Lightweight AI Workflows Outperform OpenSpec in UI Redesign Experiments

OpenSpec (Spec-Driven Development) Failed My Experiment — Instructions.md Was Simpler and Faster

Developer Incomplete Developer tested the OpenSpec framework on a .NET Razor Pages application to evaluate its efficiency against simple instruction files. The experiment revealed that the formal spec-driven process took two hours and high token counts only to deliver a UI identical to the original.

Why This Matters

Spec-driven workflows attempt to eliminate the unpredictability of vibe coding by forcing structure, but they introduce heavy overhead in the form of proposal reviews and task planning. For many engineering tasks, the technical reality is that the cost of managing these complex pipelines exceeds the time it would take a human to code the solution manually, making lightweight instruction-based iterations more practical and cost-effective for modern AI-assisted development.

Key Insights

• Spec-driven workflows involve a four-step process: proposal generation, specification review, task approval, and agent execution (Incomplete Developer, 2026).
• The Instructions.md approach successfully fixed a broken image uploader bug with minimal token usage and dramatically shorter execution time.
• Using Claude Haiku 4.5 with GitHub Copilot reduced token costs compared to GPT-5.3 Codex but still failed to improve UI results within the OpenSpec framework.
• Manual redesign by a competent developer is estimated at five days, challenging the value proposition of multi-hour AI orchestration for simple UI tasks.

Working Examples

The simplified instruction-based approach used to bypass the heavy OpenSpec framework.

Instructions.md

Practical Applications

• UI redesign using iterative feedback loops rather than static specs to achieve premium design results faster. Pitfall: Relying on heavy frameworks can result in stagnant designs despite high token expenditure.
• Bug remediation in existing backends using concise instruction files for immediate AI agent execution. Pitfall: Over-orchestration of simple tasks leads to developer fatigue and increased project latency.

References:

• https://dev.to/incomplete_developer/openspec-spec-driven-development-failed-my-experiment-instructionsmd-was-simpler-and-faster-3a5d