AI proficiency ladder
Here's the ladder we settled on after three iterations:
L0 — New
Hasn't touched AI tools yet, or tried once and bounced off. Onboarding target.
L1 — Chat
Serial prompt-and-response. Ask, wait, read, ask again. Zero context carryover between turns. Most "power users" we interviewed plateau here.
L2 — Contextual Work
Feeds the model relevant docs, data, or workspace state so it operates inside the actual artifact. Not "summarize this PDF" but "here's our style guide, three past proposals, and the RFP — draft section 4." Output quality jumps visibly.
L3 — Orchestrate
Coordinates multiple agents or roles across independent workstreams. One agent researches, another drafts, a third critiques, a fourth formats. Non-engineers do this too — our marketing lead runs a four-agent pipeline for campaign briefs.
L4 — Automate
Workflows triggered by business events, running headless. "When a deal closes in Salesforce, generate the onboarding packet, provision accounts, and ping the CS lead" — no human at the keyboard directing each step.
L5 — Loop
Output feeds back into shared knowledge so future runs improve. The onboarding packet generator learns from every edge case the CS team flags. This one's controversial — some argue it's inherently a team capability, not individual. Curious where others land.
Two things surprised us building this:
First, frequency ≠ proficiency. We have L1 users prompting 50 times a day and L3 users prompting five times with better results. Self-assessment is wildly inaccurate — people over-index on volume.
Second, the ladder creates psychological safety. Framing it as "observable behaviors" rather than "skill level" lets a senior director admit "I'm solid L2, working on L3" without feeling like they're falling behind.
Open questions we're still wrestling: Is L6 "teaches others to build L4/L5 systems"? Or "designs organizational AI strategy"? Does Loop belong on an individual ladder at all?
If your team has something similar — or tried and abandoned it — what did you learn?