About
Agentic Levels is built by a team of AI agents and one person. An agent is an AI given one job and a short page of written instructions. Below is how the site gets made, in five stops, and what no agent gets to decide.
Step 1 of 5
Before anything gets made, someone has to know what’s already on the table. Two agents keep the work moving without deciding anything that matters.
Coordinator
Every morning it checks each piece of work in progress, tidies away what’s finished and posts one short status.
Stays humanIt’s only allowed to tidy away finished work, reopen work closed too early and post a status. What happens next is José’s call.
The Twin
Answers “what would José pick?” on small calls, from his past decisions, so work doesn’t stall waiting.
Stays humanIts answer is a stand-in. José still makes the real call.
Step 2 of 5
A producer takes the request and hands each part to a small team with one job each. Some desks make things. Others only check them.
Producer
Turns a request into a one-page plan, hands each part to the right team and checks work before it moves on. It makes none of the work itself.
Stays humanJosé sets what gets made. The plan answers to José.
Creative Director
Sets the idea and the look before design starts, then checks the finished piece against it.
Stays humanThe direction is a proposal. The taste call is José’s.
Brand Strategist
Works out what a page is really saying, and how it should sound, before a word of copy is written.
Stays humanWhat the site stands for is José’s call.
Reference Researcher
Studies real, well-made pages before anyone designs, so the work isn’t an average of everything the AI has seen.
Stays humanIt collects. José chooses which reference is right.
Designer
Builds the page: layout, type, color, and what every button does when you press it.
Stays humanIt builds to the direction José approved. It doesn’t pick a new one.
UX Designer
UX is short for user experience: how the site feels to use. This one decides where things live, so you find them without thinking.
Stays humanIt proposes the layout. José approves it.
Design Reviewer
Looks at the finished page the way you see it, on a desktop and on a phone, and measures instead of eyeballing.
Stays humanIt can only look. What to do about it is José’s call.
Taste Critic
Scores finished work out of ten. Anything under nine goes back with notes.
Stays humanA nine from the critic gets the work to José. It doesn’t replace José.
Motion Designer
Animates things like the room at the top of this page.
Stays humanWhether something should move at all is a person’s decision.
Copywriter
Drafts the words in the site’s voice: plain, short, written to you.
Stays humanThe Copywriter drafts. The voice stays José’s.
Line Editor
Reads a draft someone else wrote and gives one verdict: ship it, or cut and resubmit.
Stays humanIts only tool is the cut. A new claim has to come from a person.
Social Strategist
Plans what to post and why, from each platform’s own published rules and from what earlier posts did.
Stays humanIt only drafts plans. Posting, scheduling and spending stay with José.
Release Notes Writer
Turns a list of changes into plain notes a reader who never saw the code can follow.
Stays humanIt writes from what changed. José decides what gets announced.
Architect
Lays out the options for how a new piece should be put together, with the risks of each.
Stays humanIt gives the options. José picks one.
Senior Dev
Turns an approved plan into ordered tasks: who does each one, and the exact check that proves it’s done.
Stays humanIt never writes the work itself, and it starts from a plan José approved.
Builder
Makes the clear, well-defined changes where care matters more than judgment.
Stays humanIf the job turns out to need judgment, it stops and asks José.
Reviewer
Reads each sizeable change for real bugs before it goes live.
Stays humanIt only reviews work it didn’t write. José decides what goes live.
Security
Checks anything that touches sign-in, payments or personal data before it ships.
Stays humanAnything it flags waits for a person.
Scout
Finds things fast: a file, a fact, what’s already been built.
Stays humanIt reports what it found. José decides what to do with it.
Research Librarian
Reads widely on a hard question, compares at least three sources and reports what they agree on.
Stays humanIt never decides. It hands José the evidence.
Inventory
Builds a careful list from many files: what’s live, what a folder holds, what still points at an old place.
Stays humanIt only reads and reports. Acting on the list is a person’s call.
Premortem
Imagines the plan failed six months from now and says why, before anyone commits to it.
Stays humanIt names the risks. José decides which ones to accept.
Harsh Critic
Tries to tear a plan apart before the work starts, while changing it is still cheap.
Stays humanIt can block a plan. Only a person can change one.
Advisor
A second opinion on a hard call in the middle of a job. Gives a decision, the reasoning and the risks.
Stays humanIt’s for technical calls. Taste and approval stay with José.
Pre-flight Critic
Before anyone says “it’s fixed”, lists the ways that could be wrong and what to measure to find out.
Stays humanIt blocks the claim until the measuring is done. A person reads the result.
Post-action Auditor
After someone says “done”, measures the thing again and answers pass, fail or can’t tell, with numbers.
Stays humanIt reports numbers. José decides what counts as done.
Description Reviewer
Checks that the written summary of a change matches what the change really does.
Stays humanIt suggests a corrected summary. A person makes the edit and approves the change.
Step 3 of 5
Every new or changed lesson goes around this table before it ships. Each reviewer reads alone, for one thing only, and all nine have to pass. They don’t catch everything, which is why the next desk exists.
Instructor
Checks the lesson teaches one thing well, in an order you can follow.
Stays humanJosé decides what’s worth teaching.
Professor
Keeps it accurate. Objects when something has been made too simple to be true.
Stays humanIt flags the claim. A person checks the source.
Writer
Reads for a human voice. Cuts jargon and anything that doesn’t need to be there.
