The Bottleneck Is Never the Stack.

Opening

1. How to Use This Book

   Read the argument in order once. Return by the decision in front of you.

2. The Bottleneck Is Never the Stack

   A year of writing, one argument, and a working theory of where software is going.

Part I · The New Bottleneck

1. The Frame Is the Bottleneck

   As agents get better at execution, the constraint moves upstream to the human framing of problems.

2. The Dark Factory Model

   The lights are off because the controls are on. Dan Shapiro's dark factory, StrongDM's working version, and what the model demands before you trust it.

Part II · The System

1. The Agentic SDLC: Uniting the AI Tool Sprawl

   Every tool in your pipeline is now an AI agent trying to do everything. Here is how to draw the boundaries.

2. The Spec Is the Product

   Code used to be the durable asset. In an agentic SDLC, the code is exhaust and the spec is the product.

3. Spec, Standards, Specialists: How I Actually Build with AI

   One spec, one standards file, a handful of specialist agents: the whole workflow, including the five costs it charges.

Part III · The Practice

1. Worked Example: From Vague Request to Governed Release

   A composite notification-preferences feature, traced through a proposed delivery path.

2. The Confident Defect

   The worked example showed the intended loop. This one shows what happens when it fails, and how mature incident practice extends to an agentic system.

3. Pipeline-First Is Not a DevOps Initiative

   The technical case is easy; the human case decides it.

Part IV · From One Team to the Enterprise

1. Day 31

   The loop is clear at one-team scale. The next problem is making it repeatable across an institution without scaling confusion faster than clarity.

2. Where You Actually Are

   Five conditions decide whether the operating model takes root or gets rejected. Assess them before you announce anything.

3. Redesign the Work at Enterprise Scale

   Workflows become paved paths, decision rights become policy, ownership becomes inventory, and evaluation becomes the instrument panel. The org chart still moves last.

4. The Org Model: Structure Follows Evidence

   Team shapes, the AI engineering capability, and federated standards: how to draw the boxes once the redesigned work has told you where they go.

5. Roles, Careers, and Apprenticeship

   The four responsibilities the operating model creates, the apprenticeship it almost destroys, and the capacity the institution has to fund for real.

6. Governance That Ships

   Provenance, liability, model risk, vendors, and the control environment around generated code.

7. The Scorecard: Measuring Evaluation Capacity, Not Activity

   The instruments that tell you whether the operating model works, the one-page scorecard executives review, and the gaming the design has to survive.

8. The Roadmap: From One Team to the Portfolio

   Four phases governed by evidence gates. Growth is a consequence of health, not a schedule.

Part V · The Bargain

1. The Bargain Underneath

   The subsidy that made AI feel free is ending, and the full costs of inference, supervision, lost judgment, and workforce transition are becoming visible. The bill lands on the people building the systems.