How to Counter a Boss Who Claims a Product Manager Using Cursor Can Replace Developers

When a leader suggests that a product manager using Cursor can replace programmers, the first question is what counts as a "complete project". For a simple demo that only needs to be clickable, Cursor can indeed generate a login page, list page, and even a database in minutes, making the process appear ten times faster than traditional development.

However, the author points out that the real nightmare begins when the demo is expected to run in production. In an embedded Linux scenario, generating code for a configuration‑file parser, JSON handling, logging, and unit tests may take only half a day, but the resulting code is still far from production‑ready. Critical issues such as memory leaks, thread deadlocks, network jitter, unexpected reboots, hardware compatibility, and data‑security concerns remain unaddressed.

The author illustrates this with a concrete automotive‑terminal example: a product manager defines requirements like CAN‑message collection, cloud upload, offline caching, OTA upgrades, and automatic recovery. Cursor can produce code for CAN parsing, MQTT upload, SQLite caching, and OTA flow, and the demo runs. Yet after 72 hours on a vehicle, problems surface—message‑queue overflow during CAN bursts, thread accumulation during 4G reconnections, database corruption after power loss, failed OTA checksum verification, and excessive socket reconnection attempts.

These failures arise because AI lacks knowledge of hardware constraints (e.g., 256 MB memory), strict recovery time limits (30 seconds), and domain‑specific safety requirements. Engineering is not just typing code; it involves translating vague goals like "stability" or "speed" into measurable metrics, designing monitoring, alerts, fallback, retries, and rollback strategies.

The author stresses that leaders often mistake the visible layer of software—UI, APIs, features—for the whole system, and that layer is the most vulnerable to AI‑generated shortcuts. While tools like Cursor and Copilot can eliminate repetitive CRUD work, they do not replace engineers who ensure reliability, security, and performance.

To rebut the boss, the author suggests a measured response: acknowledge that Cursor speeds up prototyping, but a full project must include requirement clarification, architecture design, data security, exception handling, performance testing, automated testing, deployment monitoring, and long‑term maintenance. Propose a pilot on a well‑scoped module with mandatory code review, automated tests, stress tests, security checks, and a 30‑day stability requirement before claiming any replacement.

In conclusion, AI can generate code, but turning that code into a stable product still requires engineers. The real value lies in using AI to automate standardized tasks while engineers focus on judgment, system design, and responsibility.