Chapter 33: Key Takeaways

Project Planning and Estimation -- Summary Card

  1. AI acceleration is asymmetric, not uniform. The Asymmetric Acceleration Principle states that AI speeds up implementation by 3-10x while leaving requirements gathering, architectural design, stakeholder communication, and production deployment largely unchanged. Planning must account for this unevenness by applying different acceleration factors to different task types.

  2. Planning is more important in the AI era, not less. Because AI compresses implementation time, the relative proportion of effort spent on planning, design, and communication increases. A project where implementation once consumed 60% of total effort may now see implementation consume only 30%, making the other phases proportionally more critical to get right.

  3. Classify tasks into AI acceleration tiers. Tier 1 tasks (CRUD, boilerplate, test generation) accelerate 3-10x. Tier 2 tasks (complex algorithms, integration code) accelerate 1.5-3x. Tier 3 tasks (requirements, architecture decisions, stakeholder alignment) accelerate only 1.0-1.5x. Apply the appropriate factor to each task rather than using a single blanket multiplier.

  4. Use the Three-Point AI Estimation Method. For each task, estimate optimistic (AI works perfectly), realistic (some iteration needed), and pessimistic (significant human intervention) scenarios, then apply the PERT formula: (Optimistic + 4 * Realistic + Pessimistic) / 6. This captures the higher variance that AI introduces into individual task estimates.

  5. Beware the Velocity Trap. Initial AI adoption often produces a dramatic surge in code output that is not sustainable. Teams that set expectations based on peak velocity will overpromise and underdeliver. Plan for sustainable velocity -- typically 30-50% improvement for mature teams -- not the 60-80% spikes seen in the first few sprints.

  6. Code review becomes the critical bottleneck. AI-augmented developers produce code faster, but code review capacity does not scale proportionally. Explicitly allocate 20-30% of sprint capacity to code review. Track the review queue as a leading indicator of quality problems.

  7. Use AI speed for quality, not quantity. The most effective use of AI-driven acceleration is not to build more features but to build fewer features better. Invest the time saved on implementation into more thorough testing, more careful code review, improved documentation, and better user experience polish.

  8. Manage five AI-specific risks. Quality consistency, architectural drift, vendor lock-in, overconfidence in estimates, and security vulnerabilities in AI-generated code are risks unique to AI-augmented projects. Each requires explicit mitigation strategies in the project risk register.

  9. Adapt Agile ceremonies for AI. Sprint planning should categorize stories by AI acceleration tier. Standups should surface AI successes and struggles. The Definition of Done should include AI-specific criteria such as human code review and security scanning. Retrospectives should recalibrate acceleration factors based on actual performance.

  10. Communicate with the Three Timelines. When presenting schedules to stakeholders, always provide aggressive (maximum AI acceleration), expected (realistic acceleration), and conservative (minimal acceleration) timelines. This transparently communicates uncertainty and sets appropriate expectations.

  11. Track AI-specific metrics alongside traditional ones. Monitor AI Utilization Rate, First-Prompt Success Rate, AI Code Retention Rate, Prompt-to-Production Ratio, and AI Rework Rate. These metrics help the team improve their AI usage over time and provide data for more accurate future estimates.

  12. Build your own acceleration baseline. Generic industry acceleration factors will not accurately predict your team's experience. Track actual task completion times, categorized by task type and AI utilization, for at least 2-3 sprints to develop team-specific factors that reflect your domain, tools, and skill levels.

  13. Quality-Adjusted Velocity reveals the true picture. Raw velocity increases can be misleading if defects also increase. Calculate QAV = Raw Velocity - (Defect Count * Defect Resolution Cost) to measure genuine, quality-inclusive productivity improvement.

  14. The Portfolio Effect smooths individual variance. While AI makes individual task estimates less certain, errors tend to cancel out across many tasks. Project-level estimates are more reliable than task-level estimates, so maintain healthy project buffers while trusting aggregate predictions.

  15. AI does not change methodology fundamentals. Regardless of whether you use Scrum, Kanban, SAFe, or Waterfall, the core principles remain: clear requirements reduce rework, iterative feedback improves outcomes, team collaboration beats isolated work, and measurable progress enables informed decisions. AI is a powerful execution tool, but it does not change why these principles matter.

Use this summary as a quick reference when planning AI-augmented projects or when calibrating your team's estimation process.