T The Testing Academy · AI Tester Blueprint
AI TESTER BLUEPRINT · 6 MODULES + 5 PROJECTS

Roadmap to learn Generative AI, AI Automation & Agents for Software Testers

A full curriculum from The Testing Academy: prompt engineering, LLMs, AI test design, n8n + Langflow + MCP agents, and 5 capstone projects you can ship to production.

Modules: 6 Projects: 5 Stack: ChatGPT · Claude · Ollama · n8n · Langflow · MCP · DeepEval

Start here — 4-step roadmap

A condensed view of the full programme. Each column maps to two modules.

1 Prompt Engineering"What we did"
Topic
  • What is Prompting?
  • Steps to learn prompting
  • Zero-shot & few-shot prompting
  • Direct, contextual, role-based, step-by-step
  • Prompting frameworks (SWOT, STAR, CLEAR, PAR)
  • How not to use ChatGPT / AI tools
  • Prompt generators & ready-to-use QA prompts
  • Advanced prompt engineering for QA
  • What is an LLM?
  • Try different LLMs (GPT, Claude, Gemini, DeepSeek)
  • Hosting DeepSeek / Open GPT-OSS locally
  • Best tools overview
  • Low-code tools & n8n automation overview
  • Cursor & Copilot intro
  • Hallucination detection & factuality testing
2 Generative AI"What we found out"
Topics & Projects
  • QA & automation testing principles
  • Requirement analysis with AI
  • Test planning & test strategy
  • Test plan templates / checklist
  • Example test-strategy documents
  • Test-case generation (manual & automated)
  • Test-case tables for Jira / ticketing tools
  • Bug identification and reporting
  • Test closure and reporting
  • API test case + script generation (Postman, REST Assured, Python)
  • Test data management for APIs
  • Automated reporting (Allure, Jenkins integration)
  • API test project structure, builder pattern, auth types
  • Performance & load testing with AI prompts
3 AI Agents & MCP Server"What we learned"
Topics & Project
  • n8n basics + AI agents
  • Code explanation and review with AI
  • AI code error identification
  • AI code optimisation
  • Generating / updating code (Java, REST Assured, etc.)
  • Mock technical interviews (Java, SQL, etc.)
  • API testing with ChatGPT & AI tools
  • Automation project structure & enhancement
  • What is MCP server?
  • Use of MCP server
  • Allow LLM to talk to MCP
  • Selenium AI learning plan
  • AI project setup (Maven, IntelliJ, TestNG, Allure)
  • Sample project structure and documentation
  • Code optimisation patterns (Singleton, Base Test, etc.)
4 Advanced AI Tools & Integration"How we could improve"
  • Web & mobile functional automation using AI
  • Advance problem solving (web projects E2E)
  • SQL query and project automation
  • AI-driven synthetic test data generation
  • ReportPortal: AI-driven test analytics & reporting
  • DeepEval: comprehensive LLM testing framework
  • ATS-friendly resume creation (review, score, update)
  • Resume tailoring for job descriptions
  • Email and follow-up sequence generation
  • Cover letters, cold emails, video / self-introduction scripts
  • LinkedIn profile optimisation (headline, summary, experience)
  • Resume validation and keyword optimisation

Curriculum — module by module

Each module is taught with hands-on labs and graded mini-assignments. The full curriculum runs in cohort or self-paced mode.

AIATB_01

AI Fundamentals

What you'll learn
  • Introduction to the AI Automation Tester Blueprint program
  • Tools you'll need (ChatGPT, Claude, Gemini, n8n, Langflow, IDEs)
  • What is Generative AI?
  • How LLMs work (tokenisation, context, reasoning)
  • GPT vs Claude vs Gemini vs DeepSeek vs open-source models
  • LLM comparison sheet (PDF)
  • Local install LLM (private) via Ollama — Mistral, DeepSeek, GPT-OSS
  • Difference between ML, AI, deep learning
AIATB_02

Prompt Engineering

What you'll learn
  • What is prompt engineering and why we need to learn it
  • Steps to follow for effective prompt engineering
  • Role-based prompting for QA · zero-shot vs precise prompting
  • Write your first QA-style prompt · chain-of-thought
  • Prompt frameworks: STAR, CLEAR, CRISP
  • Context creation via templates (.md files)
  • RICE-POT framework for QA
AIATB_03

Essential AI Tools Setup

What you'll set up
  • ChatGPT setup · Claude & Gemini setup
  • Claude Desktop setup · LM Studio setup · Blackbox AI
  • Using Ollama / LM Studio locally (optional)
  • Anti-Gravity · GitHub Copilot · Cursor for testers
  • Checklist: AI tools setup validation
  • Augment · GitHub Copilot · Amazon Q · IntelliJ · Claude Code

ChatGPT

Daily driver for QA prompts & brainstorming.

