About Services Products Work Blog Careers Contact Japanese
← BLOG
ai 2 min read

The Reality of AI-Driven Efficiency — Running 0.6 FTE Work with 0.2 FTE

Can AI truly reduce workload? Real project data and practical methods for genuine efficiency gains.

#AI#productivity#solo-founder#real-data

Introduction

You hear it all the time: “AI cut our workload by 10x.” But reality isn’t that simple. In this article, I share real numbers from actual project data on AI-driven workload compression.

Baseline

  • Based on actual results at ANDOOR (solo consultancy)
  • Period: Q3-Q4 2025
  • Comparison: Estimated effort if the same scope were handled by a traditional team of 3-4 people

Where Compression Worked

Code Generation & Implementation (Compression: 70-80%)

The combination of Claude Code + GitHub Copilot delivers reliable gains in implementation effort. The effect is especially pronounced for boilerplate code and routine CRUD implementations. However, architectural design and complex business logic decisions remain firmly in the human domain.

Documentation (Compression: 60-70%)

AI dramatically accelerates first-draft creation of proposals and reports. But the final adjustments — incorporating client-specific context and nuance — remain human work.

Research & Information Gathering (Compression: 50-60%)

AI is effective for initial screening of web research and academic papers. But the reliability assessment and contextual filtering of information falls to humans.

Where Compression Is Difficult

  • Client dialogue and relationship building
  • Structuring ambiguous requirements
  • Designing proposals that account for organizational politics

Conclusion

AI is not a “human replacement” — it’s an amplifier. Running 0.6 FTE work on 0.2 FTE is possible, but the prerequisite is having a clear division of what AI handles versus what requires human judgment.