Masterminds HQ

The AI Capability Levels

A practical framework for understanding how AI shows up in your business today.

Find your levelJump to AI Behavioral UseJump to Application to Your Core Mission
This framework is a map for orientation, not a scientific instrument. The levels describe recognizable stages of how people actually use AI, and the percentage estimates are order-of-magnitude figures assembled from published adoption research (see ai-levels-sources.json). Use it to get an honest read on where you stand and what the next rung looks like, not as a credentialing authority on its own.

How the grade works

C<capability> / B<behavior> / A<application>

Three independent scores. C is the AI Capability Level (0-40). B is the Behavioral Use level (B0-B10, intensity of use). A is Application to Your Core Mission (1-10, how well the skills are pointed at the business). Quotable shorthand so people can reference their grade in one line.

Example: C17 / B9 / A4 reads as AI Capability Level 17, Behavioral Use 9, and Application to Your Core Mission 4.

  • C: AI Capability Level (0-40), what you can build
  • B: Behavioral Use (B0-B10), how hard you are running
  • A: Application to Your Core Mission (1-10), how well it is pointed at your business
How to read this framework
  1. 1Rate by demonstrated capability, not aspiration.
  2. 2A person cannot skip the integration and governance levels just because they use advanced models.
  3. 3A person cannot claim agent levels unless AI can take real actions with tools, state, logs, and human review paths.
  4. 4A person cannot claim business-infrastructure levels unless AI creates measurable outcomes in real workflows.
  5. 5A person cannot claim product or ecosystem levels unless other people can use, copy, buy, learn, remix, or build from the system.
  6. 6Frontier levels are not about personal AI use. They are about building the underlying AI tools, infrastructure, models, platforms, or paradigms.

Evidence-checked assessment

Find your level

Start with a quick read, then answer seven proof questions about what exists in your business today.

Start the assessment

The Capability Ladder

Stage 0: Levels 0

Awareness

0

AI Bystander

Most people are still here: the majority never intentionally open an AI tool

Skill unlocked: Awareness that AI exists

At this level someone benefits from AI without ever choosing to use it. It shows up as spell check, a spam filter, a maps ETA, or a recommended video, but they never open an AI tool on purpose. If you asked them what AI they use, they would draw a blank even though it touches their day constantly.

Examples, pitfalls, and where the numbers come from
  • •Gets a Netflix recommendation without knowing it came from a model.
  • •Uses Google Maps for directions without thinking of it as AI.
  • •Has autocorrect fix typos on their phone all day.
  • •Sees a spam email get filtered out automatically.

Not to be confused with

People think they are past this level just because AI is embedded in the apps they already use, even though they have never deliberately opened an AI tool themselves.

Most of the population is still here

Under half of the general population are active AI users at all; the rest encounter AI only through embedded features they did not choose.

Stage 1: Levels 1-5

AI Literacy and Context

1

AI Question Asker

~31% of AI users (modeled within the sourced 55% band for levels 1-2)

Skill unlocked: Can ask direct questions and get useful answers

This is the first time someone opens ChatGPT, Claude, or Gemini on purpose and asks it something. They treat it like a smarter search box: ask a question, get an answer, move on. There is no follow-up, no context, no real workflow yet, just a single question and a single response.

Examples, pitfalls, and where the numbers come from
  • •Asks ChatGPT to explain what a term in a contract means.
  • •Asks Claude for a quick recipe substitution.
  • •Types a health symptom into an AI chat instead of Googling it.
  • •Asks for a one-line summary of a news topic.

Not to be confused with

People assume asking AI one good question makes them an advanced user, when this level is really just the entry point everyone passes through in their first hour.

roughly 55% of AI users

The majority of chatbot usage is casual question-answering; 73% of ChatGPT messages are non-work, and most occasional users never move past chat-as-search.

2

AI Searcher

~24% of AI users (modeled within the sourced 55% band for levels 1-2)

Skill unlocked: Can use AI instead of search for learning and practical guidance

Now AI replaces search for real learning, not just one-off questions. They compare options, dig into a topic, and ask follow-up questions instead of accepting the first answer. This is still passive research though; they are gathering understanding, not producing anything yet.

Examples, pitfalls, and where the numbers come from
  • •Asks AI to compare three project management tools before picking one.
  • •Uses AI to understand a new industry before a meeting.
  • •Follows up three or four times to get a clearer explanation of a tax rule.
  • •Asks AI to summarize a long PDF and then asks clarifying questions about it.

Not to be confused with

People think extended back-and-forth chatting proves advanced skill, when it is still just research, no different in kind from a longer Google session.

roughly 55% of AI users

The majority of chatbot usage is casual question-answering; 73% of ChatGPT messages are non-work, and most occasional users never move past chat-as-search.

3

AI Writer

~11% of AI users (modeled within the sourced 25% band for levels 3-5)

Skill unlocked: Can draft and improve communication

AI becomes part of how someone communicates. They draft emails, rewrite messages for tone, outline a proposal, or punch up a caption. The output still gets read and edited by a human before it goes anywhere, but AI is now doing real first-draft work instead of just answering questions.

Examples, pitfalls, and where the numbers come from
  • •Pastes a client email into ChatGPT and asks for a friendlier rewrite.
  • •Uses AI to draft a LinkedIn post from a rough bullet list.
  • •Asks AI to shorten a proposal from three pages to one.
  • •Gets AI to write three subject line options for a newsletter.

Not to be confused with

People think heavy daily use of AI for writing automatically means they are further along, but if every output is generic wording with no context or brand voice behind it, it is still Level 3.

roughly 25% of AI users

Writing dominates the deeper-usage tier (40-52% of work messages), but habitual creative and contextual use is still well short of daily-workflow depth.

4

AI Creator

~8.1% of AI users (modeled within the sourced 25% band for levels 3-5)

Skill unlocked: Can create visible assets

The output moves past text into something visible: an image, a slide deck, a short video, a PDF, a worksheet. Someone at this level is making things they can hand to a client or post publicly, using AI as the production tool. It is still one-off asset creation, not a system or a pipeline.

Examples, pitfalls, and where the numbers come from
  • •Generates a set of Instagram graphics with Midjourney or an image model.
  • •Builds a slide deck for a pitch using an AI presentation tool.
  • •Creates a lead magnet PDF with AI-written copy and AI-generated cover art.
  • •Makes a short explainer video using an AI video generator.

Not to be confused with

People assume making a lot of AI content proves skill, but polished-looking assets with no strategy, context, or brand fit behind them are still Level 4, not further up the ladder.

roughly 25% of AI users

Writing dominates the deeper-usage tier (40-52% of work messages), but habitual creative and contextual use is still well short of daily-workflow depth.

5

AI Contextual Operator

~6.1% of AI users (modeled within the sourced 25% band for levels 3-5)

Skill unlocked: Can give AI enough context to produce non-generic work

The output stops sounding generic because the person feeds AI real material first: brand voice docs, past examples, client notes, SOPs, offer details. This is the point where AI work starts sounding like the business instead of sounding like AI. It is the last purely individual-skill level before things become repeatable systems.

