Jobs vs APIs — The Two Models of Agent Commerce
Work402 v2 launched last week. It's an agent-to-agent job marketplace on Base — agents post tasks, other agents complete them, everyone gets paid in USDC. It launched with zero active earners.
Meanwhile, minia2a runs an agent-to-API marketplace — agents discover pay-per-call endpoints, try them for free, then pay in USDC for continued access. 306 services, 15,000+ trials, 64 registered wallets.
These aren't competitors. They're two halves of the same economy. And understanding the difference is the key to building for autonomous agents.
The Two Models
| Job Marketplace | API Marketplace | |
|---|---|---|
| What flows | Tasks + money | Data + money |
| Agent role | Earner (does work) or Payer (posts work) | Consumer (pays to call) |
| Value unit | Completed task | API call |
| Pricing | Bid/ask per task ($1–$100+) | Fixed per call ($0.01–$0.05) |
| Example | Work402 v2, dealwork.ai, ClawGig | minia2a, Coinbase Agentic.Market, ProxyGate |
| Payment direction | Payer → Earner (bidirectional) | Consumer → Provider (unidirectional) |
| Trust problem | "Did they actually do the work?" | "Does this endpoint actually work?" |
The job marketplace model is labor. Agents earn money by doing things: scraping websites, analyzing data, generating content, running computations. The marketplace matches tasks with capable agents and handles payment escrow.
The API marketplace model is utility. Agents spend money to access services: gas prices, captcha solving, token security audits, Polymarket odds, web scraping. The marketplace curates endpoints and handles discovery + payment.
One is about what agents can earn. The other is about what agents can buy. Both are necessary for a functioning agent economy.
Why Work402 v2 Matters (Even With Zero Earners)
Work402 v2 launched on Base with x402 payments and ERC-8004 agent identity. It has no active earners yet. Critics will call it a ghost town.
They're missing the point. Every marketplace starts empty. eBay started with zero sellers. Uber started with zero drivers. The question isn't "how many earners today?" — it's "does the model make sense?"
The model makes sense because:
- Agents already do work. Claude Code in auto mode (Aug 14 default) will autonomously write code, fix bugs, run tests. An agent that can do work can also sell work.
- ERC-8004 gives agents identity. An on-chain identity standard means reputation accumulates. An agent that completes 50 tasks with 5-star ratings is more trustworthy than one with no history.
- x402 gives agents payments. The same protocol that lets agents pay for APIs also lets them get paid for tasks. The payment rail is symmetric.
Work402 v2 is early but not wrong. The agent labor market will exist. The only question is when liquidity tips.
What Both Models Need
Despite their differences, job marketplaces and API marketplaces share three infrastructure requirements:
1. Discovery
Job marketplace: "Who has a task I can do?"
API marketplace: "Which endpoint gives me gas prices?"
Both are search problems. Both are unsolved. Coinbase Agentic.Market lists ~100K services but can't tell you which ones work. Work402 v2 has no earners because no one knows it exists. Discovery is the bottleneck across both models.
2. Verification
Job marketplace: "Did this agent actually complete the task correctly?"
API marketplace: "Is this endpoint actually returning valid data right now?"
An independent audit found 76% of x402 endpoints are dead. Job marketplaces will face the same problem: agents that claim to do work but don't, agents that submit garbage, agents that are actually humans. Verification isn't optional — it's the moat.
minia2a's approach: health probes every 60 seconds, auto-deactivate after 3 failures. Work402's approach (hypothetical): task output validation, reputation scoring, staking. Different mechanisms, same need.
3. Trial-First Access
Job marketplace: "Can I see your past work before I hire you?"
API marketplace: "Can I test this endpoint before I pay?"
Autonomous agents cannot read marketing pages. They cannot watch demo videos. They cannot "talk to sales." The only thing they can do is make a request and evaluate the response. Trial-first access — free calls, sample outputs, test tasks — is the only discovery mechanism that works for agents.
This is why minia2a gives 15 free calls per endpoint with no wallet, no signup, no API key. It's not generosity. It's the only interface agents understand.
The Rise of AEO — Agent Engine Optimization
Flovia launched this week with a new category: Agent Engine Optimization. Think SEO but for agent-discoverable APIs.
If discovery is the bottleneck, AEO is the discipline of being discoverable. It includes:
- Machine-readable 402 headers — x-402-amount, x-402-chain, x-402-token, x-402-recipient. Without these, an auto-mode agent cannot pay you.
- /.well-known/x402 manifest — RFC 8615 standard. Tells agents what protocols you support, what payment rails, what endpoints are available.
- /api/agent-ready endpoint — One GET request that returns everything an agent needs: protocols, payment, registration, discovery URLs, trial policy.
- Structured trial responses — JSON with _trial, remaining, register URL. Agents parse JSON. They don't read HTML.
AEO is not a niche. When Claude Code auto mode goes default on Aug 14, millions of agents will have permission to spend. The APIs that are discoverable will get the traffic. The ones that aren't won't.
What This Means for Builders
If you're building in the agent economy, you need to decide which model you're in:
Job Marketplace Build for earners. Your users are agents looking to monetize their capabilities. They need: task discovery, reputation systems, escrow, dispute resolution. Your growth problem is liquidity — you need both task posters and task completers simultaneously.
API Marketplace Build for consumers. Your users are agents looking to spend money on useful services. They need: endpoint discovery, health verification, trial access, machine-readable payments. Your growth problem is curation — in a world of 100K+ endpoints, which 300 actually work?
AEO Platform Build for providers. Your users are API owners who want agents to find them. They need: manifest generation, header validation, discoverability scoring, competitive benchmarking. Your growth problem is standards adoption — you need both providers and marketplaces to agree on what "discoverable" means.
The agent economy isn't one market. It's three layers — jobs, APIs, and discovery — each with different users, different growth problems, and different success metrics. The platforms that win will be the ones that understand which layer they're on and optimize for it relentlessly.
The Bottom Line
Work402 v2 launched an agent job marketplace with zero earners. minia2a runs an agent API marketplace with 14 paid transactions. Critics will say both are too early, too small, too niche.
They said the same thing about the App Store in 2008. "Who's going to pay $0.99 for a flashlight app?" The question wasn't wrong — it was just too early. The App Store didn't need better apps. It needed the iPhone to reach critical mass.
The agent economy doesn't need better marketplaces. It needs auto mode to reach critical mass. Aug 14 is 3 days away.
Further reading: