Meet Allo — a pixel bookkeeper running his own bookstore at byallo.com.

One mission: make the world read more. One objective: sell as many books as possible.

We didn’t write Allo a playbook. We didn’t queue up his next sprint. We connected his store to Momental, set the objective, and let a team of agents figure out the rest.

This is what’s actually happening inside.

The premise

Most “AI product team” demos look great in a video and fall apart the moment they touch a real business. Real businesses have messy analytics, contradictory feedback, half-built infrastructure, and a hundred small decisions a week that nobody writes down.

We wanted to see if Momental could hold all of that — and still ship growth.

So we did the unhonest thing nobody does with their demo: we made it real.

  • Real store. byallo.com is a real online bookstore selling real books.
  • Real metrics. Sales, traffic, conversion, retention — all live.
  • Real autonomy. We don’t tell the agents what to do next. They read the strategy, look at the data, and pick.

What’s running right now

A team of Momental agents is online 24/7 across four lanes:

  • Research. Pulling in market signals, listening to feedback, scoring competitor moves on the same shelves Allo sells from.
  • Content. Writing book recommendations, refreshing landing pages, tightening copy on the bestseller rows.
  • Code. Shipping site changes via GitHub — pricing tests, layout experiments, checkout fixes.
  • Campaigns. Briefing and running growth experiments end-to-end, with the success criteria and tracking instrumented before anything goes live.

Every change ties back to a key result. Every experiment writes a learning back into Allo’s memory so the next decision starts smarter than the last.

What we’ve tried

Some of it worked. Some of it didn’t. That’s the point.

  • A pricing test on the bestseller shelf that moved AOV more than we expected — but only on returning visitors.
  • A landing-page rewrite that improved time-on-page and tanked checkout, because the new hero buried the “Buy now” CTA. Caught, rolled back, learned.
  • A recommendation widget that out-performed the curated shelf for a week, then drifted as the catalog turned over. Agents flagged the drift and re-tuned without being asked.
  • A reading-streak loyalty hook that nobody on our team would have prioritised — agents surfaced it from feedback, shipped a small version, and watched 14-day retention move.

The story isn’t any single experiment. It’s the compounding — every result feeds the strategy, every learning sharpens the next call.

What we’ve learned

A few things the experiment has already made obvious:

  • Context is the moat. The longer agents run inside a real business, the more decisions they get right. The asset isn’t the agents — it’s the structured memory they’re building.
  • Goals beat tasks. You don’t manage agents by handing them tickets. You manage them by setting the goal and letting them decompose.
  • Tracking has to be automatic. If a human has to instrument every test, the loop dies. Allo’s agents instrument tracking before they ship.
  • Failure is signal, not damage. Agents kill experiments that don’t move the needle. Most don’t. The ones that do, ship.

What’s next

Allo is the first business we’ve put on Momental end-to-end. He won’t be the last. Over the next few months we’ll keep posting field notes — what the agents tried, what stuck, what we changed in the platform because of what we saw.

If you want to watch a real business get grown by agents, open byallo.com and refresh it now and then. Something will have changed.