AI in the Portfolio. Delivery Metrics Unchanged.

What You Will Read in This Brief

• The AI Productivity Paradox -- 84% of developers use AI coding tools, yet the 2025 DORA report ties a 25% rise in AI adoption to a 1.5% drop in delivery throughput and a 7.2% drop in stability. Faros AI saw 21% more tasks and 98% more PRs merged with zero change in organizational delivery. The brief names the root cause: AI tools layered on top of processes designed for a pre-AI world.
• Two Spec Types: What to Change vs. How the System Is -- Feature Specs describe what to change and flow through the pipeline once. Anchor Specs describe how the system is and update deliberately. AI agents need both: a clear instruction and an accurate model of the system they are touching. Documentation drift defeats AI's value at scale; the two-spec structure is the fix.
• Preserving the Why -- Working systems are easy to inspect; the reasoning behind them is not. SDD captures the why systematically: Decision Rationale (alternatives, constraints, assumptions, trade-offs), Gate Annotations (the reason a gate approved despite a flag), Post-Completion Retros (what surprised, what to repeat). The result is a system whose decisions are inspectable months and years later -- by humans and by AI agents alike.
• The Four-Gate Pipeline -- PM-Spec, Architect-Review, Implementer-Tester, Deployment. Four human-judgment checkpoints that prevent AI-generated code from compounding errors downstream. Deliberately linear with controlled feedback: when a gate rejects work, the rejection includes specific remediation, preventing endless cycling.
• The Adaptive Pipeline: Complexity Tiers That Match the Work -- Four tiers (Trivial, Standard, Complex, Critical) scale process intensity to the risk profile of the change. A config update and a payment system change get different treatment by design. Stop over-engineering simple changes and under-engineering the ones that carry real risk.
• The Coach in Every Project -- A methodology document on a shelf is overhead. A methodology your team actually applies, consistently, without a senior engineer in every room, is leverage. SDD ships as a kit that installs into a project alongside Claude Code. The methodology is taught and enforced by the methodology itself, not bottlenecked on whether your team reads the document.
• Field Evidence -- Four observations from the alpha kit, with honest framing on sample size: the coach applied the methodology unprompted on the hardest-to-explain feature; an experienced engineer recognized the governance as a feature, not friction; a missing methodology concept was independently invented in the field; the kit's own learning loop is operating on itself. Directional, not definitive -- and stronger than any scripted demo.