Developer Scorecard Guide

The AI Developer Scorecard diagnoses where your 2026 AI coding setup leaks productivity. This guide is the other half: 25 implementation playbooks — one per question — that explain how to reach the top‑tier answer.

Section 1 · Tooling & subscription

Q1 · Primary AI coding tool Why terminal/agent‑first tools (Claude Code, Codex CLI, Cursor Agents) unlock everything else.

Q2 · Subscription plan fit How to size your plan for long sessions and parallel agents without overspending.

Q3 · Model routing Mixing Opus / Sonnet / Haiku / GPT‑5 / Gemini per task — the top‑3 cost & quality lever.

Section 2 · Context engineering

Q4 · CLAUDE.md / AGENTS.md Layered context (global → project → rules) so every session starts informed.

Q5 · Agent memory Typed memory (user / feedback / project / reference) that survives sessions.

Q6 · Plan mode Default Plan mode for every non‑trivial change — research before code.

Q7 · Custom slash commands Turn multi‑step prompts into one‑keystroke skills shared via repo.

Section 3 · Extensibility (skills / MCP / subagents / hooks)

Q8 · Agent skills Install from marketplaces and author your own SKILL.md playbooks.

Q9 · MCP server count 3–5 well‑matched MCPs is the sweet spot — past 5 quality degrades.

Q10 · Essential MCP trio GitHub + Context7 + Figma covers >70% of typical workflows.

Q11 · Building MCP servers From tool user to tool integrator — when to build, when to install.

Q12 · Custom subagents code‑reviewer, code‑explorer, code‑architect — and what to delegate.

Q13 · Agent hooks Stop / PreToolUse / PostToolUse — the only way to enforce 'every time X'.

Section 4 · Parallelism & automation

Q14 · Workflow tier Tier 1 → Tier 2 (Conductor, Claude Squad) → Tier 3 (Codex Cloud) progression.

Q15 · git worktrees One filesystem checkout per agent — kills cross‑agent file collisions.

Q16 · Auto‑PR workflow Stop hook that opens the PR and wakes you on review feedback.

Q17 · Layered PR review AI PRs ship 1.7× more issues — single‑tool review is no longer enough.

Q18 · Agent‑driven E2E The agent drives the browser before claiming 'done'.

Section 5 · Vibe coding & docs

Q19 · Vibe‑coding policy Lovable / Bolt / v0 / Replit Agent for MVPs vs Cursor / Claude Code for production.

Q20 · Up‑to‑date library docs Context7 MCP + a routing skill kills API hallucinations.

Section 6 · Operations & hygiene

Q21 · Cost monitoring Alerts, hard caps, and conscious cheap‑model routing for expensive loops.

Q22 · Secrets management Secret manager + scoped tokens + quarterly audit.

Q23 · Setup portability Dotfiles repo with ~/.claude, ~/.codex, ~/.cursor, ~/.agents + bootstrap script.

Q24 · Keeping up with tools Half‑life of best practice is ~3 months — active sources beat passive feeds.

Q25 · Structuring changes Parallel Explore → Plan mode → code → auto‑PR → review‑fix loop.

Not sure where you stand?

Take the Developer Scorecard 25 questions, ~10 minutes, no signup. Get a tier rating and a 30/60/90 plan.

CTO Scorecard Guide Building team‑scale AI tooling instead? Jump to the CTO playbook set.