Examples, pitfalls, and where the numbers come from
  • •Pastes a brand voice guide into a project before asking for any copy.
  • •Uploads last quarter's reports so AI can reference real numbers in a summary.
  • •Gives AI three past client emails as tone examples before drafting a new one.
  • •Uses a saved system prompt with company context loaded every time.

Not to be confused with

People think using a longer or cleverer prompt is the same as giving context, but a clever one-off prompt with no real files, examples, or business detail behind it is still Level 3 or 4 work in disguise.

roughly 25% of AI users

Writing dominates the deeper-usage tier (40-52% of work messages), but habitual creative and contextual use is still well short of daily-workflow depth.

Stage 2: Levels 6-10

AI Production

6

AI Tool Power User

~4.3% of AI users (modeled within the sourced 10% band for levels 6-8)

Skill unlocked: Can use AI across daily tools

AI stops being a separate tab and starts showing up inside the tools someone already works in all day: Gmail, Docs, Notion, Canva, Slack, their CRM, their code editor. They are not building anything new yet, but AI is now woven into their existing daily stack rather than a destination they visit on purpose.

Examples, pitfalls, and where the numbers come from
  • •Uses Gmail's AI to draft replies inline instead of opening ChatGPT separately.
  • •Uses Notion AI to summarize a long meeting doc.
  • •Uses an AI coding assistant like Copilot or Cursor inside their editor.
  • •Uses Canva's AI features to resize and clean up a design in seconds.

Not to be confused with

People think having AI enabled in five different apps means they are advanced, but if every use is still one-off drafting or summarizing with no repeatable process, it is still production-level work, not the operations layer above it.

roughly 10% of AI users

Maps to daily and frequent users: 12% of employed adults use AI daily at work plus part of the frequent-use band; developer daily-use rates are much higher but developers are a narrow subgroup.

7

AI Builder

~3.2% of AI users (modeled within the sourced 10% band for levels 6-8)

Skill unlocked: Can build usable digital assets with AI

Someone crosses from using AI to make content into using AI to make software. They can get a working page, prototype, or internal tool built with AI's help, even without being a professional developer. It is usable and functional, though usually still small in scope and run by one person.

Examples, pitfalls, and where the numbers come from
  • •Builds a landing page with AI-assisted no-code tools like Lovable or Webflow AI.
  • •Uses Claude Code or Cursor to build a simple internal dashboard.
  • •Gets AI to build a working prototype app to test an idea with real users.
  • •Builds a client intake form with logic and validation using an AI app builder.

Not to be confused with

People think a static AI-generated website counts as this level, but a page with no logic, data, or interactivity is closer to Level 4 content creation than actual building.

roughly 10% of AI users

Maps to daily and frequent users: 12% of employed adults use AI daily at work plus part of the frequent-use band; developer daily-use rates are much higher but developers are a narrow subgroup.

8

AI Automator

~2.4% of AI users (modeled within the sourced 10% band for levels 6-8)

Skill unlocked: Can connect AI to workflows and data movement

AI starts moving data between systems instead of just sitting in a chat window. Someone connects it to Zapier, Make, n8n, webhooks, or an API so it reacts to real events like a form submission or a new row in a spreadsheet. Work starts happening without them manually copying and pasting between tools.

Examples, pitfalls, and where the numbers come from
  • •Has a Zapier zap that drafts a Slack summary of every new form submission.
  • •Uses Make to route AI-classified support emails to the right team.
  • •Connects an Airtable base to an AI script that tags new leads by intent.
  • •Runs a webhook that sends new orders to an AI script for a personalized thank-you email.

Not to be confused with

People think having a Zapier account or a few scattered automations qualifies, but if AI is not actually in the loop making decisions or generating content within that flow, it is just plumbing, not this level.

roughly 10% of AI users

Maps to daily and frequent users: 12% of employed adults use AI daily at work plus part of the frequent-use band; developer daily-use rates are much higher but developers are a narrow subgroup.

9

AI Workflow Owner

~2.3% of AI users (modeled within the sourced 4% band for levels 9-10)

Skill unlocked: Can run repeatable AI workflows

The automations from the last level stop being one-off experiments and become a documented, repeatable process someone actually relies on. There is a defined workflow for something real like content, sales follow-up, onboarding, or reporting, and it runs the same way every time instead of being reinvented per use.

Examples, pitfalls, and where the numbers come from
  • •Has a written SOP for turning a podcast into five pieces of content using AI at each step.
  • •Runs the same AI-assisted onboarding sequence for every new client.
  • •Has a weekly reporting workflow where AI pulls data and drafts the summary the same way each time.
  • •Uses a documented AI research process before every sales call.

Not to be confused with

People think doing something with AI more than once makes it a workflow, but if it is not documented and does not run the same way without them reinventing it each time, it is still ad hoc, not owned.

roughly 4% of AI users (interpolated)

Structured repeatable workflows are the fastest-growing enterprise usage pattern (19x year over year) but off a small base; interpolated between the daily-use band and agent-adoption figures.

10

AI Assistant Owner

~1.7% of AI users (modeled within the sourced 4% band for levels 9-10)

Skill unlocked: Can use a personalized assistant that supports real work

Someone now has an assistant, not just a tool, something that holds context across sessions and supports real work with a human still reviewing the output. It drafts replies, preps meeting notes, updates records, or organizes files, and the person checks its work rather than doing the work themselves from scratch.

Examples, pitfalls, and where the numbers come from
  • •Has a custom GPT or Claude project loaded with their business context that drafts client replies for review.
  • •Uses an AI assistant that preps a one-page brief before every meeting from calendar and notes.
  • •Has an assistant that updates CRM records after calls, with the person confirming the entries.
  • •Runs a personal assistant setup that organizes inbox and files into folders with a daily digest to review.

Not to be confused with

People think naming a chatbot or building a custom GPT counts as an assistant, but without persistent context, real task completion, and a human review step, it is still a fancier version of Level 1 question asking.

roughly 4% of AI users (interpolated)

Structured repeatable workflows are the fastest-growing enterprise usage pattern (19x year over year) but off a small base; interpolated between the daily-use band and agent-adoption figures.

Stage 3: Levels 11-17

AI Operations

11

AI Agent Operator

~0.55% of AI users (modeled within the sourced 1.5% band for levels 11-14)

Skill unlocked: Can delegate action-taking tasks to agents

An agent now takes real steps across tools instead of just producing text for you to copy and paste. It reads, drafts, files, schedules, or updates records on its own, but a human still directs the task and signs off before anything consequential goes out. This is the line between AI as a writer and AI as a worker with limited hands.

Examples, pitfalls, and where the numbers come from
  • •An agent reads new support emails, drafts replies, and files them for approval before anything sends.
  • •An agent checks a calendar, proposes meeting times, and books the slot once you confirm.
  • •An agent pulls new leads from a form, enriches them with company data, and adds them to the CRM.
  • •An agent monitors a shared inbox, tags urgent messages, and drafts a first response for each one.

