THE LOOP

Five stages for turning artificial intelligence into work that holds up.

Jeffrey Alleyne

Confident is not the same as correct.

You ask for help. You get a clean, confident answer. You follow it exactly. It does not work. Three rounds later you find out the whole conversation was about a Mac and you are on Windows.

Nobody lied to you. You never mentioned the island.

THE LOOP is a working method for using AI on real projects: how to load a session, how to constrain it, how to direct it, how to check what it hands back, and how to keep the record honest as things change. Every technical term gets a plain-English box. Every claim is one you can check.

Options are cheap and nearly worthless. Constraints do all the work.

The method

One loop, five stages, run in order and run again.

  1. FRAME

    The model knows nothing about your machine, your project or your life, so it guesses. Give it the environment and the names before it says a word.

  2. CONSTRAIN

    Set the limits before you ask for ideas. Name a comparative bar instead of an adjective. Kill weak candidates in one line.

  3. GENERATE

    Terse directives, scoped to one file or one decision, with the output shape locked. Short prompts work only after the first two stages are loaded.

  4. AUDIT

    Never trust the diff. Audit first, then ask for the whole corrected file, because snippets hide the damage.

  5. RE-BASELINE

    State decays. Freeze finished phases and reset what is current, so session nineteen does not quietly contradict session four.

Then the loop starts again with a truer frame.

Three ways to read it

If you build

Four systems built end to end, with the real architectures, the real constraints, the cost figures, and a section in each on where the limits showed. Runnable code that needs no network and no installed package.

If you teach

Every chapter opens with five assessable learning objectives and closes with five graded exercises. Appendix F maps the book to the UNESCO AI Competency Framework for Students and to a general undergraduate AI-literacy syllabus, with three ready-made course shapes.

If you are new to this

The narrative is written for non-technical readers, with technical sidebars for the people who want them. Jargon is explained in plain terms, with a joke where one is earned.

What is inside

26chapters
5parts
7appendices
20+runnable code blocks
9figures
Part OneGround truth. What these systems are, how they are built, where their data comes from.
Part TwoThe method. One chapter for each stage of the loop.
Part ThreeThe human side. Accountability, bias and fairness, and the physical and labor footprint.
Part FourDown the rabbit hole. Four systems, built end to end.
Part FiveThe fleet. What it bought, rights and liability, designing and evaluating a system, and a thirty-day plan.
AppendicesPattern library, audit checklist, environment header template, glossary, the stack, curriculum map, where to go next.

Get THE LOOP

Paperback, 6 by 9 inches. Start with the Environment Header, the six-line block that opens every session, and see what changes.

The author

Jeffrey Alleyne is an engineer and founder based in New Jersey. He works across software, autonomous systems and embedded hardware, and wrote this book to make the results reproducible for anyone willing to run the loop.

THE LOOP, first edition, 2026. The book is not legal or financial advice, and results are not guaranteed. Case studies are anonymized composites.