55% OFF
L3 Twin galaxy
Super AI Assistant Training Program
After four lessons, it's already working for you.
NT$ 3,590List priceNT$ 7,980
AI Assessment
Use one conversation to confirm how you work with AI
A 10 to 15 minute AI interview. Based on how you actually work, we find which level you showed.
What is the most recent resultyou finished with AIand actually used?
- Free with sign-in
- 10 to 15 minutes
- Results saved to your account
No account yet? Sign up free
Level 1
Foundation
Finish this one
Say it clearly, check and fix it, end up with something usable
Example: the weekly reportGive AI the data, ask for the weekly report, then check and fix it yourself.
- What you do
- State the need, supply the data, check and correct
- What AI does
- Drafts and revises on your instructions
- Versus the next level
- L1 gets this one task done. L2 adds a method that still works on a new topic or new material.
Level 2
Maker
Works on new topics
Turn one success into a repeatable craft
Example: the weekly reportWith a fixed template, analysis rules and style settings, each week's fresh data still yields a report that meets the bar.
- What you do
- Set specs and style, pick methods, tune and maintain them
- What AI does
- Helps produce again and again from your methods, material and specs
- Versus the next level
- At L2 a person applies the method. At L3 AI takes over a stretch of the work.
Level 3
Twin
Hand off a task
Hand AI the goal and let it finish the whole task
Example: the weekly reportAI reads the given data, compiles the report and saves it where it belongs. You assign, approve, review and take over when something goes wrong.
- What you do
- Define the task, arrange data and tools, set permissions, review and take over
- What AI does
- Carries on through the middle steps, uses tools, delivers the result
- Versus the next level
- L3 knows how to delegate and receive the deliverable. L4 designs the handling, roles and handoffs, and can explain why.
Level 4
Workflow
Design the whole flow
Redesign how the work runs, so people, code and AI each do their part
Example: the weekly reportCode handles the fixed calculations, AI explains anomalies, a person signs off on key conclusions, and you can say why each handoff and check exists. Missing data or odd numbers have a defined path.
- What you do
- Design roles, handoffs, review and exception handling
- What AI does
- Carries out several steps by design: checks, routing and handoffs
- Versus the next level
- L4 is about how the work is designed. L5 is about caring for it once real use begins.
Level 5
Product
Keep it running
Turn a one-off piece into a system that keeps carrying the work
Example: the weekly reportThe weekly report now serves the team. When a data field, a calculation or a reader's need changes, you plan the change, confirm old metrics and past reports still hold, and can roll back if it breaks.
- What you do
- Gather needs, plan changes, review, maintain and recover
- What AI does
- Keeps serving and accepts the updates a person arranges
- Versus the next level
- At L5 people mostly plan and confirm improvements. At L6 the system finds problems against criteria, coordinates the fix and verifies it.
Level 6
Evolve
Let the system improve itself
Set the criteria and limits, then let the system improve itself
Example: the weekly reportThe system checks each weekly report against calibrated quality criteria, classifies the problem, fixes it and re-verifies. Recurring problems feed back into the method, adopted once verified. Anything beyond its authority, short on data or at a limit goes back to a person.
- What you do
- Set goals, roles, quality criteria and the limits of autonomy
- What AI does
- Finds problems against the criteria, applies fixes and re-verifies
- Versus the next level
- This is the last level.
The Whole Path
Six levels, one path
- ExploreStarter
- L1Foundation
- L2Maker
- L3Twin
- L4Workflow
- L5Product
- L6Evolve
Related Work
Anthropic | AI Fluency course (4D framework)On delegating and checking
- What it does
- Anthropic built the AI Fluency course with professors from two universities, using a 4D framework: Delegation, Description, Discernment, Diligence.
- How it maps to the six levels
- The six levels likewise treat task definition, quality judgment, delegation and the limits of responsibility as abilities to look at.
Claude Academy | Anthropic's official coursesFrom daily use to delegation
- What it does
- Covers everyday work use, building Skills, working with Claude on your own files and projects, delegating to sub-agents, and trusted long-running work without supervision.
- How it maps to the six levels
- Its topics run from everyday use to delegation and long-running work, close to the shallow-to-deep direction of the six levels.
OpenAI Academy｜Apply AI at WorkReview first, then repeat
- What it does
- Build the habit of prompting and reviewing first, then practice repeatable workflows and directing agent work.
- How it maps to the six levels
- Review habits first, then repeatable practice, much like the opening order of the six levels.
Levels of Autonomy for AI Agents (Feng et al., 2025)Seen from the human role
- What it does
- A paper published by the Knight First Amendment Institute. It sorts agent autonomy into five levels by the role the user plays when working with the agent.
- How it maps to the six levels
- It also looks at the role a person plays. The paper is about agent design; the six levels are about human ability.
Future of Work with AI Agents (Shao et al., 2025)Task-level view
- What it does
- A paper by Shao et al. Taking occupational tasks as the unit, it uses a five-level Human Agency Scale to describe how much human involvement each task calls for, based on worker preferences and expert assessment.
- How it maps to the six levels
- It also takes work tasks as the unit when looking at how people and AI divide the work.
Public courses and papers we consulted when defining the six levels and the courses. We have no partnership with, or licence from, the institutions or authors.
How It Works
Three steps, from conversation to result
-
01
Answer the questions
Start with one piece of work you recently finished with AI. Follow-up questions ask for specifics and reasons, based on your answers.
-
02
Find your level
When the conversation ends, Claude and OpenAI's GPT each rate your level independently.
-
03
See your result
See which level you showed, what it was based on, and what is missing for the next level.
