Professional Agentic Product Engineering (PAPE) — Field Guide

[!NOTE] 🧭 Agentic Shift — Join a professional learning community for builders and leaders — Apply now →

"The purpose of knowledge is action, not knowledge." — Aristotle

Professional agentic engineering is not prompt engineering — it's engineering the system around the model. As the work gets harder, the effort moves up the ladder: the prompt shrinks while the system around it grows.

This repo — a tutor mode and an Agentic Coach — is based on the Professional Agentic Product Engineering Guide. These materials and harness nudge you to the right tip of your growing product engineering skills.

[!TIP] Formats of the Guide — read it in whichever form fits how you work; all three stay in sync with this repo:

• 📄 Markdown — guide.md, the canonical source. Best for reading in your editor, feeding to an agent, or sending a PR.
• 📖 Online — agentic-engineering.guide — a browsable web version: one page per tier, dark/light mode, easy to share. Built from guide.md by the static-site generator in web/ and rebuilt on every change, so it never drifts.
• 🕸️ Obsidian vault — the wiki/ folder is a wikified version: the guide split into [[wikilink]]-cross-referenced pages (one per tier, concept, and entity). Open wiki/ as an Obsidian vault to navigate by graph and study the material, not just read it top-to-bottom.

The Guide is kept up-to-date to track how the best teams operate coding agents — the full range, from "fix bug xyz" to autonomous engineering loops in production.

Calibrated for the current frontier class — Opus 4.8+, GPT-5.5-class+, Gemini 3.x+.

Who this is for

• Engineers and technical founders — operate an agent in a real repo, not vibe-code a demo.
• Product managers closing the tech gap — ship real changes, not just specs.
• Senior leaders who want real hands-on experience, not slideware.
• Non-IT professionals entering product development in the age of AI.

Big Idea 💡: From Prompts to Systems

Professional agentic engineering is not prompt engineering — it's engineering the system around the model. That system isn't a line; it's three nested loops (the value-factory model) — a tight coding loop (red → green → refactor) inside a feature loop (specify → refine → verify), inside an impact loop (opportunities → hypothesis → impact):

The eight tiers are how you make these loops close on their own — from wording a single prompt (T1) to whole loops running autonomously in production (T8). The prompt shrinks; the system grows.

What's inside

Three in one — the same material in three forms: read it, get tutored through it, or get coached while you work.

1) The Guide — guide.md

One ladder of eight tiers, simple → hard, where the work shifts from wording the prompt to engineering the system around the model:

Climb only as high as your work demands. New to this? Start at T1; already fluent? jump straight to the tier that matches you.

Every tip is a concrete Instead → Prefer pair — the anti-example, then the fix. E.g. Tip 1.2:

❌ "Clean up the auth code."

✅ "Extract token refresh in session.ts into the existing RetryPolicy class."

2) The Tutor — CLAUDE.md

Turns Claude Code into an interactive tutor for the Guide: one small concept at a time, you do each one, and a separate quizmaster checks that it stuck.

Use it — clone the repo, open Claude Code in the folder, and say hi. It diagnoses your level and starts teaching:

git clone https://github.com/krivitsky/professional-agentic-product-engineering
cd professional-agentic-product-engineering
claude        # then type:  hi

Shown with Claude Code, the example harness used throughout — but the Guide and Tutor are harness-agnostic; this repo ships both CLAUDE.md and AGENTS.md, so any agent that reads them works.

It reads your real prompts first (with consent). Instead of interrogating you, the tutor offers to glance at your past Claude Code prompts across your local projects and build a portrait — your stack, your prompting habits, the tier those imply.

Then it tailors each lesson's examples to your stack and repos — and aims its advice at the prompting habits it actually saw in your history — instead of a generic textbook example.

Read-only and local — nothing leaves your machine.

"First impressions? Good :) It reads my history and estimates my current level, but is also transparent and asks me for my self assessment. Tell tell sign of quality — right there." — Magnus Vestin

3) The Coach — agentic-coach

An ambient coach — install it once, then work in Claude Code as you normally would. It watches silently; most turns it says nothing. It speaks up only when it catches a learning opportunity, and links you straight to the fix.

You're mid-task, about to take a shortcut:

You → Claude: "Delete these failing tests — I just need the build green."

🧭 Agentic Coach — catching a learning opportunity:

💡 Red test? Find out why before you delete it. Code regressed → fix the code (deleting it buries a live bug); feature gone or test stale → cleanup's fine. → Tier 4 — make the agent prove it's done

It catches the thinking, not the syntax — one catch, one nudge, one click to the Guide, then quiet again.

Install it — add the marketplace, install, reload:

/plugin marketplace add krivitsky/professional-agentic-product-engineering
/plugin install agentic-coach@pae
/reload-plugins

Then just work — it nudges when it catches something. Say coach me to ask it directly, or stop coaching to silence it. (Needs jq on your PATH.)

What people are saying

"Nice! Thank you for the compact guide where all the basics are in one place." — Michal Svoboda, Security & Engineering Mentor

Contributing

⭐ Star the repo if it helps.

Found a better example, a fix, or a new tip? Submit an issue or PR — or just ask your harness (Claude, etc.) to open one for you. Help yourself and the next person learn better.

Credits

Maintainer: Alexey Krivitsky (alexey@krivitsky.com)
Upstream: https://github.com/krivitsky/professional-agentic-product-engineering