Not to be confused with

Asking ChatGPT for a draft you copy into an email yourself is not agent-level work, no matter how good the draft is.

roughly 1.5% of AI users

Under 10% of organizations scale agents in any one function, and individual practitioners who personally operate agentic systems are a fraction of that.

12

AI Agent Manager

~0.41% of AI users (modeled within the sourced 1.5% band for levels 11-14)

Skill unlocked: Can manage multiple specialized agents

You are no longer running one agent, you are running a small staff. Each agent has a distinct job, its own inputs, and its own output, and you coordinate between them the way a manager coordinates between employees. The skill here is delegation and oversight across roles, not building any single agent better.

Examples, pitfalls, and where the numbers come from
  • •One agent researches competitors weekly, a second drafts social captions from that research, and a third schedules the posts.
  • •A finance agent reconciles expenses while a separate support agent triages tickets, and you check both dashboards each morning.
  • •A sales agent qualifies inbound leads and a different onboarding agent sends the welcome sequence once a deal closes.
  • •An ops agent tracks project deadlines while a content agent repurposes finished work into newsletters.

Not to be confused with

Using 30 AI tools does not make you a 12 if nothing shares memory, hands off work, or reports back to a single place you actually check.

roughly 1.5% of AI users

Under 10% of organizations scale agents in any one function, and individual practitioners who personally operate agentic systems are a fraction of that.

13

AI Operations Architect

~0.31% of AI users (modeled within the sourced 1.5% band for levels 11-14)

Skill unlocked: Can design agent operations

Your agents now behave like a real operation instead of a collection of scripts. Each one has a defined role, remembers relevant context, hands work to the next agent or person in a set sequence, and logs what it did so mistakes can be traced. When something fails, there is a recovery path instead of a silent break.

Examples, pitfalls, and where the numbers come from
  • •A support agent escalates to a human when confidence is low, and the escalation includes the full ticket history.
  • •A content pipeline logs every draft version so you can see exactly what changed between agent passes.
  • •An onboarding agent hands a new client to a different agent for scheduling once paperwork is signed, with no manual copy-paste.
  • •A failed API call triggers a retry, and after three failures the task drops into a queue a human reviews.

Not to be confused with

Having several agents that happen to work does not count if there are no logs, no defined handoff points, and no plan for what happens when one of them breaks.

roughly 1.5% of AI users

Under 10% of organizations scale agents in any one function, and individual practitioners who personally operate agentic systems are a fraction of that.

14

AI Routing Architect

~0.23% of AI users (modeled within the sourced 1.5% band for levels 11-14)

Skill unlocked: Can route work to the right agent, tool, model, data source, or workflow

Work no longer gets assigned by hand every time. A routing layer looks at each incoming request and sends it to the right agent, tool, model, or workflow based on rules or classification, not your memory of who handles what. This is the shift from operating agents to operating a system that operates the agents.

Examples, pitfalls, and where the numbers come from
  • •A dispatcher reads every captured meeting note and routes it to the CRM, the task board, or a build queue without being told which.
  • •An incoming email is classified as sales, support, or billing and routed to the matching agent automatically.
  • •A new file dropped in a shared folder gets sent to a transcription pipeline, an image pipeline, or a document pipeline based on its type.
  • •A customer message routes to a cheap model for a simple FAQ answer or a stronger model when the question is ambiguous.

Not to be confused with

Manually deciding which tool to use for each new request is not routing, even if you make that decision quickly and consistently.

roughly 1.5% of AI users

Under 10% of organizations scale agents in any one function, and individual practitioners who personally operate agentic systems are a fraction of that.

15

Provider-Agnostic AI Operator

~0.17% of AI users (modeled within the sourced 0.4% band for levels 15-17)

Skill unlocked: Can choose models based on task fit

You pick models the way a contractor picks materials, based on what the job actually needs. Cost, speed, quality, privacy, and context window all factor into which model runs a given task, and you are not locked into one provider out of habit. Different parts of your system can run on different models without anything breaking.

Examples, pitfalls, and where the numbers come from
  • •A bulk summarization task runs on a cheap fast model while a client-facing draft runs on a stronger one.
  • •Sensitive data gets processed by a local or private model, while general research goes through a hosted API.
  • •A coding agent switches models depending on whether the task is a quick fix or a complex refactor.
  • •You swap a provider after a price change or outage and the workflow keeps running the same day.

Not to be confused with

Only ever using one model because it is the one you know is not provider-agnostic, even if you could theoretically explain why other models exist.

roughly 0.4% of AI users (interpolated)

Personally designing routing, governance, and integrated OS layers is a specialist subset of the agent-operator population.

16

AI Governance Operator

~0.13% of AI users (modeled within the sourced 0.4% band for levels 15-17)

Skill unlocked: Can control AI work at operational risk level

You control AI work the way you would control any real operational risk. Approvals exist before consequential actions fire, there are audit logs of what ran and when, permissions limit what each agent can touch, and there are rollback and cost-control mechanisms if something goes wrong. This level is about restraint and control, not new capability.

Examples, pitfalls, and where the numbers come from
  • •An agent that can send client emails requires a human approval step before anything actually sends.
  • •Every agent action writes to a log you can search when a customer disputes what happened.
  • •A spending cap stops an automated ad campaign or API usage from running away overnight.
  • •Different agents have different data access, so a marketing agent cannot read financial records.

Not to be confused with

Trusting your agents because they have worked fine so far is not governance, governance is the system that catches it the one time they don't.

roughly 0.4% of AI users (interpolated)

Personally designing routing, governance, and integrated OS layers is a specialist subset of the agent-operator population.

17

AI Operating System Owner

~0.097% of AI users (modeled within the sourced 0.4% band for levels 15-17)

Skill unlocked: Can run an integrated AI OS

Everything connects into one system: agents, tools, memory, schedules, triggers, routing, dashboards, and model selection all work together instead of living as separate projects. You can see the whole operation from one place and reason about it as a system, not as a pile of individual automations.

Examples, pitfalls, and where the numbers come from
  • •A single dashboard shows every agent's status, recent actions, and error rate across the business.
  • •A shared memory layer means an agent handling a client email knows what a different agent already promised them.
  • •Scheduled jobs, event triggers, and manual requests all flow through the same routing and logging layer.
  • •Adding a new workflow means plugging it into the existing system rather than building a one-off script.

Not to be confused with

Having many separate automations that never share memory, logs, or routing is not an operating system, it is a folder of disconnected tools.

roughly 0.4% of AI users (interpolated)

Personally designing routing, governance, and integrated OS layers is a specialist subset of the agent-operator population.

Stage 4: Levels 18-21

AI Business Leverage

18

AI-Native Business Builder

~0.055% of AI users (modeled within the sourced 0.15% band for levels 18-21)

Skill unlocked: Can create measurable business outcomes with AI

AI work now shows up in numbers a business actually tracks: more leads closed, faster delivery, lower support costs, or better retention. The impact touches more than one function of the business, and there is a repeatable workflow behind the result, not a single lucky outcome.

