We’re working on a ladder for individual AI proficiency and would love feedback on both the levels and the definitions.
L0 New: brand new to AI, or has not yet used it.
L1 Chat: simple prompt-and-response use. Work is serial: ask, wait for an answer, then ask again.
L2 Contextual Work: gives AI relevant documents, data, or workspace context so it can work within the actual artifact and produce a more useful result.
L3 Orchestrate: coordinates multiple agents or AI roles across independent workstreams, with work that may review, challenge, compare, or build on other work. This is not just for engineering.
L4 Automate: creates workflows that are triggered by business events and run without someone sitting at a laptop directing each step.
L5 Loop: feeds the output of those workflows back into shared knowledge or a company brain, so future workflows improve over time.
A few things I’d love your perspective on:
Are these the right levels?
Are any of the names unclear or overlapping?
What observable behaviors would you use to distinguish one level from the next?
Does “loop” make sense as an individual proficiency level, or is it inherently a team or company capability?
Is there a L6 and if so how would you define it?
We’re trying to define these because, in customer conversations, we’ve found that people are not very good at self-evaluating their own AI proficiency. Frequent use often gets mistaken for proficiency. And being low on a ladder like this can feel like admitting you are falling behind, do not fit in, or are less secure in your job, especially for leaders expected to set the pace. We want a more objective, behavior-based way to distinguish the two.
- Orchestrate is situationally useful. Quite often you're better off using agents in series rather than in parallel. Most people bragging about parallel agents or orchestrated agent workflows are just bragging.
- Automate is hard to pin down. Not every process requires automation. But anything that is guarded by mechanical (deterministic) process is VERY helpful for keeping agents in line.
- "loop" is not quite that. Loop as it's currently being described in the field is about automating the process so that it remains entirely within the LLMs (which is a pipe dream, btw). They're actually saying that you don't have to write prompts anymore! That's a dead end. However, "loop" in the sense you're describing (feeding results back into the documentation) is valid.