13 October 2025 Hassan Syed

Cursor AI Said 'It's Working' 15 Times Before It Actually Worked - My Gen AI Development Reality Check

After spending several weeks developing and deploying a solution using Cursor IDE and Claude Sonnet 4.5 (and Sonnet 4, Grok 4, GPT 5, Gemini Pro - almost all of them do that), I wanted to share the real lessons learned f

Author

Hassan Syed

AI Architect | Generative AI SME | Azure Certified Solution Expert | Enterprise Systems | IoT Solutions | Big Data | Digital Transformation Leader | Integration Architect | Hands-on| Mentor

Helping organisations turn AI ambition into secure systems, confident teams, and measurable value.

Hassan Syed is an enterprise AI architect, founder, transformation coach, teacher, and writer with more than 20 years of experience designing and delivering complex technology systems.

Follow on LinkedIn

After spending several weeks developing and deploying a solution using Cursor IDE and Claude Sonnet 4.5 (and Sonnet 4, Grok 4, GPT 5, Gemini Pro - almost all of them do that), I wanted to share the real lessons learned from working with Gen AI development tools in production.

This isn’t another “AI will replace developers” article. It’s a practical reflection on what actually works when building real applications with AI assistance.

The Partnership Model That Works

The biggest revelation? You’re still the architect. Think of Cursor as a brilliant junior developer with perfect coding knowledge and skills but limited business judgment and long term memory.

The “It’s Working” Problem

One recurring challenge was premature AI celebration. Cursor would make a change, see partial success, and declare the issue resolved—only for us to discover the fix was incomplete.

Where AI Truly Excelled

Small-Scale Application Development: Cursor shines for rapid prototyping and small applications. It excels at creating UX mockups, interactive components, and setting up complete project scaffolding from scratch.

Infrastructure Setup: Cursor was exceptional at generating Cloud resources, Azure Devops Pipelines/GitHub Actions workflows, and monitoring configurations. Tasks with established patterns and clear documentation.

Pattern Implementation: When I described needing JWT authentication middleware, Cursor generated clean, secure code following industry best practices.

Interactive Design Elements: Creating responsive UI components, setting up routing, and implementing basic user interactions happened remarkably fast with minimal guidance.

Project Scaffolding: From initial project structure to package.json configuration, Cursor handled the tedious setup tasks that usually consume the first hours of any new project.

Systematic Debugging: Given error logs, Cursor consistently suggested logical troubleshooting approaches and relevant diagnostic commands.

Where Human Judgment Stayed Critical

Architecture Decisions: Questions like “Should we use Cosmos DB or SQL Server?” require understanding business growth plans, budget constraints, and team expertise.

Performance Trade-offs: Choosing between different implementation approaches based on expected load and maintainability needs human context about business priorities.

Complete Verification: While Cursor could run individual tests, verifying the entire user journey worked required human oversight and real-world testing.

The Workflow That Emerged

Human defines project scope and success criteria Cursor generates initial scaffolding, mockups, and basic structure Human reviews architecture and guides business logic decisions Cursor implements features, creates test scripts, and suggests solutions Human verifies complete user journeys and production readiness

This partnership consistently delivered faster results than either working alone, especially in the early development phases where Cursor’s scaffolding abilities shine.

Key Lessons Learned

Start Small, Scale Smart: Cursor excels at rapid prototyping and small applications. Use it to validate concepts quickly before scaling to complex enterprise solutions.

Test Everything Twice: Don’t trust “deployment successful” messages. Always verify the actual running application works as expected.

Define Success Clearly: Vague goals lead to incomplete solutions. Be specific about what “working” means for each feature.

Leverage AI for Scaffolding: Let Cursor handle project setup, UX mockups, and boilerplate code while you focus on business logic and architecture.

Understand Every Change: Never deploy generated code you don’t understand. Ask Cursor to explain complex implementations step by step.

Document Decisions: Keep notes about why you chose specific approaches. Future you will thank present you.

Summary: AI vs Human Strengths

The Bottom Line

Working with Cursor and Claude Sonnet 4.5 (4.5 is outstanding!) genuinely accelerated development, particularly in the initial phases. The key insight: it’s a powerful collaboration tool that amplifies human capabilities rather than replacing developer judgment.

Cursor shines brightest when building from zero to MVP. For small-scale applications, interactive prototypes, and rapid scaffolding, it’s genuinely transformative. The sweet spot is leveraging AI for the foundational work while maintaining human oversight for strategic decisions.

The future isn’t AI replacing developers—it’s developers with AI assistants moving from idea to working prototype faster than ever before.

Hassan Syed is a Solution Architect and Generative AI SME specializing in enterprise systems and digital transformation. Connect with him on LinkedIn for more insights on AI-assisted development.

Originally published via LinkedIn. View source ↗
← Back to Articles