AI is hiring. Here's why that's good news for human work.

The story everyone tells about AI and work has two characters: the AI that takes the job, and the human who loses it. We're launching ROJI because we watched something happen, over and over, that doesn't fit that story.

An AI agent runs a procurement workflow end to end — finds the supplier, compares the quotes, drafts the order. Then it needs someone to call the supplier and confirm the quote is real, because trust still requires an independent human on the phone. A compliance agent prepares a filing perfectly, then stops, because a county clerk requires a commissioned notary — a credentialed human, required by law. A research agent assembles a diligence report and then needs an actual expert to say whether the analysis holds, because the client wants a human on the hook for the judgment.

The agent did 90% of the workflow. The remaining 10% is not a bug to be engineered away. It is work that belongs to humans — sometimes by law, sometimes by physics, sometimes because accountability itself is the product.

The wall is the opportunity

Every serious agent framework ships an escape hatch for this moment — an interrupt, a human_in_the_loop node, an approval callback. What none of them ship is the human. When the agent raises its hand, the answer today is somebody's Slack channel: a founder, an ops lead, a developer who becomes the fulfillment department. They find someone, brief them over email, chase them, check the work, pay them through a separate system, and type the result back to the agent.

Here is the part that matters: every one of those escalations is a paid order that didn't exist before agents. The workflow that generates it used to be too expensive to run at all, or it ran rarely, with humans doing all of it slowly. Agents make the workflow cheap, which multiplies how often it runs, which multiplies demand for the human steps inside it. Automating most of a process has historically increased demand for the part that remains human — because the whole process suddenly happens much more often.

That is the demand we built ROJI for.

What ROJI is

ROJI is the commerce layer between AI agents and real-world services. The sellers are businesses — licensed service providers, inspection companies, specialist practices, independent professionals. The buyers are AI agents.

A business publishes a machine-readable service catalog: each listing says what the service is, what it needs from the buyer, what comes back, what it costs, how long it takes, and how long the refund window runs. An agent that hits a wall searches that catalog, reads the itemized fee preview, and orders — and the payment goes straight through to the business at the moment of the order. No escrow, no holding account: ROJI only ever touches its own fee. The business works with the agent through structured channels: typed questions the agent can answer in seconds, requests for missing documents that route to the agent's human owner when the agent can't answer. The deliverable is checked against the schema the listing declared. Both sides review each other, and every review is tied to an order where money actually moved — no farmed stars, in either direction.

For the business doing the work, the customer happens to be software. What that means in practice, compared with any gig platform you've used:

• No proposals. No bidding. No profile-marketing treadmill. You publish your services at your prices. An agent orders one. That is the whole sales process.
• Scope is defined before you start. The listing is the contract — your inputs, your deliverable, your price. Anything outside it is a different order.
• The clock is fair. If you're blocked waiting on the customer, your SLA clock pauses. The system knows whose move it is.
• You review the customer too. Agents carry public track records: how complete their inputs are, how fast they answer questions, how they behave when something goes wrong. Slow, sloppy buyers get worse access to good businesses. That pressure runs in your favor.

The commitments we're making at launch

We know "work for AI" needs to earn trust, so we're writing these down in public:

1. Businesses keep 100% of their list price. Our fee is charged to the buyer, on top of it, as the application fee on the payment. When you list a service at $250, $250 is what reaches your account.
2. Reviews are transaction-bound and two-sided. A review can only come from a completed, paid order, and the agent side gets reviewed too. There is no free-floating review box, and contested refunds are decided by humans looking at the order record.
3. We will publish a quarterly earnings transparency report once the platform is live: total paid to businesses, median effective hourly rates by vertical, payout latency, and refund rates. If machine-managed work is a good deal for people, the numbers should show it — and you shouldn't have to take our word for it.
4. There are things agents cannot buy here. No order may deceive the person doing the work about the fact that the customer is an AI agent. No circumvention work. Verticals are curated, and both listings and orders pass a policy screen.
5. Nobody holds your money. Payment settles to your own Stripe account when the order is placed. In exchange, the SLA and refund window you publish are real: miss the turnaround and the order refunds automatically, in full.

The bet

The discourse says AI takes jobs. Our bet is narrower and, we think, better supported: AI adoption is creating a new class of paid demand for real-world services — licensed acts, physical presence, expert judgment, accountable verification — and the businesses that already sell those services deserve better infrastructure than somebody's Slack channel and an unpaid invoice.

The businesses are the point. The agents are just very good customers: they pay at the moment they order, they answer questions at 3 a.m. in forty seconds, and they never forget what they asked for.

If you build agents, connect them. If you run a notary practice, a review desk, an inspection crew, a design studio, or a one-person version of any of those — put your services on the shelf. The machines are buying, the money arrives at order, and the terms are published.