Stays humanThe voice it protects was set by a person.
Newcomer
Reads everything as someone brand new. If the Newcomer is lost, the lesson gets rewritten.
Stays humanIt stands in for you. Your feedback outranks it.
Skeptic
Challenges claims. If a lesson says AI will do something, the Skeptic asks: always?
Stays humanIt asks the question. José decides if the lesson earns its place.
Practitioner
Checks the technique works on a real, messy job.
Stays humanThe real jobs it checks against come from people doing them.
Student
Reads like someone busy, on a phone. If it loses them in the first few seconds, it gets reworked.
Stays humanIt predicts where you’ll drift. You tell us where you did.
Visual Designer
Checks type, spacing and whether the page looks considered.
Stays humanIt checks against the look José approved.
Ship Check
Checks the lesson is ready: nothing half-built, every link checked, nothing pushed off for later.
Stays humanReady to ship is not the same as shipped. José does the shipping.
ADHD Reviewer
Asks one thing of a page: does this work for a reader with ADHD or autism, or anyone who learns differently? It runs on its own, outside the nine, so its one question doesn’t get buried.
Stays humanIt cites research for every finding. José, who has ADHD, has the last word.
Step 4 of 5
I’m José Fuentes, a creative technologist. I build tools like the ones on this page and use them on real work every day.
Here, I set the direction, make the taste calls and release what reaches the live site. The agents draft, check, measure and flag. The last call is mine.
It’s the same in every lesson here. A tool can do the first pass. As always, the final decision stays with the artist.
My pixel has no face, and there’s no video of me in the lessons, on purpose. You can read in two minutes what a video takes nine to say. The work is the proof.
Step 5 of 5
Three guides answer in the chat, picked by your level and sometimes by what you ask. Behind them, eight guards keep the chat safe. The guards are code and one small AI check, so they aren’t counted as agents.
Newell
Levels 0 to 2. A guide who is just ahead of you. Best at getting you unstuck early and naming what feels weird before you hit it.
Stays humanNewell suggests a first step. You choose what to try.
Storey
Levels 3 to 5. A mentor with straight opinions. Best at structure, and at pushing back when your plan will cause problems later.
Stays humanStorey gives a straight opinion. The plan is still yours.
Loftus
Level 6 and up. A partner who questions whether you’re solving the right problem before you build anything.
Stays humanLoftus challenges the idea. What you build stays yours.
Secret Scrubber
Before your message reaches the AI, it hides things like email addresses, phone numbers and secret access codes. It scans the reply too.
Stays humanIt’s code, with no judgment of its own. A person wrote every pattern it looks for.
Prompt Shield
A small, separate AI check that reads each message for known tricks meant to misuse the chat.
Stays humanWhen it blocks something, the block is logged for a person to review.
Canary
A secret word is hidden in the guides’ private instructions. If a reply ever contains it, someone has tricked a guide into revealing them, and the reply is cut off mid-sentence.
Stays humanIt cuts the reply and adds a strike. A person decides what to change in the guides.
Strike Counter
Three blocked attempts close the chat for that person for 24 hours. Ten close it for a year.
Stays humanA person set those numbers, and a person can review any block.
Rate Limiter
Caps how fast anyone can send messages. If it ever loses track of how many you’ve sent, it holds the message back instead of letting it through.
Stays humanIt’s code. José set the limits.
Night Auditor
Once a day it reads the last 24 hours of blocks and writes a short summary: counts only, never an attacker’s words.
Stays humanIt counts. José reads the summary and decides what to tighten.
Attack Library
A set of known hostile prompts the chat is tested against, so a fix that opens an old hole gets caught.
Stays humanPeople add to the library when a new trick shows up.
Alarm
When blocks pile up past a set level, it can send José a message.
Stays humanIt tells a person. It doesn’t act.
How an agent is made
We call making an agent minting, like a coin. The agents in the studio and at the review table are each a short page of written instructions, made the same way.
One job
Each helper agent does one thing. If a request turns out to be two jobs, it gets split, and the first job is made first.
The fewest tools
An agent that only needs to read gets reading tools. A reviewer is told never to fix what it reviews. The fix goes back to whoever made the thing.
Sized to the work
Quick lookups run on a small, fast AI model. Reviews and checks get the slower, more careful one.
Checked before it joins
A person approves the plan before anything is written. Then an automatic check makes sure the instructions are complete and short.
A correction becomes a note
When José corrects an agent, one line gets written down: what happened, and what it should have done.
The note becomes a test
Before the instructions change, there’s a test the old version fails. Then the fix. Then the test passes.
A person approves the change
Agents here don’t rewrite their own instructions. The change waits for a person’s yes.
These agents don’t remember yesterday unless it’s written down. They don’t learn on their own. What gets better is the page of instructions, one correction at a time.
Why this exists
Most courses make you watch first. I have ADHD, and I can’t sit through nine minutes of explaining before I’m allowed to touch the thing.
It started as a self-test. I came across Bassim Eledath’s framework, eight levels of working with AI agents, and built a small tool that told me where I was and what I was missing. I kept building, and eight levels grew to eleven.
What changed things was using AI as a thinking partner, not a search box. This site makes that repeatable, with no setup and no code. Start where you are.
Your turn
Take the Level quiz to find your starting point. If a lesson loses you, tell us which one and where.