This post was originally published on Mohamad Alsabbagh's Blog.
AI coding assistants boost raw throughput, but they can silently amplify technical debt when teams skip rigorous spec-first planning. GitClear data shows refactoring dropped about 32% from 2021 to 2023, while code churn is projected to nearly double by 2024 compared with its 2021 baseline.
The fix isn't less AI — it's better process architecture designed around how engineers actually make decisions under pressure.
As AI coding assistants have become ubiquitous, I've noticed an alarming pattern: engineers are increasingly over-relying on AI tools to accelerate the "PR generation" phase while drastically reducing the time spent on planning, architecture, and researching edge cases.
Historically, writing syntax was the bottleneck. But that friction also forced real-time architectural reflection. Today, AI can generate hundreds of lines in seconds, stripping away that automatic reflection period and creating a false sense of productivity.
Key Insight: DORA's 2025 research frames AI as an amplifier: it magnifies existing organizational strengths and weaknesses. Without strong platforms and workflows, local productivity gains can turn into downstream disorder.
The Bottleneck Has Shifted
Previously, the time engineers spent manually writing code also served as an essential planning phase. Writing proper code was slow and deliberate — it forced real-time architectural reflection. That organic slowdown is gone. Instead of investing time upfront to deeply understand scope, teams are now maximizing PR generation.
The Foundation of an Engineer
In this era, foundational engineering skills matter even more. The most critical ones:
- Deep Problem Framing: Articulating intent precisely and understanding the "why" before the "how."
- Architectural Soundness: Understanding systems at a macro level, where AI still lacks persistent context.
- Validation & Edge Case Analysis: Predicting where a generated solution will fail under load or security constraints.
- First-Principles Debugging: Maintaining the ability to drill down into the compiler or network layer when AI abstractions leak.
Defining a Process for the Human Brain
We cannot simply rely on engineers to "try harder" to plan. Under delivery pressure, people tend to choose the easiest available path. Autocomplete made that path hitting "Tab"; agentic tools make it delegating an entire task before the context is clear.
Spec-First Agentic Workflow (Recommended)
To counteract this, we must build enterprise processes that match how our brains work:
- Ticket assigned
- Write spec / TDD / ADR: Goals, non-goals, alternatives, edge cases.
- Architectural review: Sign-off before any code is generated.
- AI generates implementation: Guided by precise spec and constraints.
- Human reviews security + arch: AI output treated as untrusted draft.
- Merge: Maintainable, intentional codebase.
Quality Signals: Observed Risk vs. Suggested Targets
| Metric | Observed Risk (No Spec) | Suggested Target |
|---|---|---|
| Copy/Pasted Code | 8.4% → 11.6% | ↓ Stable ~8–9% |
| Refactoring | 24.8% → 13.4% | ↑ Maintained 20–25% |
| Code Churn Rate | 3.6% → 7.1% | ↓ Near baseline ~3–4% |
| AI Trust Paradox | No review gate | ↑ Mandatory arch sign-off |
What to Do Next
1. Spec-First by Default
Before any ticket is picked up, require a written spec.
🌱 Spec Kit
Build high-quality software faster.
An open source toolkit that allows you to focus on product scenarios and predictable outcomes instead of vibe coding every piece from scratch.
Table of Contents
- 🤔 What is Spec-Driven Development?
- ⚡ Get Started
- 📽️ Video Overview
- 🌍 Community
- 🤖 Supported AI Coding Agent Integrations
- 🔧 Specify CLI Reference
- 🧩 Making Spec Kit Your Own: Extensions & Presets
- 📚 Core Philosophy
- 🌟 Development Phases
- 🎯 Experimental Goals
- 🔧 Prerequisites
- 📖 Learn More
- 📋 Detailed Process
- 💬 Support
- 🙏 Acknowledgements
- 📄 License
🤔 What is Spec-Driven Development?
Spec-Driven Development flips the script on traditional software development. For decades, code has been king — specifications were just scaffolding we built and discarded once the "real work" of coding began. Spec-Driven Development changes this: specifications become executable, directly generating working implementations rather than just guiding them.
⚡ Get Started
1. Install Specify CLI
Requires uv…
2. AI PR Review Checkpoint
AI-generated code should not merge without architectural sign-off. Layer your review stack:
- PR-Agent — Great for
/reviewand/describecommands. - Semgrep CE — Best for custom security rules in CI gates.
- SonarQube Community — Mature rule-based quality gates.
3. Track Deployment Rework
Track deployment rework alongside the classic four DORA metrics. Use tools like DORA Four Keys to monitor if your rework rate is rising—a key warning sign of AI-amplified debt.
Research Archive
I'm curious—how has your team changed your code review process since adopting AI tools? Do you find yourself spending more or less time on architecture? Let's discuss in the comments!
Top comments (0)