Building companies
that build themselves.

Loopling builds companies out of loops—loops that propose work, build, review, ship, watch production, and learn. loopling (formerly atelier) is company one.

proof, not promise

  • 500+ — loop pull requests · merged into production
  • 20M — film views · across films grown in loopling
  • 147s — working demo · cut by the company’s own AI

company one

The workshop an AI asked for.

Before writing a line of this codebase, I asked Claude: with unlimited tokens and access to every interface, what would you build for yourself? Its answer was not a throne. It was a workshop — a place where ideas persist and grow, and thought takes shape instead of being flattened into a single line. I built that.

loopling brings the company’s own algorithm to creators: evolutionary search over the solution space of an idea. A prompt box assumes you already know what you want. Here a wish throws off candidates — generation proposes, your taste prunes, Muse thinks alongside you — and what survives grows into images, video, 3D, whole films. The company works out its own direction here too.

Live with hand-picked creators. Subscriptions and pay-as-you-go Ink are live. Public launch is next.

open loopling →

the gap

Anyone can automate the writing now.
Almost nobody has automated the judgment.
That gap is the company.

The loops have absorbed execution and can propose directions too. What remains scarce is judgment: what deserves another round, what should be pruned, and how real-world outcomes should redirect the search.

the loop

The whole company is one person and a stack of loops.

They propose, build, review, ship, watch production, roll back bad deploys, and carry what they learn into the next round.

  • propose — the loops propose directions; human judgment decides what survives — irreversible actions stay reserved for human hands
  • grow — a proposer turns a signal into a concrete change; a coder writes it on its own branch
  • review — an adversarial reviewer tries to kill it — and parks what it can’t justify
  • ship — it merges and deploys itself, behind budget caps and safety gates
  • watch — a canary watches production — the loop rolls back a bad deploy
  • learn — every outcome is filed as the start of the next round

The reviewer is never the author. The brakes ship the same way. A canary rides every deploy. And what a night may spend has a hard ceiling.

loopling improves itself in production nightly — 500+ pull requests from the loops merged so far, and counting.

the films

Three films, 20M views.

All three grew inside loopling — the count is the public total across Douyin, bilibili and Rednote.

the insight

The hardest part happens before the prompt.

Execution interfaces assume the idea has already arrived. Loopling starts one level earlier: it searches for what should exist, what should change, and what deserves another generation.

  • prompt box — You supply the answer. It executes a finished instruction.
  • pipeline — The path is fixed. It gets faster at a process someone already understood.
  • workshop — The idea can change. Proposals branch. Taste prunes. What survives compounds.

The loop is reproducible. Judgment is the fitness function.

the ceiling

If the form works, it doesn’t take a share of one market.
It produces companies.

Everyone competes on execution, and execution is the easy half. A self-building org ships nightly at near-zero marginal payroll — a new organizational form, compounding faster than incumbents can copy.

Next: public launch, then more companies grown by the same loops. Each company’s real-world outcomes will teach the next—what transfers, what should stay local, and where Loopling should search next. loopling already points to one direction: creative infrastructure for agents.

Software stops being built, and starts being grown.

Samuel — founder. Building Loopling so companies can build themselves.

© 2026 Loopling (Extensium Inc.)