education
Menu

SpecStory Education

Field notes · October 2026

Confidence Check.

Real-world conversations with engineering leaders about how their teams handle QA and verification when AI writes more of the code.

Get the field notes

The field guide

25 Patterns in
Agentic Engineering

When agents write the code, your work becomes defining the goal, steering the run, and proving the result. These 25 patterns show you how.

By Greg Ceccarelli · SpecStory Press

Read the book Free PDF · 64 pages
Prefer hands-on learning? Explore the course
Explore the 25 patternsSix themes, from intent to evidence

Part I

Verification Is the Job

  1. Calibrated Distrust
  2. Keep a Source of Truth Outside the Agent’s Reach
  3. The Premise Auditor
  4. The Read-Only Turn

Read how to test AI-generated code.

Part II

Steering, Not Typing

  1. The Interrupt Is the Keyboard
  2. Assert the Ground Truth, Collapse the Branch
  3. Steer by Reference, Not Spec
  4. The Human Is the Runtime Sensor
  5. License the Agent to Ask Before It Acts

Part III

The Brief Is the Work

  1. The Prompt Is an Engineered Brief
  2. Structure Compounds, Incantation Depreciates
  3. Pin the Work to a SHA
  4. The Self-Grading Spec (Goal + Rider)
  5. Commit at Phase Boundaries, Never Push

Part IV

Docs Are the API Between Turns

  1. Write for a Reader Who Remembers Nothing
  2. The Incident Doc Programs the Next Agent
  3. The AS-BUILT Map Is a Test That Can Go Stale

Part V

Code Is Cheap, Understanding Is Dear

  1. Delete and Regenerate from a Clean Base
  2. Band-Aid Is a Verdict, Not a Default
  3. Fix the Generator, Defend the Shape

Part VI

You Run an Org, Not a Pair

  1. The Model Is a Swapped Dependency
  2. The Human Is the Message Bus
  3. Agents Self-Assign; You Pick the Lane
  4. Fan Out Along Non-Overlapping Seams
  5. Make the Agent Show Its Rails

The hands-on course

Hardcore
Agentic Engineering

Build an agent harness, steer real work, and engineer the checks that prove it’s correct. Live, hands-on learning for people who already ship software with coding agents.

Explore the course

View the syllabus and upcoming cohorts on Maven.

  1. Brief the goal.

    Define what “done” means before the agent starts.

  2. Steer the work.

    Shape the context. Catch the wrong turn. Keep the run on track.

  3. Verify the result.

    Check the running software against evidence you can trust.

    Read why coding agents say “done” too early.

Learn with the
SpecStory founders

  • Greg Ceccarelli
  • Jake Levirne
  • Sean Johnson

The paper

Evaluators Let You Automate Your Coding Agents

Agents cannot reliably grade their own work. This paper makes the case for a separate evaluator in the loop, so quality becomes repeatable and agents can run longer with less supervision.

SpecStory Press · Free PDF · 26 pages

Read the paper

The tutorial

Team-based Loop Engineering

A walkthrough for people who already live in Linear. Go from zero to a working plan → generate → evaluate loop that builds a text adventure game using Claude + Linear.

By Jake Levirne · 83 slides · Follow along as you read

Start the tutorial

The reference

Learning Center

Short, plain explanations of the ideas behind the book and the course: how to verify what a coding agent built, what makes a test worth running, and where each check belongs.

Browse the Learning Center

Topics