Navigating AI Productivity: Implementation vs. Delivery Speed

Managing Expectations in the AI Era

Steve McDougall addresses the gap between AI demos and real-world engineering where implementation is rarely the primary bottleneck. While tools like GitHub Copilot offer 30% speed increases for coding, they do not resolve the complexities of legacy integration and architectural review.

Why This Matters

The productivity revolution promised by AI demos often ignores the reality of mature codebases where security, compliance, and long-term maintenance are paramount. Overestimating AI’s impact on total delivery speed leads to misaligned roadmaps and compromised quality standards.

Key Insights

• A GitHub Copilot study suggests a 30% implementation speed increase for well-defined patterns.
• Implementation speed differs from delivery speed, which includes alignment, review, and integration bottlenecks.
• Greenfield AI demos fail to account for existing codebase constraints, dependencies, and performance requirements.
• Shape Up’s appetite model allows for explicit conversations about scope and time compared to traditional sprint velocity.
• Maintaining quality bars is essential as AI tools increase the volume of code produced without improving its inherent safety.

Practical Applications

• Use Case: Adjusting Shape Up appetites for implementation-heavy tasks while preserving review time.
• Pitfall: Reducing cycle times based on AI speed without accounting for the increased overhead of reviewing AI-generated output.
• Use Case: Communicating capacity shifts to stakeholders by explaining how saved time is reinvested into architectural discipline.

References:

• https://dev.to/juststevemcd/managing-expectations-in-the-ai-era-ogi