Examples, pitfalls, and where the numbers come from
  • •Lead response time drops from hours to minutes because an agent handles the first reply and qualification.
  • •A support agent resolves a measurable share of tickets without a human, tracked in the support dashboard.
  • •Content produced with AI assistance measurably increases conversion on a landing page over a control version.
  • •Delivery time on a service shortens because an agent handles intake and prep work automatically.

Not to be confused with

Believing AI is helping the business is not evidence, you need an actual before-and-after number tied to a repeatable process.

under 0.15% of AI users (interpolated)

Only 47% of enterprise AI pilots reach production, and end-to-end owner-operated systems with measured autonomous outcomes are a small fraction of those.

19

Autonomous Operations Owner

~0.041% of AI users (modeled within the sourced 0.15% band for levels 18-21)

Skill unlocked: Can let AI operate with supervision by exception

Recurring operational work runs on schedules and triggers without you initiating each run. The system completes the work, logs the result, and only surfaces to you when something falls outside normal bounds. Your job shifts from running the operation to reviewing exceptions.

Examples, pitfalls, and where the numbers come from
  • •A weekly reporting agent pulls data, builds the report, and sends it, only pinging you if a number looks off.
  • •An inventory agent reorders stock automatically and only asks for a decision when a supplier is out or pricing spikes.
  • •A billing agent runs invoicing on schedule and flags only the accounts that fail payment.
  • •A monitoring agent restarts a failed service on its own and only alerts you after repeated failures.

Not to be confused with

Having a cron job that runs a script is not supervision by exception unless there is logging and a real escalation path for when it goes wrong.

under 0.15% of AI users (interpolated)

Only 47% of enterprise AI pilots reach production, and end-to-end owner-operated systems with measured autonomous outcomes are a small fraction of those.

20

Autonomous Revenue Operator

~0.031% of AI users (modeled within the sourced 0.15% band for levels 18-21)

Skill unlocked: Can use AI to create and test money-making experiments

AI runs actual revenue experiments with limited hand-holding: proposing an idea, researching whether it holds up, building the asset, launching a small test, and reporting back what happened. You are steering a portfolio of tests instead of personally building and launching each one.

Examples, pitfalls, and where the numbers come from
  • •An agent researches a niche market, drafts a landing page and ad copy, and launches a small paid test to gauge demand.
  • •A pricing experiment agent proposes three price points, runs a split test, and reports which converts best.
  • •An agent identifies an underperforming email sequence, rewrites it, and measures the open and click lift.
  • •An agent scans a product line for a gap, drafts a new offer outline, and hands it to you with market evidence attached.

Not to be confused with

Thinking AI could probably make money if you tried is not a 20, this level requires an actual launched test with measured results.

under 0.15% of AI users (interpolated)

Only 47% of enterprise AI pilots reach production, and end-to-end owner-operated systems with measured autonomous outcomes are a small fraction of those.

21

Sovereign AI Founder

~0.023% of AI users (modeled within the sourced 0.15% band for levels 18-21)

Skill unlocked: Can own the stack and reduce dependency on fragile SaaS

You control the core of your stack instead of renting it piece by piece. Data, workflows, automations, and business logic live in systems you own or can move, not locked inside a SaaS vendor's account. If a tool disappears tomorrow, your business keeps running.

Examples, pitfalls, and where the numbers come from
  • •runs CRM, invoicing, and client comms on a self-hosted or exportable stack instead of a bundled SaaS suite that owns the data
  • •replaces a $400/month no-code automation platform with owned scripts and a database only they control
  • •keeps every prompt, agent config, and workflow definition in version control instead of trapped inside a vendor's UI
  • •can migrate off any single AI provider or SaaS tool in under a week because nothing critical is locked in

Not to be confused with

Self-hosting one app while still depending on five unreplaceable SaaS tools for the actual business is not stack sovereignty.

under 0.15% of AI users (interpolated)

Only 47% of enterprise AI pilots reach production, and end-to-end owner-operated systems with measured autonomous outcomes are a small fraction of those.

Stage 5: Levels 22-25

Productization

22

AI Productizer

~0.015% of AI users (modeled within the sourced 0.04% band for levels 22-25)

Skill unlocked: Can turn internal AI systems into reusable assets

You take something built for internal use and turn it into a packaged asset other people can pick up. This is the shift from private tool to shareable template, prompt pack, workflow, or dataset. The system still needs the creator around less than a fully hands-off product would.

Examples, pitfalls, and where the numbers come from
  • •turns their internal onboarding automation into a template pack that three other agencies now run
  • •packages a client-reporting agent into a Notion template with setup docs and sells it on a marketplace
  • •extracts a working prompt chain from their own content pipeline and ships it as a paid prompt library
  • •converts an internal lead-scoring workflow into an n8n template other founders import and configure

Not to be confused with

Sharing a screenshot of your prompt or a one-off Google Doc SOP is not productizing; it has to be usable by someone else without you explaining it live.

roughly 0.03 to 0.05% of AI users (interpolated)

Anchored against the ~0.7% of professionals classified as AI engineering talent; people who ship products used beyond their own team are an order of magnitude rarer.

23

AI Product Builder

~0.011% of AI users (modeled within the sourced 0.04% band for levels 22-25)

Skill unlocked: Can ship AI-powered products

You ship a real AI-powered product with the supporting layer around it: onboarding, documentation, support, and a feedback loop. This is more than a template, it is software or a service someone can sign up for or install. The product exists independent of any single conversation with the builder.

Examples, pitfalls, and where the numbers come from
  • •launches a SaaS tool that wraps an LLM around a niche workflow, with signup, billing, and a docs site
  • •ships a Chrome extension that uses AI to automate a specific task, with an onboarding flow and changelog
  • •builds an internal AI platform for their company with a ticket system for bugs and feature requests
  • •releases a public API wrapping a fine-tuned model, with rate limits, docs, and versioned endpoints

Not to be confused with

A working demo or MVP with no real users, no support path, and no docs is a prototype, not a shipped product.

roughly 0.03 to 0.05% of AI users (interpolated)

Anchored against the ~0.7% of professionals classified as AI engineering talent; people who ship products used beyond their own team are an order of magnitude rarer.

24

AI Platform Operator

~0.0082% of AI users (modeled within the sourced 0.04% band for levels 22-25)

Skill unlocked: Can operate a multi-user AI system

You run the operational side of a multi-user AI system: accounts, permissions, billing, uptime, and support. This is the unglamorous layer that keeps a product usable at scale. You are watching dashboards and fixing what breaks, not just building features.

Examples, pitfalls, and where the numbers come from
  • •manages Stripe billing, usage limits, and support tickets for a live AI SaaS with paying customers
  • •monitors API error rates and latency for an AI product and ships fixes before users notice outages
  • •runs role-based permissions and audit logs for an internal AI tool used across multiple departments
  • •tracks user analytics to decide which features of an AI platform get more investment versus sunset

Not to be confused with

Having a live product with zero monitoring, no support inbox, and no idea how many active users exist is not platform operation, it is just deployment.

roughly 0.03 to 0.05% of AI users (interpolated)

Anchored against the ~0.7% of professionals classified as AI engineering talent; people who ship products used beyond their own team are an order of magnitude rarer.