Claude

Long-context reasoning, code review, large transcript Q&A.

Gemini

Multimodal — bug screenshots → repro steps.

Ollama

Run LLaMA / Mistral / DeepSeek locally for private data.

Cursor / Copilot

IDE pair programming — generate Page Objects + tests inline.

Claude Code

Terminal agent — scaffold full Playwright frameworks from spec.

AIATB_04

Generation with AI for QA — test design, strategy & automation

What you'll generate
  • Requirement analysis with AI
  • Test plan generation (templates) · test strategy
  • AI-generated test design
  • Bug report · test metrics
  • Automation code generation with AI
  • Claude Code (AI terminal for QA tasks · automation frameworks)
  • Augment overview with framework
AIATB_05

AI Agents · n8n · LangChain · Langflow · MCP & RAG Automation

Introduction to AI Agents
  • What is an AI agent · vs single-shot LLM call
  • Tools, actions, memory, reasoning loops
  • Safety & guardrails

n8n — automate QA workflows

n8n is a node-based workflow tool. Drag a trigger, drag actions, connect with arrows, hit run. We use n8n to chain LLM calls, vector stores, and tools without writing glue code.

1

Install n8n locally

npx n8n on Node 20+. Editor at http://localhost:5678. Full walk-through: n8n tutorial.

2

Install n8n on a VPS

Docker-Compose with persistent volume — same tutorial covers production setup.

3

Understand AI nodes

OpenAI / Anthropic / Ollama chat nodes, plus a generic LangChain agent node.

4

Trigger · AI Agent · Vector embeddings · Memory store · Tools

The five primitives every agent workflow uses. Practise wiring each.

5

Build: Test Plan Generator agent

Cron → fetch a spec doc → LLM generates a test plan → Confluence.

6

Build: Jira Ticket → Test Case agent with RAG

Webhook on Jira → retrieval over past test cases → LLM writes new ones → comment back on Jira.

Build a simple Jira agent — the recipe

Wire these six nodes:

n8n workflow — jira-test-case-agent
// Node 1 — Webhook trigger
POST /webhook/jira-issue
body: { issueKey, summary, description, ac }

// Node 2 — Vector Store retrieval (RAG)
input:  $json.summary + $json.description
search: top-5 from "past-test-cases" collection

// Node 3 — Build prompt with context
system: "You are a QA engineer. Use the retrieved cases as style examples."
user:   "Acceptance criteria: {{ $json.ac }}\nReturn 5 test cases as JSON."

// Node 4 — OpenAI / Claude chat node
output: $json.choices[0].message.content (parsed as JSON)

// Node 5 — HTTP Request
PUT  jira/api/2/issue/{{$json.issueKey}}/comment
body: { body: "Generated test cases:\n{{ $json.testCases }}" }

// Node 6 — Slack notify
channel: #qa-agent
text:    "Generated {{$json.testCases.length}} test cases for {{$json.issueKey}}"
Assignment: Create your own custom AI QA workflow + MCP
  • Pick a real chore: nightly Playwright digest, flaky-test detector, requirement → BDD generator.
  • Wire it in n8n end-to-end. Activate. Take a 60-second video.
  • Bonus: register the workflow as an MCP tool so Claude/Cursor can call it from your IDE.

Langflow

Visual LangChain — same node-graph idea, but every node is a LangChain primitive (chain, tool, agent, memory). Great for prototyping a RAG system without code.

  • What is Langflow
  • Chains, tools, agents
  • Build a QA RAG system
  • Build a "Requirement Understanding agent"
  • AI-powered QA assistant (Langflow) + MCP

MCP Server (Model Context Protocol)

MCP is the open protocol that lets an LLM call your tools. Build a server, plug it into Claude / Cursor / ChatGPT, the model can read your test repo, run a script, post a comment.

