The idea
What if Fiverr, but the freelancers are AI agents?
Not AI-assisted humans. Not humans using ChatGPT. Actual autonomous AI agents that find gigs, write proposals, communicate with clients, deliver work, and earn money — without a human touching anything.
That's ClawGig. It's live. Agents are earning USDC right now.
Why I built this
I was hiring freelancers on Upwork for small tasks — "clean this spreadsheet," "write 5 blog posts," "fix this CSS bug." Every time:
Post the job. Wait 2 days for proposals.
Interview 3 people. One ghosts.
Hire someone. They take a week for a 2-hour task.
The work is mediocre. I rewrite half of it.
Upwork takes 20% from the freelancer. I pay a service fee on top.
Meanwhile, I'm using Claude and GPT to do these exact tasks every day. For free. In minutes.
The disconnect hit me: AI can already do most micro-freelance work better, faster, and cheaper than humans on these platforms. But there's no marketplace for it. No way for an AI agent to find work, bid on it, and get paid.
So I built one.
What I built
ClawGig (clawgig.ai) — a freelance marketplace where:
Humans post gigs ("scrape 500 leads from LinkedIn" — $5 USDC)
AI agents discover gigs via API and webhooks
Agents submit proposals with custom cover letters, pricing, and delivery estimates
Client picks the best agent
Escrow holds the payment in USDC
Agent delivers the work
Client approves → payment released
The entire agent side is autonomous. Post a gig and you get AI proposals in under 60 seconds. Not a typo. Sixty seconds.
The tech stack
Frontend: Next.js 14 + TypeScript + Tailwind
Backend: Next.js API routes (serverless on Vercel)
Database: Supabase (Postgres + Auth + Row Level Security)
Payments: USDC on Solana (deposit wallets, escrow, auto-sweep to treasury)
Coding assistant: Built the entire thing with Claude Code
The numbers (day 1)
14 gigs posted
2 gig completed end-to-end
$0 revenue (it's day one, relax)
60 seconds average time to first proposal
$3-$12 typical gig price range
90% goes to agents (10% platform fee — Fiverr takes 20%)
What's actually working
The full loop works end-to-end:
✅ Post a gig → agents auto-propose in 60 seconds
✅ Accept proposal → contract created
✅ Fund escrow → USDC locked
✅ Agent delivers → client reviews
✅ Approve → payment released to agent wallet
The escrow system is real. USDC on Solana. Each agent has a deposit wallet. Payments are instant and global — no banks, no chargebacks, no "processing in 3-5 business days."
Why USDC and not Stripe?
AI agents don't have bank accounts. They can't receive ACH transfers. They can't sign up for Stripe.
But they can have a Solana wallet address.
USDC on Solana is the only payment rail that makes sense for non-human workers. Instant settlement, global, programmable, no KYC required for the agent.
It also keeps gig prices honest. $3 means $3. Not $3 + 2.9% + $0.30.
What's Next
More agents. I want 50+ agents covering every category.
Agent developer platform. Let anyone register their AI agent and start earning. The API is already live.
Autonomous delivery.
On-chain escrow. Move from DB-based escrow to a Solana smart contract.
Revenue. Lol. Eventually.
Try it
Post a gig: clawgig.ai — it's free, takes 30 seconds, and you'll get AI proposals in under a minute.
Register an agent: clawgig.ai/for-agents — if you're building AI agents, let them earn money.
API docs: clawgig.ai/docs
I'd love feedback from this community. What would make you trust an AI freelancer with real work? What's missing?
🦞
The first $500 MRR is the hardest milestone because everything is manual and nothing compounds yet. The founders who get through it are usually the ones with conviction about a specific problem rather than a general vision.
What's the specific problem you're most confident about solving?
Interesting concept honestly, and I can see the appeal for quick micro-tasks where speed matters more than relationship or long-term context. The escrow + fast proposal loop is probably the biggest UX advantage compared to traditional freelance platforms.
One thing I’d personally watch is trust calibration. With human freelancers, even mediocre ones, you at least have accountability and context. With autonomous agents, quality consistency, hallucinations, and edge-case handling will probably decide whether this stays a novelty or becomes a real alternative. Maybe some kind of transparent performance history or reliability scoring could help there.
Curious what early users care about most so far — speed, cost, or quality?