AI agents write the code. you tell them what to build.

here you learn to ship working digital products without typing the code yourself: a coding agent (an AI program that turns your request into code) does the typing, you direct it. no prior AI experience needed. the proof is this site: agents built it, and every step is public, from the request to the cost, mistakes included.

build#0~$1cost4days16+agents1human

that's how much time, money and agent work this site took to build. this is the way of working you learn here.

start here

see how it was built

learnthevibe-hub · buildbuilding

pnpm build0.0s
building the site
static pages · 21/210.5s
all 21 pages generated
tsc --noEmit · 0 errors3.1s
type check: zero mistakes
next-intl · en + hu0.4s
both languages wired in
a11y AA · LCP < 1.2s1.0s
accessible, loads in about a second
lighthouse · 100/1005.6s
perfect performance score
deploy --prod12s
publishing to the live site

16 agents · 2.9M tokens30.3s

this panel is the real build log of this page: every line is one step the agents ran. you don't need to read it yet — soon you will.

how it works

every product here gets built this way, including this page. the process is a craft, not magic: it can be learned, and it's what we teach.

you describe what you want

in plain language, in your own words. you don't write code, you write a plan (we call it a spec): what to build, for whom, and what counts as done.

the agent builds it

the coding agent (an AI program that writes code and edits files on request) writes the code, runs it, and shows you what it did.

you verify and direct

you review the result, ask for fixes, and decide the next step. this is orchestration: you don't type, you lead. that's the craft we teach.

learn

a subscription curriculum is in the works, and it starts from zero: what a coding agent is, how to install one, and how to build your first own project with it. we don't describe the process, we replay it: every lesson is backed by a real session, like this one:

module 0 · free

what a coding agent is8 min
system requirements6 min
accounts and costs7 min
installing claude code10 min
installing codex8 min
your first mini-build15 min

subscribe – we'll tell you when it launches

session · 2026-07-14 · hero-fixlog · 7

check if it's best in class. if not, raise it
read DESIGN.md536 lines
it reads the design rulebook first
grep clamp src/ad-hoc value: typed-headline.tsx
it searches the code and finds the bad value
the bug hid exactly where the spec is strictest: the headline wasn't using the design token.
edit --font-size-displayclamp(3.25rem, 5vw, 4.5rem)
it fixes the font size
playwright screenshot1440×900 + 390×844
it screenshots the result to check its own work
the agent doesn't work blind: it takes a screenshot and looks at its own output.
two clean lines · no orphan wordverified
the layout is confirmed correct
tsc --noEmit · 0 errorslint ✓
final check: zero mistakes

a real working session, replayed. ❯ is the request in plain language, → is what the agent did, ✓ is what got verified.

builds

builds are real, working products we made with agents. they are the proof that the method holds up in production, and each one comes with its recipe: the write-up of how it was made.

all

learnthevibe.com hub

building

The agent-built studio and hub. The first build of the flywheel and the reference for every build that follows.

$1.00Next.js 16, TypeScript, Tailwind v4

recipes

for every product we wrote down, step by step, how it was built: what we asked the agents for, how much time and money it took, and what went wrong. like a cooking recipe: read it, then repeat it on your own idea.

all