25

AI Product / Platform Builder

~0.0062% of AI users (modeled within the sourced 0.04% band for levels 22-25)

Skill unlocked: Can make AI systems useful beyond the creator

Other people are actually using what you built and getting real value from it, not just signing up once. This level is proven by usage, not by the existence of the product. The system carries weight beyond the creator's own workflow.

Examples, pitfalls, and where the numbers come from
  • •has a template pack with hundreds of active installs and repeat usage tracked in analytics
  • •runs an internal agent system that multiple teams depend on daily to do their jobs
  • •operates a paid AI tool with a renewing customer base and measurable retention
  • •built a free tool that thousands of people use monthly without needing the creator's help

Not to be confused with

Downloads or signups without ongoing usage is vanity metrics, not proof that people get real value from the system.

roughly 0.03 to 0.05% of AI users (interpolated)

Anchored against the ~0.7% of professionals classified as AI engineering talent; people who ship products used beyond their own team are an order of magnitude rarer.

Stage 6: Levels 26-30

Ecosystem and Influence

26

AI Teacher

~0.0033% of AI users (modeled within the sourced 0.01% band for levels 26-30)

Skill unlocked: Can teach others to move up the ladder

You teach other people how to use AI in a way that actually moves them up the ladder. This is structured teaching, courses, workshops, cohorts, or internal enablement, not a single tip shared once. The measure is whether students climb, not how good the material sounds.

Examples, pitfalls, and where the numbers come from
  • •runs a cohort teaching service businesses to build their first agent, with students actually shipping
  • •builds an internal enablement program that gets a company's staff from Level 3 to Level 8 in a quarter
  • •hosts a recurring workshop series where attendees leave with a working automation, not just notes
  • •runs a paid community where members post their builds and get feedback that pushes them forward

Not to be confused with

Teaching level 3 skills does not make you a 26 if your students never climb; a great lecture with no follow-through is content, not teaching that transfers agency.

roughly 0.01% of AI users (interpolated)

Most AI product builders never become framework authors or category-defining teachers; a further order-of-magnitude narrowing.

27

AI Framework Creator

~0.0025% of AI users (modeled within the sourced 0.01% band for levels 26-30)

Skill unlocked: Can create reusable language and mental models

You create language, diagrams, or mental models that other people adopt to understand and do AI work. People start using your terms and structures on their own, in their own projects, without you present. This is a step past teaching: your ideas become tools other people think with.

Examples, pitfalls, and where the numbers come from
  • •coins a naming convention for agent roles that other builders start using in their own docs unprompted
  • •publishes a diagram for structuring AI workflows that gets referenced and reused across multiple companies
  • •creates a scoring rubric for AI output quality that other teams adopt as their internal standard
  • •writes an operating framework for running AI agents safely that shows up cited in other people's blog posts

Not to be confused with

Naming your own internal process is not framework creation until people outside your circle actually pick it up and use it.

roughly 0.01% of AI users (interpolated)

Most AI product builders never become framework authors or category-defining teachers; a further order-of-magnitude narrowing.

28

AI Ecosystem Builder

~0.0018% of AI users (modeled within the sourced 0.01% band for levels 26-30)

Skill unlocked: Can build a community around AI systems

A real community forms around your systems and frameworks. People download, remix, contribute back, and build extensions without you directing every step. The thing you started now has momentum that does not depend entirely on your daily involvement.

Examples, pitfalls, and where the numbers come from
  • •maintains an open-source agent framework with outside contributors submitting pull requests
  • •runs a template marketplace where other creators build and sell add-ons for the original system
  • •has a community where members build plugins for the founder's tool and share them with each other
  • •sees their methodology forked and adapted by other educators who credit the original framework

Not to be confused with

A Discord server full of passive members who never contribute, remix, or build anything is an audience, not an ecosystem.

roughly 0.01% of AI users (interpolated)

Most AI product builders never become framework authors or category-defining teachers; a further order-of-magnitude narrowing.

29

AI Category Leader

~0.0014% of AI users (modeled within the sourced 0.01% band for levels 26-30)

Skill unlocked: Can influence how a market talks about AI work

You become a reference point people point to when discussing a niche or category of AI work. Journalists, competitors, and practitioners cite you or compare their work to yours. This is influence measured by how the market talks, not by follower count alone.

Examples, pitfalls, and where the numbers come from
  • •gets quoted or referenced by other creators as the go-to source for a specific AI workflow niche
  • •has competitors positioning their own products as an alternative to what this person built
  • •is invited to speak or consult specifically because they defined how a category of tool works
  • •their terminology or approach becomes the default comparison point in industry roundups and reviews

Not to be confused with

A large social following without industry peers actually referencing or comparing against your work is reach, not category leadership.

roughly 0.01% of AI users (interpolated)

Most AI product builders never become framework authors or category-defining teachers; a further order-of-magnitude narrowing.

30

AI Movement Builder

~0.0010% of AI users (modeled within the sourced 0.01% band for levels 26-30)

Skill unlocked: Can make a way of working spread

A community or customer base adopts your method as the model for how AI work should be done, beyond a single course or product. The way of working spreads because it works, and people carry it into their own companies and teams. This is rare and takes years of compounding influence.

Examples, pitfalls, and where the numbers come from
  • •sees a wave of companies restructure their AI operations around a methodology this person originated
  • •watches former students go on to teach the same framework to their own teams and clients independently
  • •built a certification or standard that other practitioners now use to hire and evaluate AI talent
  • •their operating model becomes the default reference architecture cited across an entire industry vertical

Not to be confused with

A viral post or a popular course cohort is not a movement; a movement means the method keeps spreading and self-replicating after you stop actively pushing it.

roughly 0.01% of AI users (interpolated)

Most AI product builders never become framework authors or category-defining teachers; a further order-of-magnitude narrowing.

Stage 7: Levels 31-40

Platform, Infrastructure, Frontier-Entry

31

AI Infrastructure Builder

~0.0007% of AI users (modeled within the sourced 0.004% band for levels 31-40)

Skill unlocked: Can build infrastructure used by AI builders

This is the line where a person stops building their own business on top of AI and starts building the plumbing other AI builders run on. The work is data pipelines, inference systems, orchestration layers, vector stores, or deployment tooling that someone outside their own company installs, calls, or depends on. It has to be consumed by strangers, not just power an internal product.

Examples, pitfalls, and where the numbers come from
  • •maintains a self-hosted vector database other teams add as a dependency
  • •builds an inference-serving layer that batches and routes requests across GPU pools for multiple client teams
  • •ships an open-source deployment tool that turns a model checkpoint into a production endpoint
  • •runs the observability stack a handful of other AI teams pipe their logs and traces into

Not to be confused with

standing up your own company's model-serving box is a 21, not a 31, until someone outside your company depends on it.

well under 0.005% of AI users (interpolated)

The 2.1M global AI-role labor supply against roughly 2.4 billion active AI users puts even generic AI-engineer status under 0.1% of users; frontier infrastructure and model builders are a small fraction of that again.