What You Get
Understand your level,and know the next step
AI Ability Test ・ Result
Hand AI the goal and let it finish the whole task
Why this level
Why L3
After the conversation, Claude and OpenAI's GPT each rate you independently against the six-level criteria, checking how you work and the calls you make. When needed, SAT. KNOWLEDGE staff confirm the result.
Example｜Daily morning briefing
He has AI compile a morning briefing every day and save it to Notion. His description shows these three L3 calls:
- Asks AI to compile the briefing daily and save it to Notion.
- Decides which news sources and read/write permissions it may use.
- Steps in to check the sources and summaries himself.
What is missing for the next level
To reach L4: arranging the whole workflow
Look at the work end to end. Practise deciding who does what, how work is handed over, and what happens with exceptions.
Next step from L3
- Decide what code, AI and people each do, and explain why
- Set up handovers and checks between steps
- Plan how exceptions are handled
Suggested galaxy course
The skills you need start here
Recommended from this result, so you can build the skills you need right now, step by step.
55% OFF
Your Nebi
Let Nebi
learn and evolve with you
Finish the assessment and you get your own companion, Nebi.
- Pick your companionChoose the one you like from nine series.
- Give it a namePick a name you like so it has one of its own.
- Help Nebi evolveShow a higher level and it evolves into a new form.
-
NowL1
-
L2
-
L3
-
L4
-
L5
-
L6
Nine series. Pick one (tap to try)
Forms from L2 on appear only once you show that level.
Before You Start
The time it takes for a coffee to see your AI ability
How long it takes
About 10 to 15 minutes. A dozen or so questions, answered one at a time in writing.
Explain how you work
What you asked for, how you checked it, why you set it up that way: all of it counts.
Write freely
Under the providers' commercial terms, your answers are not used to train models.
What did you last get done with AI, and how did you do it?
I gave it interview transcripts and had it sort out needs, concerns and next steps so sales could see the key points.
Have you had AI handle new transcripts on a schedule, without asking each time?
No, I still open a chat myself each time.
Say new transcripts arrive every day. How would you set AI up to sort them regularly?
For Teams
With one shared language,
With one shared language,
see where your team stands
Match each member's AI level to a galaxy course and plan your team's training.
Course completion
Finished the set lessons and projects
Level placement
Shows the core understanding and judgment of that level in a real task or situation
Hands-on evidence
Has real experience or work, and can explain what they arranged, ran and verified
Ready to own a task independently
Also depends on the domain knowledge, hands-on experience, resources and scope of authority that job needs, decided by a manager
FAQ
FAQ
About the assessment
Who is this assessment for?
Anyone who has used AI at work, or is about to start. The questions begin with the work you actually do.
Is this an exam?
No. There is no pass or fail. The result is the level you showed this time.
Does it test tool skills or jargon?
No. It looks at your understanding and judgment in real work, not how many tools you know or whether you can define terms.
How long does it take, and how many questions?
10 to 15 minutes, a dozen or so questions, one at a time, answered in text.
What do I need before I start?
No materials needed. Just think of one piece of work you recently finished with AI: how you briefed it, how you checked the result, and, if something went wrong, how you dealt with it.
What if I'm asked about something I haven't done?
Just say so. The question will switch to a concrete scenario and ask how you would handle it.
What if I don't understand a question?
Press "Ask another way" and you'll get the question phrased differently.
Can I leave partway through?
Yes. Your answers are saved automatically and you pick up from the last question.
How soon will I get my result?
Rating starts when the conversation ends. You can leave and check back later. The few results that need confirming by SAT. KNOWLEDGE staff take a little longer.
What will the result tell me?
The result shows the level you demonstrated (or Starter), and explains the basis by referring to how you described your work. You also see what stands between you and the next level, and which galaxy course to start with.
Does it cost anything?
Your first assessment is free. For every Level you buy, you get one more assessment. You can also buy one on its own.
Can I take it again?
Yes. Each new assessment uses one of your assessment credits. To add to your last one, use "Add more details" on the report page, which costs no credit.
Do I have to take the assessment before buying a course?
No. The result is a suggestion for choosing courses. You can also pick courses directly.
About the levels
What is Starter exploration?
It means this session did not yet show the core understanding and judgment of L1 Foundation. It is not one of the six levels, and it is not a verdict on you.
If I've used many AI tools, will my level be higher?
No. Your level reflects how you delegate, check and take over. The number of tools alone doesn't count.
Can I reach a higher level without coding?
Yes. Content creators, administrators, operations leads and product builders can all show different levels of ability in their own work.
Does everyone need to reach L6 Evolve?
No. Going deep on the level your work needs is a worthwhile path.
Is the result a certificate or work authorization?
No. A level is not a certificate, a professional qualification or work authorization. Whether you can own a job independently also depends on domain knowledge, hands-on experience, resources and scope of authority.
Does finishing a galaxy give me that level?
No. Finishing a galaxy means finishing its set lessons and projects. Whether you show that level's ability is still decided by the assessment.
About your data
Where do my answers go?
Your conversation is sent to Anthropic's Claude, which asks the questions. For grading, Claude and OpenAI's GPT each rate your level independently.
Will my answers be used to train AI?
No. Under both services' commercial terms, this content is not used to train models.
Where is my conversation stored?
Conversations are stored in SAT. KNOWLEDGE's cloud database in Singapore.
Who can see my conversation?
You can. If you take part through your company, it sees your level and course suggestions, but not your answers or the conversation. When a result needs confirming, a reviewer at SAT. KNOWLEDGE reads the conversation.
My work is confidential. How should I answer?
Replace or leave out client names, amounts and unpublished data. You don't need those details to explain how you work.
Can I have my data deleted?
Yes. Email service@sat.cool and we will delete your conversation and result within 14 working days.