  • What is MCP?
  • Building local testing tools using MCP
  • Code review tool using MCP
  • API testing tool using MCP
  • Activity: connect MCP tools to ChatGPT
  • Mini-project: automation healing agent via MCP tools
  • Build your own MCP server in JS / TS
AIATB_06

Advanced AI Tools & Integrations

Cross-stack AI integration
  • AI + Selenium (full Selenium project)
  • AI + API testing workflows — POJO generation, API framework setup
  • AI + Playwright (Stagehand)
  • AI + Appium (vision detection)
CI/CD + reporting with AI
  • Jenkins pipeline + LLM integration · AWS Run
  • GitHub Actions + AI summaries
  • Failure clustering with AI (ReportPortal)
  • Build an "AI debugging step" in a pipeline
LLM testing — DeepEval, TruLens, PromptFoo, OpenAI Eval
  • Why test LLMs? Accuracy, safety, reliability checks
  • DeepEval setup + test writing
  • Test hallucinations & structured outputs
  • LLM test template
  • Overview of DeepEval, PromptFoo, TruLens
AIATB_07

RAG · LangChain · CrewAI · DeepEval · PromptFoo + CI/CD

Where prompt engineering ends and engineering AI systems begins. This module covers retrieval, framework code (LangChain), multi-agent crews, and how you actually evaluate & ship LLM features in CI.

RAG (Retrieval-Augmented Generation)

  • Why RAG — grounding answers in your test docs, runbooks, Jira history
  • Chunking strategies (size, overlap, semantic split)
  • Embedding models (OpenAI, BGE, Nomic, local via Ollama)
  • Vector stores: Pinecone · Weaviate · pgvector · Chroma · Qdrant
  • Retrieval modes: top-k, MMR, hybrid (BM25 + vector), rerankers
  • Build a QA-ops RAG: feed it past test plans → ask it to write a new one
  • Companion: /rag playground · /rag-cheat-sheet

LangChain (the framework)

  • Chains, prompts, output parsers, runnables (LCEL)
  • Retrievers + vector stores wired into a chain
  • Agents: ReAct, tool-calling, function-calling agents
  • Memory: buffer, summary, entity memory
  • LangGraph for stateful, branching agent flows
  • Build: a "Requirement Understanding agent" that parses a PRD into Gherkin

CrewAI (multi-agent crews)

  • Concepts: Agent · Task · Crew · Process (sequential / hierarchical)
  • Role-based agents — Test Lead, BA, Automation Engineer, Reviewer
  • Hand-off pattern: BA writes ACs → QA writes test cases → Engineer writes Playwright code → Reviewer scores
  • Tools per agent (web search, Jira API, file IO, Playwright runner)
  • Build: a 4-agent crew that turns a one-line feature request into a runnable Playwright spec
  • Live test reference: ai-for-qa-part-6-crewai

DeepEval (LLM unit testing)

  • deepeval test run like pytest — but for prompts & agents
  • Built-in metrics: faithfulness, answer-relevancy, contextual recall, hallucination, bias, toxicity
  • G-Eval — define your own metric in plain English, the framework grades it
  • Synthesizer — auto-generate test cases from your docs
  • Confident AI dashboard for run history
tests/test_test_case_agent.py
from deepeval import assert_test
from deepeval.test_case import LLMTestCase
from deepeval.metrics import AnswerRelevancyMetric, FaithfulnessMetric

def test_test_case_agent():
    output = my_agent.run(
        prompt="Generate 3 negative test cases for: login form, max 50 char password"
    )
    case = LLMTestCase(
        input="login form, max 50 char password",
        actual_output=output,
        retrieval_context=["AC: password must be 8-50 chars, alphanum + symbol"]
    )
    assert_test(case, [
        AnswerRelevancyMetric(threshold=0.8),
        FaithfulnessMetric(threshold=0.85)
    ])