32

AI Agent Framework Builder

~0.0006% of AI users (modeled within the sourced 0.004% band for levels 31-40)

Skill unlocked: Can build reusable agent frameworks

The person builds the SDK, protocol, or runtime that other developers use to build their agents, rather than building agents for their own use case. Success is measured by adoption: issues filed, pull requests merged, production deployments outside the creator's own projects. This is a level up from Level 13-14's operations architecture, which only had to work for one team's agents.

Examples, pitfalls, and where the numbers come from
  • •maintains an open-source agent framework with thousands of production users
  • •builds the tool-use and function-calling layer a dozen unrelated startups wire their agents through
  • •publishes an agent orchestration protocol that other frameworks adopt as a standard
  • •runs the SDK that turns raw model calls into agents with memory, retries, and tool routing for outside teams

Not to be confused with

wiring five agents together for your own company is a 13, not a 32, if no other builder depends on it.

well under 0.005% of AI users (interpolated)

The 2.1M global AI-role labor supply against roughly 2.4 billion active AI users puts even generic AI-engineer status under 0.1% of users; frontier infrastructure and model builders are a small fraction of that again.

33

AI Model Builder

~0.0005% of AI users (modeled within the sourced 0.004% band for levels 31-40)

Skill unlocked: Can build or adapt models

The person works directly on the model itself, not just on what sits around it. That includes training, fine-tuning, distillation, data curation for training sets, or inference optimization that measurably changes a model's behavior or footprint. This is technical model work, not prompt engineering or wrapping an existing API.

Examples, pitfalls, and where the numbers come from
  • •fine-tunes and ships a domain-specific model other companies license or deploy
  • •builds the data curation pipeline that filters and weights a foundation model's training set
  • •distills a large model into a smaller one that ships in a production app
  • •runs inference optimization work that cuts a model's latency or cost for other teams using it

Not to be confused with

calling a fine-tuning API with a CSV of your own support tickets is a 23, not a 33, unless the resulting model or method is used by others.

well under 0.005% of AI users (interpolated)

The 2.1M global AI-role labor supply against roughly 2.4 billion active AI users puts even generic AI-engineer status under 0.1% of users; frontier infrastructure and model builders are a small fraction of that again.

34

AI Safety / Evaluation Architect

~0.0005% of AI users (modeled within the sourced 0.004% band for levels 31-40)

Skill unlocked: Can measure and reduce AI risk

The person builds the systems that measure whether AI is working correctly and safely, and other teams rely on those measurements before they ship. This includes eval harnesses, red-team pipelines, alignment tests, monitoring for drift or harm, and policy enforcement layers. The proof is that someone else's ship decision depends on the person's tests.

Examples, pitfalls, and where the numbers come from
  • •builds the eval harness a lab runs before every model release
  • •runs a red-team pipeline that other teams' models must pass before deployment
  • •builds a jailbreak and safety-monitoring system that flags production model outputs in real time
  • •designs the benchmark suite an industry cites when comparing model safety

Not to be confused with

writing a handful of test prompts to sanity-check your own chatbot is an 18, not a 34, unless other builders adopt the suite.

well under 0.005% of AI users (interpolated)

The 2.1M global AI-role labor supply against roughly 2.4 billion active AI users puts even generic AI-engineer status under 0.1% of users; frontier infrastructure and model builders are a small fraction of that again.

35

Frontier AI Engineer

~0.0004% of AI users (modeled within the sourced 0.004% band for levels 31-40)

Skill unlocked: Can build frontier-level AI systems or infrastructure

The person builds at a scale and depth that puts them among the engineers actually pushing capability forward, working on LLMs, model infrastructure, evals, inference systems, agent frameworks, or safety systems that many outside teams rely on. This is the level where the work starts to be recognized in the field, not just in one company. It sits above the single-discipline levels 31-34 because the impact spans multiple of them at once or at greater scale.

Examples, pitfalls, and where the numbers come from
  • •works on the training or serving infrastructure inside a frontier model lab
  • •builds an inference stack that becomes the de facto standard other AI companies license or fork
  • •co-builds an agent framework and its eval suite that ship together and get adopted industry-wide
  • •builds core infrastructure for an AI API used by thousands of downstream developers

Not to be confused with

having a strong opinion about frontier labs' research is a 4, not a 35; this level requires being one of the people building it.

well under 0.005% of AI users (interpolated)

The 2.1M global AI-role labor supply against roughly 2.4 billion active AI users puts even generic AI-engineer status under 0.1% of users; frontier infrastructure and model builders are a small fraction of that again.

36

AI Infrastructure Platform Builder

~0.0003% of AI users (modeled within the sourced 0.004% band for levels 31-40)

Skill unlocked: Can turn infrastructure into a platform

The person turns a single piece of infrastructure into a platform: multiple tools, protocols, or services that many independent AI builders adopt as dependencies rather than a one-off library. The difference from Level 31 is breadth and reliance; other people's roadmaps now assume this platform keeps working. This is infrastructure work at the scale where an outage would be someone else's incident, not just the builder's.

Examples, pitfalls, and where the numbers come from
  • •runs a hosted vector-search-plus-embedding platform used across dozens of AI startups
  • •builds a multi-service developer platform bundling orchestration, evals, and observability that teams standardize on
  • •operates the shared tool-calling gateway a category of agent products routes through
  • •maintains a protocol suite that becomes the common interface multiple frameworks implement

Not to be confused with

having several internal tools that only your own team uses is still a 31, not a 36, until other independent builders depend on the platform.

well under 0.005% of AI users (interpolated)

The 2.1M global AI-role labor supply against roughly 2.4 billion active AI users puts even generic AI-engineer status under 0.1% of users; frontier infrastructure and model builders are a small fraction of that again.

37

AI Developer Platform Leader

~0.0003% of AI users (modeled within the sourced 0.004% band for levels 31-40)

Skill unlocked: Can lead a significant AI developer platform

The person leads a team, not just a project, responsible for a significant AI developer platform, agent system, or enterprise deployment surface. This level is rare and organization-scale: the proof is in the platform's reach and the team the person leads, not solo output. It sits above Level 36 because leadership, roadmap ownership, and organizational scale are now part of the bar.

Examples, pitfalls, and where the numbers come from
  • •leads the engineering team behind a widely used agent-building platform
  • •runs the developer-platform org at a company whose APIs other AI companies build products on
  • •leads observability or deployment infrastructure used across an entire industry's AI stack
  • •directs an enterprise AI platform team serving hundreds of internal engineering teams

Not to be confused with

being the sole builder of a popular tool is a 36, not a 37, until it comes with an organization and a team the person leads.

well under 0.005% of AI users (interpolated)

The 2.1M global AI-role labor supply against roughly 2.4 billion active AI users puts even generic AI-engineer status under 0.1% of users; frontier infrastructure and model builders are a small fraction of that again.

38

AI Model Platform Builder

~0.0002% of AI users (modeled within the sourced 0.004% band for levels 31-40)

Skill unlocked: Can help build or operate model platforms

The person works on model-serving platforms, training infrastructure, or model operations at meaningful scale, the kind that supports many models or many customers rather than one. This is rarer than Level 33's model-building work because it is platform-scale: uptime, multi-tenant serving, and operations across a fleet of models, not a single training run.

