◆ Stages

AI proficiency stages

Based on Steve Yegge's original post. Stages 1–4 sit before our telemetry begins; 5.5 and 6.5 are AIQRank's own half-step granularity.

Ascent map

Pre-CLI Measurable Claude only

Detail

1.0

No AI

Pre-CLI · Not measurable

No AI in the loop. Maybe the occasional ChatGPT tab, but the IDE is unaugmented and code is written by hand.

2.0

Pre-CLI · Not measurable

A narrow coding agent lives in a sidebar and asks permission for every tool call. Low trust; output is reviewed before anything lands.

3.0

Pre-CLI · Not measurable

Permissions off. The agent edits files directly inside the IDE. Works well for single-function changes, less well for cross-file refactors.

4.0

Pre-CLI · Not measurable

The shift from inspecting every diff to steering the conversation. You're trusting direction over line-by-line verification, but still IDE-bound.

5.0

Measurable

You're using Claude Code or Codex with built-in tools, but not yet customizing your environment. Single-threaded sessions, no skills authored.

5.5

Measurable

You've started shaping the tool. CLAUDE.md, AGENTS.md, or one or two custom skills. Plan mode in regular use.

6.0

Measurable

Multiple agents running in parallel within a session. Skills and MCPs span beyond the defaults. Async work via Task / Cron / Monitor.

6.5

Claude Code only

Concurrent Claude Code sessions across distinct workstreams. Bespoke skills have replaced manual processes.

7.0

Claude Code only

You're building agents and skills others use. Your Claude Code config is itself a product.

8.0

Claude Code only

Operating at the edge of what's possible — pushing on Claude Code itself, contributing patterns the broader community adopts.