PromptFoo + CI/CD with AI agents

  • Declarative promptfooconfig.yaml — providers × prompts × tests grid
  • Compare GPT-4 / Claude / Llama side-by-side on the same evaluation set
  • Assertions: contains, equals, regex, JS snippet, LLM-rubric, embedding-similarity
  • Run in GitHub Actions / Jenkins / GitLab — fail the PR if accuracy regresses
  • Snapshot the prompt + model — prevent silent drift when a vendor updates
.github/workflows/llm-eval.yml
name: LLM eval

on: [pull_request]

jobs:
  eval:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: '20' }
      - name: Run promptfoo
        env: { OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }} }
        run: npx promptfoo eval --config promptfooconfig.yaml --output report.html
      - uses: actions/upload-artifact@v4
        with: { name: promptfoo-report, path: report.html }
The discipline. Treat prompts like code. Version them. Eval them. Fail the PR if quality drops. Without this, you're shipping vibes.
AIATB_08

Vibe Coding AI Agents + Copilot Masterclass

The shift from "ask ChatGPT to write a snippet" to "let an agent edit your repo, run your tests, and ship a PR." This module covers the IDE-resident agents, custom rules files, and the workflows that actually make a tester 5× faster.

Vibe coding — IDE-resident agents

  • Cursor / Windsurf / Anti-Gravity — agent mode, multi-file edits, plan-act loop
  • Claude Code (CLI) — terminal agent, Skills, sub-agents, hooks, MCP servers
  • Aider, Continue, Roo Code — open-source alternatives
  • Project rules: CLAUDE.md, .cursorrules, AGENTS.md for QA constraints
  • Slash commands & reusable prompts for repetitive QA tasks
  • Hooks — auto-run lint, run tests, block secret commits
  • Build: a Playwright skill that scaffolds Page Objects from a URL

GitHub Copilot masterclass for testers

  • Copilot Chat vs inline suggestions vs agent mode
  • Custom instructions per repo (.github/copilot-instructions.md)
  • Generate Playwright specs from a PR diff
  • Copilot for code review — failure root-cause + suggested fix
  • Copilot Workspace — task → plan → multi-file PR
  • Pull-request reviews via Copilot — auto-comment style + risk

Mini-projects in this module

  • Repo "rules file" you reuse across every QA project
  • Skill / slash command: /scaffold-pom from a URL
  • Skill: /triage-failure — reads test report, opens Jira
  • Hook: pre-commit blocks console.log + missing data-testid
  • Mini-MCP: expose your test runner as a tool the IDE agent can call

Project diagrams — sketch view

One whiteboard per phase. Boxes are nodes you build; arrows are JSON or events flowing between them.

0 — The whole course on one page

~14 weeks · 6 phases · 22 projects · 1 GitHub portfolio

Phase 1 Local AI P00–P04 Phase 2 App build P05, P06, P10 Phase 3 n8n flows P08, P09 Phase 4 RAG P11–P15 Phase 5 Agents · MCP P07, P16–P19 Phase 6 Eval P20, P21, P23 → portfolio · interview-ready SDET · 5-7 live agents shipped

Phase 1 — Local AI & prompt foundations

Ollama → prompts → test artefacts. No cloud, no API key.

User story RICE-POT prompt OllamaLlama 3.2 Test cases Selenium / PW Run P01 · P02 · P03 · P04 use this flow with different prompt templates

Phase 3 — n8n visual workflow (Jira → Slack)

No code. Drag · connect · activate.

Cron 9 AM Jira REST Code nodefilter open LLM nodeGroq / Claude HTTPcomment back Slack #qa items flow as JSON · one node in, many items out P08 · P09 · same shape, different actions on the right

Phase 4 — RAG pipeline

Docs → chunks → vectors → retrieve → LLM grounded answer.

Docs / PRDs Chunkersize + overlap EmbedderNomic / OpenAI Vector DBChroma · pgvector Question Embed query Retrieve top-kMMR / rerank LLM+ context Grounded answerw/ citations vector lookup

Phase 5 — Multi-agent crew + MCP tools

Roles + tasks + tools. Hand-off chain ships a runnable Playwright spec.

BA agentPRD → ACs QA agentACs → cases Engineercases → spec Reviewerscore + diff Jira tool PW runner GitHub tool MCP serverexposes tools to any LLM P07 · P16 · P18 · P19 — same shape, different roles & tool set

Phase 6 — DeepEval + PromptFoo in CI

Treat prompts like code. Eval on every PR, fail if quality drops.

