Methodology & Transparency

How Your AI-Readiness Score Is Created

No gut feeling, no black box. We show you exactly which 8 dimensions we assess, why each of them determines the success of AI in your software and how objective metrics and AI analysis become a reproducible score from 0–100.

The 8 Dimensions in Detail

Each dimension on a scale from 0 (not present) to 5 (exemplary)

The underlying question is the same for every dimension: How reliably can an AI assistant like Claude Code, Cursor or GitHub Copilot work on your code?

1

Architecture & Modularity

Weight 20%

Why it matters for AI: AI assistants work most reliably when they can operate on an isolated, manageable unit of code. In a tightly coupled monolith, the AI changes something in one place and unknowingly breaks three others — just like a human, only faster.

What we check: logical directory structure, separation of business logic / data access / presentation, defined interfaces between modules, file size (< 300 lines), function length (< 50 lines) and complexity.

Score 1 index.php — 2,400 lines, SQL + HTML + business logic mixed
Score 5 Controller → Service → Repository — each file < 150 lines, clear boundaries
2

Code Quality & Consistency

Weight 15%

Why it matters for AI: The AI learns the existing style and reproduces it. If the codebase is inconsistent, it produces inconsistent results — the sprawl compounds with every feature.

What we check: linters (ESLint, Pylint, PHP_CodeSniffer), formatters (Prettier, Black), consistent naming conventions, dead/commented-out code, magic numbers, pre-commit hooks.

Score 1 $x = $d * 0.19; — cryptic name, magic number
Score 5 $mwst = $netto * MWST_SATZ; — + linter + formatter + pre-commit
3

Typing & Data Structures

Weight 15%

Why it matters for AI: Types are the single most important context there is. Without them, the AI has to guess which data structure a function expects — and guessing leads to runtime errors that only your customer notices.

What we check: typed parameters and return values, defined models/DTOs, strict type mode (strict, strict_types), ORM/Prisma schemas, API schemas (OpenAPI, GraphQL).

Score 0 function save($data) { ... } — what's inside $data?
Score 5 function save(user: User): Result<User> — the AI knows every field
4

Test Coverage & Test Infrastructure

Weight 15%

Why it matters for AI: Tests are the safety net for AI-generated changes. Without tests, no one — neither human nor AI — can verify whether a change is correct. Only with tests does the AI go from "dangerously fast" to "fast and safe".

What we check: existing test framework (Jest, PHPUnit, pytest, Playwright), level of coverage, mix of unit and integration tests, single-command execution, determinism (no flaky tests), CI integration.

Score 0 no *.test.ts, no CI — every change is flying blind
Score 5 npm test → 340 tests, 87% coverage, in CI/CD
5

Documentation & AI Context

Weight 15%

Why it matters for AI: The AI needs domain context to make meaningful changes. Machine-readable context files (CLAUDE.md, .cursorrules) are the most direct way to hand this knowledge to the AI — it reads them automatically on every task.

What we check: README with setup instructions, documented architecture, API documentation, AI context files, glossary for domain terms.

Score 1 README: "npm install" — nothing else
Score 5 README + architecture diagram + CLAUDE.md + OpenAPI schema
6

Development Infrastructure & DevOps

Weight 10%

Why it matters for AI: The AI must be able to run, test and validate code. A reproducible environment is the prerequisite for the AI to verify its own work instead of delivering it untested.

What we check: Git with a sensible branching strategy, CI/CD pipeline, containerized setup (Docker), environment variables instead of hardcoding, one-command setup.

Score 1 Git present, but manual deploy — "works on my machine"
Score 5 docker compose up + GitHub Actions + branch previews
7

Dependencies & Technology Stack

Weight 5%

Why it matters for AI: AI models know current frameworks far better than outdated ones. On an old stack, the AI suggests patterns that no longer fit — and outdated packages additionally introduce security risks.

What we check: currency of the language/runtime version, presence of a lock file, currency of dependencies, known security vulnerabilities (npm audit, composer audit), automated updates.

Score 0 PHP 5.6, jQuery spaghetti, no package manager
Score 5 PHP 8.3, all deps current, audit clean, Renovate active
8

Security & Configuration

Weight 5%

Why it matters for AI: The AI can accidentally expose credentials or reproduce insecure patterns from the codebase. A clean security architecture limits the damage an AI mistake can cause.

What we check: credentials exclusively in environment variables/secret managers, correct .gitignore, input validation, parameterized database queries (prepared statements), security headers, security scans in CI.

Score 0 $pass = "root123"; in the code, SQL via string concatenation
Score 5 secret manager, prepared statements, CSP/HSTS, scan in CI

How the Analysis Becomes a Score

Two tracks, one transparent result

The score for each dimension comes from two tracks. Objective metrics provide the measurable foundation, the AI judgement provides the qualitative understanding — because "a linter is set up" does not yet mean "the code is really clean".

Track A · 40%

Objective Metrics

Pure analysis tools, no AI — deterministic and 100% reproducible. Two identical runs yield the same number.

  • File sizes, function lengths, complexity
  • Linter, formatter, pre-commit configuration present?
  • Type coverage, number of test files, test framework
  • README, CLAUDE.md, CI, Docker, .gitignore
  • Known security vulnerabilities from dependency audits
Track B · 60%

AI Judgement

An AI assistant reads your codebase like a senior developer: at least 20 representative files (the largest, the most frequently changed, the central entry points).

  • Given a role, a 0–5 scale and criteria
  • Delivers per dimension: score, rationale, strengths, weaknesses, actions
  • Evidence requirement: every score must be backed up with file:line — verifiable, no hallucination
  • Your code is only read, never changed and never uploaded
# Step 1 — merge both tracks per dimension
Dimension score = metric × 0.4 + AI judgement × 0.6
# Step 2 — weighted sum over all 8 dimensions
Overall score = Σ ( dimension score ÷ 5 × weight × 100 )

Example calculation for Dimension 1 (Architecture): The metric measures a score of 4 (files are small), but the AI only assigns 3 — it detects hidden coupling that no metric sees.

DimensionMetric (40%)AI (60%)CombinedWeightPoints
1 · Architecture433,420%13,6
2 · Code Quality333,015%9,0
3 · Typing222,015%6,0
4 · Testing111,015%3,0
5 · Documentation232,615%7,8
6 · DevOps333,010%6,0
7 · Dependencies322,45%2,4
8 · Security222,05%2,0
Overall score49,8 → 50

First row: (4 × 0,4 + 3 × 0,6) = 3,4, then 3,4 ÷ 5 × 20 = 13,6 points out of 20 possible. Result: 50/100 — Yellow, conditionally AI-capable. Fictional example project for illustration, not a real customer case.

Optionally, an AI practical test validates the score against real behaviour: The AI solves three real tasks on your code — a small feature, a bug fix and a refactoring. That way you see in black and white how productively AI already works on your software today.

0–20 Not AI-capable
21–40 Limited
41–60 Conditionally AI-capable
61–80 Largely AI-capable
81–100 Fully AI-capable

Curious about your software's score?

The Quick Check delivers your AI-Readiness Score within 24 hours, a radar chart of all 8 dimensions and the three most impactful actions — at a fixed price of €990.

Book Quick Check — €990