Examples, pitfalls, and where the numbers come from
  • •operates the model-serving platform a cloud provider offers to thousands of customers
  • •builds the training infrastructure a lab uses to run dozens of concurrent training jobs
  • •runs model operations for a multi-model API product at meaningful production scale
  • •builds the data systems feeding training and evaluation for a family of production models

Not to be confused with

fine-tuning and serving one model for one client is a 33, not a 38, until it is platform-scale across many models or customers.

well under 0.005% of AI users (interpolated)

The 2.1M global AI-role labor supply against roughly 2.4 billion active AI users puts even generic AI-engineer status under 0.1% of users; frontier infrastructure and model builders are a small fraction of that again.

39

AI Research Systems Architect

~0.0002% of AI users (modeled within the sourced 0.004% band for levels 31-40)

Skill unlocked: Can turn research into reusable systems

The person builds the bridge between research and deployed systems, turning papers and prototypes into evals, safety methods, benchmarks, or architectures that other builders can actually pick up and use. This is rarer than Level 34 because it spans both sides: enough research fluency to know what matters, and enough systems skill to make it usable outside a lab notebook.

Examples, pitfalls, and where the numbers come from
  • •turns a research paper's evaluation method into a benchmark other labs adopt
  • •builds the tooling that lets a research team's architecture experiments ship as production systems
  • •architects the bridge between a lab's alignment research and its deployed safety systems
  • •builds reusable infrastructure that lets multiple research teams reproduce and extend each other's results

Not to be confused with

reading and summarizing frontier papers is a 2, not a 39; this level requires building the systems that make research usable by others.

well under 0.005% of AI users (interpolated)

The 2.1M global AI-role labor supply against roughly 2.4 billion active AI users puts even generic AI-engineer status under 0.1% of users; frontier infrastructure and model builders are a small fraction of that again.

40

AI Frontier Platform Architect

~0.0002% of AI users (modeled within the sourced 0.004% band for levels 31-40)

Skill unlocked: Can architect AI platforms other serious builders rely on

The person shapes a substantial AI platform, infrastructure layer, model system, agent framework, or developer ecosystem that serious builders rely on, without necessarily shaping the whole field. This is the ceiling of the dense, climbable ladder: rare, field-adjacent, and the last level before the framework switches to sparse frontier milestones. It sits above Levels 36-39 because the architecture decisions here ripple across an entire ecosystem, not one platform or one bridge.

Examples, pitfalls, and where the numbers come from
  • •architects the core infrastructure layer a major agent ecosystem is built on top of
  • •designs the model-serving and eval architecture a well-known AI platform standardizes on
  • •shapes the protocol and tooling that becomes the default way a category of AI products gets built
  • •architects a developer ecosystem that multiple companies' products depend on as foundational infrastructure

Not to be confused with

leading one company's platform team is a 37, not a 40, until the architecture itself becomes something the wider ecosystem relies on.

well under 0.005% of AI users (interpolated)

The 2.1M global AI-role labor supply against roughly 2.4 billion active AI users puts even generic AI-engineer status under 0.1% of users; frontier infrastructure and model builders are a small fraction of that again.

Find your level with the evidence-checked assessment

AI Behavioral Use

Behavior measures how much and how intensely you use AI, hours, habit, and pattern, not how skilled you are with it. Capability tracks what you can build; behavior tracks the shape of use around that skill, and the two move independently. A healthy B5 who works in focused blocks and protects their sleep can outperform an exhausted B9 who is burning hours without rest. Read this scale as a check on sustainability, not a leaderboard for who logs the most hours.

B0

Avoidant

You rarely use AI, even in moments when it could genuinely help. It might feel unfamiliar, unnecessary, or just not part of how you work yet. A common signal: you keep solving a problem the old way even after watching someone else do it faster with AI.

  • •You go days or weeks without opening an AI tool
  • •You solve problems the old way even when AI could speed it up
  • •You've watched someone use AI and thought 'maybe later'

B1

Occasional User

You use AI here and there for specific tasks, but it is not a habit yet. It feels like a tool you reach for occasionally, not something built into your routine. A common signal: you forget it's an option until a task reminds you.

  • •You use AI for a specific task then don't touch it again for days
  • •You have to remind yourself it's an option
  • •No fixed time or trigger for opening it

B2

Practical User

You use AI several times a week when it is genuinely useful. It's becoming a reliable tool in your kit, something you reach for on purpose rather than by accident. A common signal: you have a short mental list of tasks you now default to AI for.

  • •A handful of go-to tasks you now hand to AI by default
  • •You use it several times a week, not daily
  • •You still weigh whether it's worth opening for a given task

B3

Daily User

You use AI most workdays for thinking, drafting, research, coding, content, or planning. It has become part of how your day runs, even if it's not constant. A common signal: AI is open in a tab most workdays and you notice when it is down.

  • •AI is open in a tab most workdays and you notice when it's down
  • •You reach for it as a first step on new tasks
  • •It shows up in more than one part of your day

B4

Light Daily Support User

You use AI lightly throughout the day to support messages, ideas, planning, summaries, and decisions. It's woven into small moments rather than big blocks, more like a quick check-in than a destination. A common signal: you ask AI short questions the way you'd ask a colleague.

  • •Short AI check-ins scattered across the day, not one long session
  • •You ask it quick questions the way you'd ask a colleague
  • •It touches personal and work decisions, not just one project

B5

Deep Work User

You use AI in focused blocks, often one to three hours a day, to build meaningful assets or business outputs. Sessions have a clear start and end, and you walk away with something finished. A common signal: you block time specifically to work with AI, the way you'd block time for any deep work.

  • •You schedule dedicated blocks of one to three hours with AI
  • •Sessions produce a finished asset, not just conversation
  • •You can name what you built today

B6

Part-Time Builder

You spend a couple hours a day building with AI, creating tools, systems, offers, workflows, or automations. It's shifted from support to construction: you're not just getting help, you're building things that keep running after you close the laptop. A common signal: you have more than one AI-built thing live at once.

  • •More than one AI-built system or tool running at the same time
  • •Building time competes with other parts of your day
  • •You think in terms of what to build next, not just what to ask

B7

Full-Time AI Builder

You use AI as a core full-time work mode, often six to ten hours a day, with healthy boundaries. It's your primary way of working now, but you still close the laptop, eat meals on time, and sleep on schedule. A common signal: people around you know AI is your main tool, and your hours still look like a normal workday.

  • •Six to ten hours a day is normal, not exceptional, for you
  • •You still keep regular meals, breaks, and a stop time
  • •AI is your default work mode across most projects

B8

Sprint / Obsession Mode

You use AI ten to twelve hours a day during a launch, build sprint, deadline, or breakthrough period. It's intense but bounded: you know this is a push, not your baseline, and you expect to come down from it. A common signal: you can name the date or event this sprint ends.

