AI Is Not Your Architect

A field guide for the AI-assisted developer, by Michael Marquez.

Back to Press

A Field Guide for the AI-Assisted Developer — AI Is Not Your Architect

By Michael Marquez · First Edition · 2026 · Crumby Labs Press

About the Book

AI is generating code faster than we could have ever imagined. AI Is Not Your Architect is a field guide for building well anyway. It grew out of three decades of production engineering, from early network servers on Solaris and FreeBSD through ecommerce infrastructure, payment systems, digital asset security, and the decentralized proof-of-origin platform Bitcrumb, built at Crumby Labs.

Its argument has not changed with the arrival of AI coding tools: architecture decides which systems last and which ones collapse under their own shortcuts. What has changed is how quickly you find out whether you made the right call. AI Is Not Your Architect lays out the discipline that AI cannot supply on its own, the abstraction, ownership, and judgment that decide whether a system survives past its first working demo, and shows how to keep that discipline in place without giving up the speed AI offers.

What's Inside

Four parts, ten chapters, moving from the argument itself to a complete application built end to end.

Part 1: The Reality of AI-Assisted Development

Establishing the thesis: AI generates syntax, not architecture, and the discipline that used to happen by accident now has to happen on purpose.

  1. AI Did Not Remove Software Engineering

Part 2: Designing Software AI Can Safely Help Build

Architecture patterns and principles for the moment before generation starts: abstraction, layering, contracts, and documentation.

  1. Abstraction Before Implementation
  2. Designing AI-Friendly Architectures
  3. Documentation Is No Longer Optional
  4. Working With AI Without Becoming Dependent

Part 3: Real Engineering Practices

Where the theory meets code: maintainability, testing, and security for systems AI helps build.

  1. Building Maintainable Systems
  2. Testing AI-Generated Code
  3. Security and Trust

Part 4: Practical Workflows

Tactical and end-to-end: the eight-step workflow, and a complete build that proves it.

  1. A Real AI Development Workflow
  2. Building a Real Application (End-to-End)

Chapter 10 is built alongside a companion repository — a working application you can run, read, and modify.

Read an Excerpt

From Chapter 1, AI Did Not Remove Software Engineering

Here is what a bad AI-assisted project looks like from the outside: fast commits, a green CI badge, a demo that works. Here is what it looks like from the inside, six months later. Nobody can explain why the UserService imports the PaymentGateway. Three different files implement "the same" validation logic slightly differently. The last four bug fixes each broke something in a module nobody remembers writing.

Ask an AI coding tool to add user authentication and it will hand you working code. It will not hand you a decision about whether authentication belongs in a middleware layer, a dedicated service, or scattered across route handlers. It will simply pick one, silently, based on whatever pattern is statistically nearby in its training data.

AI is a junior engineer with infinite speed and zero accountability.

Key Takeaways

  • AI generates syntax, not architecture. Every structural decision it does not see explicitly, it makes for you, silently, based on statistical convenience, not on your system's actual constraints.
  • Skill is largely the ability to recognize bad structure, not just produce code. AI accelerates production for everyone equally. It does nothing for recognition, which is why the gap between disciplined and undisciplined developers widens, not narrows, with AI assistance.
  • Prototypes and products have different jobs and different acceptable trade-offs. AI-generated code looks equally confident either way. The distinction has to come from you, decided before you start generating.

About the Author

Michael Marquez is the founder of Crumby Labs, where he built Bitcrumb, a decentralized proof-of-origin platform, and TapIn Wallet on top of it. He wrote his first program on a Commodore 64 in 1987, and three decades later still works from the belief that started there: architecture, not tooling, is what decides whether software lasts. AI Is Not Your Architect is his case for building that way even when AI makes it tempting not to.

Want to talk about the book, or about working together? Get in touch.

Get the Book

AI Is Not Your Architect is available now as a PDF.

  • Full PDF — all 10 chapters
  • Companion repository with the working application from Chapter 10
  • No subscription, no expiry — yours to keep

Processed securely through Stripe. You'll receive an instant download link and repository access once payment is confirmed.