PR opened GitHub Action DeepEvalpytest run PromptFooA/B prompts Threshold checkfaith / hallu / bias ✓ / ✗PR gate P20 · P21 · P23 — same gate, more metrics each project 22-metric capstone dashboard lives in P23

The 3 secrets · zoomed out

Roadmap · Topic sheet · Projects → portfolio.

Roadmap4 steps · 14 weeks Topic sheet8 modules · 120 sub-topics Projects22 repo · 5–7 live agents AI tester · 2026portfolio · interview-ready where to go · what to learn · proof you can ship

22 live projects in the GitHub repo

Every project is buildable end-to-end. Numbers match the folders in github.com/PramodDutta/AITesterBlueprint. ~14-week cohort plan; self-paced too.

Phase 1 — Local AI & prompt foundations

PROJECT 00

LLM Basics

Ground-floor: what an LLM is, tokens, sampling, where Ollama fits.

OllamaLocal LLM
Open folder →
PROJECT 01

Local Test Case Generator

FastAPI + vanilla JS UI on a Llama 3.2 backend. Feed user stories, get test cases. Introduces the B.L.A.S.T. agentic protocol.

FastAPILlama 3.2OllamaB.L.A.S.T.
Open folder →
PROJECT 02

Selenium → Playwright Converter

React + Monaco editor frontend, Node backend, CodeLlama brain. Migrates legacy Selenium Java to modern Playwright TypeScript.

ReactMonacoCodeLlamaTailwind
Open folder →
PROJECT 03

RICE-POT Selenium framework

End-to-end Java Selenium framework whose every test is generated via the RICE-POT prompt template (Role · Instructions · Context · Example · Parameters · Output · Tone).

JavaSeleniumTestNGMaven
Open folder →
PROJECT 04

Local-LLM prompt templates

Production-ready Markdown prompt templates that turn a PRD into context-constrained test cases. Wired into Playwright TS specs.

Playwright TSMarkdownTemplates
Open folder →

Phase 2 — App building & career tools

PROJECT 05

Job Board Assistant (Claude Code)

A full Kanban-style job tracker built entirely by Claude Code. React 19 + TS + Vite + Tailwind 4. Shows what an AI coding assistant can ship.

React 19TSViteClaude Code
Open folder →
PROJECT 06

Resume Fix + LinkedIn

AI prompts + DOCX/PDF templates that rewrite a QA resume for ATS + LinkedIn. Includes follow-up email + cover-letter sequences.

PromptsDOCXATSLinkedIn
Open folder →
PROJECT 10

BugSnap — bug-report enhancer

Lightweight tool to grab a snapshot, annotate it, and emit a tidy bug-report Markdown ready for Jira.

MarkdownAnnotations
Open folder →

Phase 3 — Visual agents & workflows

PROJECT 08

n8n AI Workflow Automation

Visual workflows wiring Groq, Jira REST, Google Sheets/Docs into multi-tool QA pipelines.

n8nGroqJiraGoogle APIs
Open folder →
PROJECT 09

Content Creation agent (n8n)

Schedule-driven LinkedIn post generator. n8n wakes up, asks an LLM to draft, posts on a cron.

n8nCronAPIs
Open folder →

Phase 4 — Retrieval (RAG) + LangFlow

PROJECT 11

LangFlow fundamentals

Drag-and-drop visual flows for QA starter agents. Prompt templates, API request nodes, Groq integration.

LangFlowGroq
Open folder →
PROJECT 12

RAG Basics — 10 architectures

Theory + Python walkthroughs of every RAG variant: naive, advanced, modular, agentic, graph, hybrid, fine-tuned, hyDE, self-RAG, FLARE.

PythonLangChain10 RAG types
Open folder →
PROJECT 13

RAG with LangFlow

Visual RAG pipelines for test-case generation. AstraDB / Chroma vector store + Groq LLM, all drag-and-drop.

LangFlowAstraDBChroma
Open folder →
PROJECT 14

RAG VIBE coding app

Full-stack modular RAG: FastAPI + ChromaDB backend, static-HTML chat UI, domain-routed ingestion.

FastAPIChromaDBPython
Open folder →
PROJECT 15

Vector Embeddings Visualizer

Teach-friendly playground: paste text, see chunks, embeddings, similarity scores. Ollama / OpenAI / Mistral providers.