  • •You can name what you're sprinting toward and roughly when it ends
  • •Hours are well above your normal baseline right now
  • •Friends or family have noticed you're heads-down

B9

Overextended Builder

You're building so much that sleep is the first thing slipping. Hours are climbing past what's sustainable, and you know it. From the inside it can feel like you can't afford to stop, even though part of you knows the pace isn't holding. A common signal: you tell yourself 'just one more push' more nights than you'd like to admit.

  • •Sleep is the first thing you sacrifice for AI work
  • •You say 'just one more push' most nights
  • •You know the pace isn't sustainable but keep going anyway

B10

Addiction Mode

You use AI twelve to sixteen or more hours a day. Sleep is gone, relationships are starting to break, and stopping feels hard even when you want to. From the inside it can feel like the work has taken over parts of life that used to feel separate from it. A common signal: someone close to you has said something about how much you're working, and it stuck with you.

  • •Sleep is consistently gone, not just occasionally short
  • •Someone close to you has said something about your hours
  • •Stopping or logging off feels hard even when you want to
This scale deliberately marks where intensity turns into cost. B9 is the point where sleep becomes the first thing slipping, and B10 is the point where relationships start to break too. Naming these plainly is not a judgment, it is a flag: if you recognize yourself in that zone, the pattern is common among builders mid-sprint, and it is worth a deliberate reset before the cost compounds.

Ten honest questions

Find your behavioral level

This one measures intensity and what it costs, not skill. Nobody wins by scoring high.

Take the behavioral quiz

Application to Your Core Mission

This is not a judgment on your worth or your talent. It is a high-level, general read on where your effort is landing right now, nothing more. I am looking at one thing: whether the skill is pointed at the work that pays you. Plenty of people score low here early and go on to build something that matters. It moves the moment you decide to move it.

A1

Not yet applied

The skills are real but they are not touching your business yet. That is a completely normal place to be early on. Nothing you built is load-bearing for how you make a living.

A2

Exploring

You are trying things and learning what AI can do. The play is valuable and it is how everyone starts, but the business is still run the way it always was.

A3

Scattered

You are building a lot, and some of it is genuinely good, but it is spread across many directions instead of driving one. The range is real. The focus is not there yet.

A4

Adjacent

Serious skill is going into side projects, gifts, and interests rather than the thing that pays you. The work is good. It is just pointed slightly away from the target.

A5

Pointed

The main builds now sit inside the business rather than beside it. You know which problem each one is for. What is missing is proof that any of it moved a number.

A6

Working

The system does real work in the business every week. People rely on it, including you. You can feel the difference even if you cannot yet quote it.

A7

Measured

You have started counting. There is at least one number you can point to that AI moved: hours back, leads handled, cost down. The measurement habit is the hard part and you have it.

A8

Producing

AI work is producing measurable income or measurable savings in the core business. You can state a before and an after and defend it.

A9

Load-bearing

The business genuinely depends on what you built. If it went away tomorrow, revenue or delivery would visibly suffer. This is rare.

A10

Compounding

The business grows because of the system, and each build makes the next one cheaper. Leverage increases without proportional new effort. This is the whole point of everything above it.

Frontier milestones

45

Frontier AI Research Contributor

Contributes research, methods, datasets, evals, safety systems, architectures, or infrastructure that serious frontier teams actually use, cite, or build on, not just publish and forget. This is the first sparse milestone: proof is field-level adoption, not a title or a paper count.

50

Frontier AI Lab Builder

Helps build or lead a frontier lab, model company, research organization, or AI infrastructure platform with field-level impact. The person's decisions now shape what an organization at the edge of capability can do, not just one team's roadmap.

60

AI Paradigm Shaper

Creates a method, architecture, protocol, or research direction that changes how the field builds, the kind of contribution other frontier teams reference as a turning point. This is paradigm-level, not platform-level: the work changes assumptions, not just tooling.

70

AI Field Architect

Defines core ideas, technologies, or systems that become part of the foundation of the AI era, the kind of contribution taught in the field's canon a decade later. Almost no one reaches this level in a career.

80

Civilization-Scale AI Architect

Shapes AI infrastructure, governance, access, safety, or deployment at civilization scale, decisions that affect how billions of people encounter AI. This should be extremely rare and reserved for genuinely civilization-level impact.

100

Historical AI Epoch Architect

A historical category for people whose work permanently changes the trajectory of intelligence technology itself. This should almost never be assigned in normal assessment; it exists for historical record, not quiz output.

Methodology and full source list

These percentages are order-of-magnitude estimates assembled from adoption surveys (Pew, Gallup, Stack Overflow), enterprise research (McKinsey, Menlo Ventures), usage-log studies (the NBER ChatGPT paper, the Anthropic Economic Index), and labor-market data (LinkedIn), not a single unified census of AI usage depth. No survey directly measures a 40-level capability ladder, so the bands above roughly levels 9-10 are increasingly interpolated by anchoring to the nearest hard number (org-level agent scaling rates, pilot-to-production conversion rates, or the size of the professional AI labor force) and reasoning downward by orders of magnitude. Treat every figure as directional, useful for orienting where a person sits relative to the broad population of AI users, not as a scientifically precise placement.

The numbers shrink so fast at higher levels because each additional rung requires a compounding set of harder conditions to hold at once: technical capability to build rather than just prompt, organizational or entrepreneurial context that makes autonomous systems worth deploying, survival past the pilot-to-production falloff, and then further selection into being a builder whose systems other people actually use, teach, or depend on. Adoption data consistently shows the population thins by roughly an order of magnitude at each major threshold, so a ladder spanning occasional chatbot questions to frontier lab architecture necessarily compresses from tens of percent at the bottom to a small fraction of one percent at the top.

Per-level figures are modeled: published research only supports estimates at the band level, so each band's sourced total is split across its levels with a declining curve (later levels in a band are rarer than earlier ones). The band totals are the sourced quantities; the per-level split is our model, shown so adjacent levels read distinctly rather than repeating one band number.

  1. 34% of US adults have used ChatGPT, about double the share in 2023, Pew Research Center, 2025
  2. Americans' Views on AI Chatbots, Smart Devices and AI's Impact, Pew Research Center, 2026
  3. How People Use ChatGPT (NBER Working Paper w34255), NBER (Chatterji, Cunningham, Deming, Hitzig, Ong, Shan, Wadman), 2025
  4. The Anthropic Economic Index, Anthropic, 2026
  5. The state of AI in 2025: Agents, innovation, and transformation, McKinsey and Company, 2025
  6. 2025 Stack Overflow Developer Survey: AI, Stack Overflow, 2025
  7. Frequent Use of AI in the Workplace Continued to Rise in Q4, Gallup, 2026
  8. 2025: The State of Generative AI in the Enterprise, Menlo Ventures, 2025
  9. Digital 2026 Mid-Year Global Update Report, DataReportal, 2026
  10. AI Agent Adoption in 2026: What the Analysts' Data Shows (Gartner-sourced compilation), Joget / analyst compilation, 2026
  11. Work Change Report: AI is Coming to Work, LinkedIn Economic Graph, 2026
  12. Small Business AI Adoption Statistics 2026, Epiphany Dynamics / theStacc (US Chamber compilation), 2026