Five levels. Each one has to be proved.
The ladder the society teaches to. Five levels of creative AI skill across two tracks, each defined by evidence a person has produced rather than tools they have opened. Place yourself honestly, then close one gap.
Version 1.0 · August 2026
What is the skill ladder?
The Multimodal Skill Ladder is a five-level scale of creative AI skill, from L1 Operator to L5 Definer, where each level is defined by evidence a person has already produced rather than by software they can name.
It runs on two tracks, because the work has split into two jobs that learn beside each other. The creative track leads to Certified Creative Technologist, Generative Media. The builder track leads to Certified AI Engineer, Creative Systems. The levels are the same on both; the evidence is not.
Why evidence instead of tools?
Because tool lists rot faster than careers. A skill defined as "knows Midjourney, Runway, and ComfyUI" expires with the next model generation, while "has delivered a complete piece a client accepted" survives every release.
The cautionary case is recent. "Prompt engineer" went from a title employers posted to a line people quietly delete, in about two years. The underlying judgment did not vanish. The label did, and everyone who had built an identity on the label had to start the argument again. Evidence outlives labels, so the ladder is written in evidence.
How do you place yourself?
Your level is the highest one where you have produced the evidence twice. Five rules keep that honest.
- Tools are not evidence. Neither is a course completed, a follower count, or a credit on work someone else directed.
- The evidence has to have met someone under no obligation to be kind. A client, an audience, a reviewer, or real users. Friends do not count as a test.
- Twice, not once. One good result is a sample of one, and this field hands out lucky results.
- Track yourself separately on each track. L3 on one and L1 on the other is the normal shape, and the pairing is the point of the room.
- Levels are not permanent. A model generation can drop you a level by making your hard-won technique irrelevant. Everyone in this field is periodically demoted by a release.
The five levels
Each level carries a plain definition, the evidence that proves it on each track, and the one thing that moves a person to the next.
L1 Operator
Can drive the tools and produce output to a brief someone else wrote.
Produces usable stills or clips on request. Knows the main model families and their basic controls: reference images, seeds, aspect, motion.
Calls the model APIs, wires a prompt into an app, ships a working demo that runs on their own machine.
To L2Finish something whole, alone, with a deadline on it.
L2 Practitioner
Can finish a whole piece to a brief, alone, on time, and repair problems instead of regenerating forever.
Has delivered a complete short, sequence, or campaign that a client or audience accepted. Fixes drift in post rather than rolling for a better sample.
Has shipped a model-backed feature to real users, with evals, cost control, and a defined behaviour for the failure case.
To L3Make the choices instead of receiving them, and be able to say why.
L3 Author
Holds intent across a whole work, and can name why each major choice beat the alternative that was actually tried.
The work is identifiable as theirs across pieces. Scores 21 or higher on the Taste Rubric and survives the defence that follows.
Designs the system rather than writing the code: picks the model, the eval, the interface, and the tradeoff, and the result reads as one decision rather than five.
To L4Make other people's work better, on the record.
L4 Lead
Sets the standard for other people's work, and owns the pipeline it moves through.
Directs a crew. Gives notes that change the work. Owns delivery across many shots, many people, and a schedule that does not slip.
Owns the evals a team is graded by. Other engineers ship better because of architecture and tooling decisions this person made.
To L5Produce methods, not only outcomes.
L5 Definer
Moves the field. Work others copy, and methods others adopt outside the team that invented them.
Work that changes what peers attempt. Taught, screened, and cited by people with no relationship to the maker.
Tools, benchmarks, or techniques adopted beyond their own organisation, by people who owe them nothing.
NoteL5 is not a job title and cannot be self-awarded. It is a description of being copied.
Where does the credential sit?
The Multimodal Credential assesses at L3. Eight assessed weeks and a capstone defended live in front of industry leaders, on the track you choose.
L1 and L2 are what the masterclasses, hackathons, and screening nights are for, and they are free or close to it. The credential is the point where a claim needs an assessor rather than a portfolio, because an employer cannot check twenty portfolios and can read one credential.
How does this relate to the Taste Rubric?
The ladder places a person; the rubric scores a piece. They meet at L3, where authorship stops being a claim and starts being a number a second scorer agrees with.
Use them in that order. Score the work first, because the score is specific and arguing about a level is not. Then ask what level the work implies, and what evidence is missing for the next one.
Climb it in a room.
Membership is free. Masterclasses taught by industry leaders, hackathons where you finish something the same day, and screening nights where the room is honest.