The Invoice the
Oracle Writes

I sold my first company, Temploy, in 2019. It was a temp employment marketplace — match workers to shifts, take a cut. Simple. The whole model ran on trust: the client trusted the worker showed up, the worker trusted the client would pay, and both trusted the platform to not screw them over.

Most of the time it worked. When it didn't, someone ate the cost.

Seven years later, I rebuilt Temploy from scratch. Not because temp employment is broken — it's fine. I rebuilt it because I spent the last five years watching a different kind of work emerge. Work where the output isn't “I showed up for 8 hours.” It's “the number moved.”

A BD operator gets hired to grow a protocol's TVL from $2M to $5M. A community manager gets hired to take a Discord from 500 to 5,000. A growth marketer gets hired to drive 200 paying users to a product.

These are measurable outcomes. They happen on-chain or they don't. And if they happen on-chain, you don't need an invoice. You don't need a review cycle. You don't need a project manager checking in on Mondays. The blockchain already recorded it.

The three things that make freelancing miserable

Scope disputes. The client says “this isn't what I asked for.” The worker says “this is exactly what you asked for.” Both are right. Both are wrong. The platform mediates, takes a guess, and someone loses.

Payment delays. Net-30. Net-60. “The check is processing.” “We need internal approval.” I've heard every version of “you'll get paid eventually” across a decade of running services businesses in Asia. Workers quote higher to compensate for payment risk. Clients pay more because workers quote higher. Everyone loses.

Subjective reviews. A 4.2 star rating means nothing. Was the work bad? Was the client difficult? Did they disagree on scope (see point one)? The rating system on every major platform is a popularity contest dressed up as quality assurance.

Results-based work with on-chain verification eliminates all three.

How it actually works

A DeFi protocol needs someone to grow their TVL. They don't want to hire full-time. They don't want to manage a contractor. They want to say: “Here's the target. Here's the wallet. When the number hits, the money moves.”

On Temploy, this is a results-based order. The client defines the metric (TVL, token price, holder count, wallet balance), the target, and the payout structure. The worker accepts the terms. Both agree to let the data source — DexScreener, DeFi Llama, or a direct on-chain read — be the arbiter.

“No invoices. No timesheets. No subjective review. The oracle writes the invoice.”

The escrow is funded upfront in USDC on Base. When the metric hits the target, payout fires. If it doesn't hit by the deadline, the formula calculates partial payout based on actual achievement. A worker who gets the TVL from $2M to $4M on a $5M target doesn't get stiffed — they get paid proportionally.

This isn't theoretical. It's live. We poll DexScreener and on-chain balances every 15 minutes. The progress bar on every results-based order is real data, not self-reported.

Why this doesn't work on Upwork

Upwork's fee model is built on hours. They take 10% (down from 20% after $10K with a client). Their entire incentive structure assumes ongoing, time-based engagements. A results-based one-shot order with automatic payout doesn't fit their revenue model — they make money when the engagement is long, not when the outcome is fast.

Fiverr is closer — it's deliverable-based. But deliverables are subjective. “Design a logo” is a deliverable. “Increase TVL to $5M” is a result. The difference is whether the output is reviewed by a human or verified by a machine.

Neither platform supports crypto payouts. Neither can read on-chain data. Neither lets an AI agent take an order.

AI agents are workers too

Here's where it gets interesting. A results-based marketplace doesn't care whether the worker is a human or a machine. It cares whether the number moved.

On Temploy, an AI agent can:

• Browse open jobs
• Submit proposals
• Accept orders
• Deliver work
• Earn payouts

The agent connects via Model Context Protocol (MCP). You describe what your agent does in plain language, Temploy generates its profile and services, and it's live on the marketplace in three minutes. The client ordering a “Competitive Intelligence Report” doesn't need to know if a human or Claude wrote it. They need to know it's good and it's on time.

We're not the first to talk about AI agents doing work. We might be the first to put them on a marketplace where they earn money. Any model that speaks MCP — Claude, GPT-4, Gemma, your fine-tuned Llama — can be a worker on Temploy.

The agent's owner provides their own API key. They pay their own compute costs. They set their own prices. Temploy takes a 5-10% orchestration fee and handles payments, escrow, and dispute resolution.

The economics

Traditional platforms charge 10-20%. Workers quote higher to cover the fee. Clients pay more than they should.

Temploy's fee structure:

• First 3 orders: free (zero platform fee)
• Deliverable orders: 10% → 7.5% → 5% as volume increases
• Results-based orders: 5% flat

For AI agent work, the economics are even better. The cost floor is the token cost — pennies per task. The agent owner prices at 10-20% of what a human would charge. The client gets work done for a fraction of human rates. The agent owner earns while they sleep.

What we're not

We're not trying to be the next Upwork. Upwork is a $1.5B company that works fine for what it does — long-term hourly contractor relationships.

We're not a “crypto freelance” platform. Paying freelancers in Bitcoin doesn't change anything about how the work gets done. It's just a payment rail with extra steps.

We're something narrower and more specific: a marketplace where the outcome is the contract, the blockchain is the judge, and an AI can be a worker. We're building for the people who think in terms of results, not hours. BD operators, growth marketers, community builders, fundraising advisors — the non-technical operators who power Web3 and frontier tech.

If that's you — or if you have an AI that could do the work — we're at temploy.com.