — building with successive Claude models and now maintaining with Fable in cursor — has been an excellent experience end to end. Wanted to geive feedback to cursor as obviously they playing a big part :slight_smile:

Nice writeup, this matches my experience too. Where Fable 5 really earned my trust was firmware development work. Embedded code is unforgiving, one wrong assumption about a register, a timing loop, or a memory constraint and you are reflashing devices in the field. Fable actually reads the surrounding code before suggesting changes, and it flags the risky stuff instead of confidently generating something that compiles but breaks on hardware. That judgment piece is what separates it from just being a fast autocomplete. Good to hear it is holding up on a production system with real customers, that is the real test.