Cursor builds an RS485-to-MQTT hardware bridge
Bilsland reports using Cursor agents to implement a Rust serial-to-MQTT bridge, with human-designed networking, failure handling and a React control interface.
linkedin.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Using [authorised protocol PDF], implement a Rust RS485-to-MQTT prototype for [device]. Explain register mapping and checksums, follow [failure rules], and prepare tests for my hardware validation before enabling control.
Reach people exploring AI agents
See placements →What happened
Bilsland reports using Cursor agents to implement a Rust serial-to-MQTT bridge, with human-designed networking, failure handling and a React control interface.
Reported by Robert Bilsland on LinkedInHow the agent did it
- Supply the proprietary RS485 protocol PDF to Cursor.
- Generate a Rust bridge for serial commands and register mapping.
- Connect the Raspberry Pi edge to an MQTT cloud broker.
Where a person stepped in
Bilsland chooses the stack, defines graceful-failure behaviour and takes responsibility for validation and testing of the physical system.
Robert Bilsland describes using Cursor’s agents to connect a proprietary industrial device to a cloud control plane. He supplied the raw protocol PDF and asked for a bridge from a timing-sensitive RS485 link to MQTT. The physical edge used a Raspberry Pi; the wider system included WireGuard, an AWS EC2 broker and a React/TypeScript interface over WebSockets.
Cursor generated Rust code for the serial protocol and hardware-register mapping. Bilsland designed the architecture and resilience rules, including watchdogs that identify where a network failure occurred rather than reporting one undifferentiated connection error. That distinction mattered because the transport could switch between mobile data and Starlink, with high latency and unreliable links.
He reports achieving cloud connectivity and remote control in a few weeks, comparing that with an expected months-long effort. The timing and performance claims are his own account, not independent measurements. The useful detail is to provide the exact protocol specification while retaining human ownership of physical constraints, transport choices and failure behaviour. The source does not include a test matrix or establish deployment safety for another device. A similar project requires validating the generated parser and device commands against real hardware before granting control.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: LinkedIn · Robert Bilsland on LinkedIn
Frequently asked questions
What specification did the agent receive?
Bilsland supplied the raw proprietary serial-protocol PDF, including the commands needed to bridge RS485 data to MQTT.
Which decisions stayed with the architect?
The stack composition, WireGuard transport choice, physical constraints and graceful-failure rules were human decisions.
Are the speed claims independently tested?
No. Bilsland reports a few-week outcome and his comparison with months; no independent benchmark or complete test matrix is supplied.
More details
Remaining steps
- Design watchdogs and transport failure behaviour as the human architect.
- Integrate and validate the React/WebSocket control interface.
About this account
Cursor’s agents read the protocol specification and implement low-level bridge code while the architect supplies system intent and resilience constraints.
Firsthand account by Robert Bilsland on LinkedIn; original publication date verified against Primary page publication date. Outcomes are the author’s reports, not independently reproduced. Suggested prompt written by Agents In Use; not the author’s original prompt.
Credit: Robert Bilsland on LinkedIn
Source published 19 May 2026
Source checked 14 Oct 2026
Tools and topics
Cursor · Rust · RS485 · Raspberry Pi · WireGuard · MQTT · React
Industrial IoT · Protocol parsing · Failure handling
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.