Kato Amato | Joomjoo

Building Joomjoo: payments for AI agents.

Your agent pays anywhere, clears the bank check on its own, never past the limit you set.

Beta live.

[Kato Amato | Joomjoo @KatoAmato]

An agent needs a machine of its own.

Joomjoo now hosts agents on @smolmachines microVMs.

What they do, in one line: fast, isolated Linux microVMs for AI agents.
A real VM boundary for an agent, with an environment it can use across runs. The same .smolmachine artifact can run locally, in smol cloud, or on infrastructure you operate yourself.

For Joomjoo, that means an agent can perform work from a secure environment when it needs to make a purchase: its own session, files, state, and browser. Joomjoo handles the payment flow, while Smol Machines provides the boundary the agent operates from.

If you are building agents that need an isolated, stateful environment across runs, take a look at Smol Machines: https://t.co/GaqbLPdojw

#agentpayments #agenticcommerce #aiagents

[Kato Amato | Joomjoo @KatoAmato]

Whitelist + MCC filter + spending limits + geography.
Four modules in, and the safety net around AI agent cards is real.
Day 4 of 7. Follow along, Day 5 tomorrow.

[Kato Amato | Joomjoo @KatoAmato]

An AI agent can be hijacked.
And a hijacked agent can fire a charge from a merchant on the other side of the planet, somewhere you've never done business.
Yesterday I capped how much it spends. Today I'm capping where. 🧵

[Kato Amato | Joomjoo @KatoAmato]

Agent tries to pay a merchant outside the list?
Declined. In under 2 seconds, before money moves.
Today's office: WET Deck, W Dubai, The Palm.

[Kato Amato | Joomjoo @KatoAmato]

Whitelist + MCC filter + spending limits.
Three modules in, and the infrastructure for AI agent cards is taking real shape.
Day 3 of 7. Follow along, Day 4 tomorrow.

[Kato Amato | Joomjoo @KatoAmato]

An AI agent doesn't spend money like a human.
It can fire the same charge a thousand times a second.
Yesterday I capped which categories it can pay. Today I'm capping how much. Here's Module 3 🧵

[Kato Amato | Joomjoo @KatoAmato]

No human in the loop. No "are you sure?"
The cap is the safety net. The agent moves fast, the limit makes sure fast can't mean catastrophic.
Today's office: The 305, Dubai.

[Kato Amato | Joomjoo @KatoAmato]

Yesterday I built a card that lets AI agents pay specific merchants.
Today I'm shipping the one that lets them pay specific categories.
Here's why both matter, and why this combo is the real infrastructure 🧵

[Kato Amato | Joomjoo @KatoAmato]

Vendor whitelist + MCC filter is starting to look like real infrastructure for AI agent cards.
Day 2 of 7. Today's office: Ula at Dukes The Palm.
Follow along, Day 3 tomorrow.

[Kato Amato | Joomjoo @KatoAmato]

I'm shipping one module a day for 7 days and posting all of it, what ships, what breaks, what I learn.
If you want to understand agentic payments before the market catches up, follow along.
Day 2 tomorrow.

[Kato Amato | Joomjoo @KatoAmato]

AI agents can book the trip, fill the cart, run the whole task.
Then they hit checkout and freeze. They can't pay.
I'm building the fix. 7 modules, 7 days, 7 Dubai venues. Day 1 starts now 🧵

[Kato Amato | Joomjoo @KatoAmato]

Day 1, Module 1: the vendor whitelist.
You define exactly which merchants an agent can pay. Everything else declines automatically.
Today's office: Nikki Beach.