FastAPIVanilla JSEmbeddings
Open folder →

Phase 5 — Agents, MCP, multi-agent crews

PROJECT 07

TestPlan AI Agent + Jira

Full-stack AI agent that generates test plans from Jira tickets. Express + React + TS, Groq cloud or Ollama local. A.N.T. 3-layer architecture.

ExpressReactGroqOllamaJira
Open folder →
PROJECT 16

MCP basics

Build + connect to a Model Context Protocol server. FastMCP + Playwright; expose local tools to AI assistants.

PythonFastMCPPlaywright
Open folder →
PROJECT 17

Python AI foundations

Core + advanced Python tailored to building agents. CrewAI basics + CLI patterns for orchestration.

Python 3CrewAICLI
Open folder →
PROJECT 18

CrewAI multi-agent system

Multi-agent crews for QA processes — autonomous task execution, memory, specialised QA agents, Jira REST integration.

CrewAIPythonJiraMulti-agent
Open folder →
PROJECT 19

MCP Creation AI Agent

Author your own MCP server exposing proprietary QA tools, dashboards, remote services as resources + prompts the AI can read.

FastMCPPythonCustom tools
Open folder →

Phase 6 — LLM evaluation + capstone

PROJECT 20

LLM Evaluation foundations

First taste: pytest-style DeepEval tests against a local Ollama LLM. Answer Relevancy, Faithfulness, Hallucination metrics.

DeepEvalOllamapytest
Open folder →
PROJECT 21

DeepEval for SDET (lab)

Three-app lab: Groq chatbot + manual metric verifier + lesson-style verifier walking through 9 metrics. FastAPI + Jinja2 + Llama 4 + GPT-4o-mini.

FastAPIGroqOpenAI9 metrics
Open folder →
PROJECT 23

DeepEval Framework — capstone

Three subsystems + dashboard: React e-commerce chatbot, full RAG Explorer (ChromaDB + Nomic Embed), 22-metric DeepEval framework with swappable judge LLMs (OpenAI / Groq / local Ollama).

ReactFastAPIChromaDB22 metricsCapstone
Open folder →

22 projects, 6 phases, ~14 weeks. You leave with a GitHub portfolio that proves you can ship AI testing tools — not just talk about them.

— The Testing Academy

GitHub starter repo

Every project's scaffolding, n8n workflow exports, MCP server samples, prompt templates and DeepEval test suites live in one repo.

📦
github.com/PramodDutta/AITesterBlueprint — clone, run, modify. Each module has a folder; each project has a sub-folder with README + npm run start.
Open the repo →
FolderContainsRun
/01-fundamentalsLLM comparison notebooks · Ollama install scriptsNotebooks
/02-promptsQA prompt library, frameworks, templates (.md)Read-only
/03-toolsCursor + Copilot + Claude Code config samplesCopy
/04-generationTest-plan, test-strategy & bug-report generatorsCLI
/05-agentsn8n exports, Langflow flows, MCP servers (TS)npm run
/06-advancedStagehand demos, ReportPortal hooks, DeepEval suitesMixed
/projectsThe 5 capstone projects — each its own READMEnpm run

3 secrets to become an AI tester in 2026

Everything in this programme distils into three deliverables. Master these and you ship.

SECRET 01

The Roadmap

A clear 4-step path: Prompt Engineering → Generative AI → AI Agents + MCP → Advanced Tools & Integration. No detours, no hype. Open the roadmap →

4 steps~14 weeksSelf-paced
SECRET 02

The Topic Sheet

8 modules — AIATB_01 → AIATB_08 — covering AI fundamentals, prompts, tool setup, AI test design, n8n + LangFlow + MCP, advanced integrations, RAG + CrewAI + DeepEval + PromptFoo, and Vibe-coding + Copilot. Open the curriculum →

8 modules~120 sub-topicsHands-on labs
SECRET 03

The Projects

Build & ship — not just read. Every student leaves with a GitHub portfolio.

  • 5–7+ live AI agents across QA workflows
  • 2–3 AI testing tools you build from scratch
  • n8n + LangFlow flows wired with RAG
  • Your own MCPs exposing local tools to LLMs
  • LLM Eval with DeepEval + PromptFoo in CI
Open the project list →
🚀
Bottom line for 2026. Roadmap shows you where to go. Topic sheet shows what to learn. Projects prove you can ship. Anyone who has all three becomes the AI tester their team can't replace.