×5agents at work · each horse is one agent
03 horsesspeed ×1.00 s
Drawing your rider…
Playing as ·
Leaderboard
Lead an army of agents that self‑improve.
AI chat keeps you switching context and chasing new capabilities, and it never improves on its own. Loopr’s agents keep what they learn, so your work compounds and needs less of you every week.
A small game: a person's two hands hold the reins of a herd of horses galloping toward the horizon. Press Play, then steer with the mouse or the arrow keys: each dollar goal the front row reaches adds a horse and points; one risk ends the game. Until then an autopilot plays a demo.
How it learns
Fix it once. It remembers.
Like a new hire: in week one you correct almost everything. A month later, they just know how you like it. Each loop below is one day of work. The smaller the loop, the less you had to fix.
A line of 24 loops for one demo agent. Early runs needed about 40 fixes and draw wide, uneven loops; run 23 needed 12 fixes and is the best so far; run 24 is being written.
Demo data One example agent, 24 days. The last loop is today, still being written.
In chat, every fix becomes another rule. Loopr learns the pattern.
Maya fixes ads in Claude
Her fixes0
Rules she keeps0
Claude· New chat
Reply to Claude…
Loopr learns from her fixes
Her fixes0
Lessons it learned0
Ad drafts for this week
Learning agent
Maya wants more sales from her ads. Who does the improving?
Day 0
Maya drives it in chat
Sales from ads+0%
Cost per sale$31
Her hours0 h
sales from adscost per sale
Day 0Day 30
Claude· Ads help
Reply to Claude…
Loopr runs small bets on its own
Sales from ads+0%
Cost per sale$31
Her hours0 h
sales from adscost per sale
Day 0Day 30
Maya is away in Hawaiino messages from her
Background jobs
An example shop, made-up numbers.
One horse at a time. Or every part of the shop running at once.
Week 1 of 12Maya’s time a week 40 h