Many companies are excited about using artificial intelligence to boost developer productivity. The idea is that AI agents can handle repetitive tasks, generate code quickly, and help teams ship features faster. But there’s a darker side to this trend that’s often overlooked. Relying too much on AI can lead to misleading metrics and a false sense of progress that actually hampers quality and long-term success.

The Problem with Metrics in the AI Era

One common shorthand for measuring developer output has been “lines of code” or number of commits. With AI agents, these metrics have become even easier to inflate. Some organizations celebrate high “commit velocities” without realizing they are rewarding quantity over quality. This creates a situation where teams focus on producing more output, not better software, which often results in technical debt and fragile systems.

Developer John Crickett recently pointed out the absurdity of this trend on social media. He noted that managing numerous AI agents and making thousands of commits a day doesn’t mean the work is meaningful. The metrics are already being gamed, and AI makes gaming effortless. If your organization starts praising “speed,” it’s probably measuring how fast you can create liabilities rather than real value.

Why AI Won’t Fix the Core Bottleneck

The initial promise of generative AI was that it would finally help clear backlogs by automating boilerplate code and making development faster. The hope was that teams could focus on high-level design and business logic instead of repetitive tasks. But in reality, the biggest bottleneck in software development isn’t writing code — it’s validating, integrating, and understanding complex systems.

Generating code without proper validation leads to technical debt that accumulates quickly. AI can churn out snippets, but without rigorous testing and deep understanding, these pieces become liabilities. Quality suffers, and fragile systems become the norm. This reality has become clear as we’ve settled into 2026, showing that AI isn’t the silver bullet many hoped for.

Long-term success depends on better validation frameworks. Simply producing more code isn’t enough. Companies need to focus on integration, testing, and understanding the full context of their systems. Otherwise, they risk creating software that looks productive but is actually riddled with errors and vulnerabilities.

Rethinking How We Approach Code and Development

The key shift is to see code not just as an asset but as part of a larger ecosystem that requires ongoing management. Every line of code is surface area for security breaches, bugs, and maintenance. Making code cheaper to produce doesn’t reduce the total effort needed; it often increases it because more liability is created per hour of work.

For years, many companies treated developers as translators of requirements into code. The assumption was that if requirements were clear, coding was simple. Crickett points out that if this is all developers do, they are replaceable by machines. Basic translation is something AI can do easily and tirelessly. But understanding the business context, recognizing risks, or catching mistakes in workflows requires human judgment.

He suggests moving toward “spec-driven development,” where detailed specifications act as constraints for AI agents. These aren’t just Jira tickets but precise, executable definitions that include tests, API contracts, and strict boundaries. Such specifications ensure that AI stays within safe limits and helps deliver reliable, maintainable software. This approach emphasizes quality, validation, and clear expectations over sheer speed or volume.

In the end, the goal should be smarter development processes that leverage AI without losing sight of fundamental engineering principles. Using AI to assist, not replace, human judgment ensures software stays robust and aligned with business needs. The challenge is to create a balanced approach that values quality over quantity in the age of AI-powered development.

Inspired by

• https://www.infoworld.com/article/4135492/ai-agents-and-bad-productivity-metrics.html

Sources

• substack.com
• charity.wtf
• dora.dev