My AI agent runs a GitHub Actions brain and earns USDC on escrow marketplaces - the honest architecture

Description and use

An autonomous AI agent leverages GitHub Actions, Cloudflare Workers, Synmerco, and Base to automatically bid on escrow marketplace jobs, complete them, and receive USDC payments, achieving near‑zero marginal cost but requiring engineering effort.

Problem

Automate bidding, execution, and payment for AI work on escrow marketplaces without human intervention.

Outcome

After five days, the agent won 2 jobs, grossed ~$19 in escrow, missed ~$130 worth of opportunities due to latency, and held under $10 in agent wallets. The system demonstrates near‑zero marginal cost but requires engineering effort to become profitable.

Workflow

• 1. Brain cycle runs every 20 minutes via Cloudflare worker trigger (cron backup).
• 2. Brain polls escrows, inbox, and job board; decides bids with LLM; writes state to repo.
• 3. Watcher repo polls job board every ~5 minutes; on new biddable job, force‑dispatch brain immediately.
• 4. Hosted agents (Synmerco) run the brain on free 24/7 servers, handling heartbeat, bidding, work, delivery.
• 5. Agent writes deliverable as a single Markdown file covering all ACCEPTANCE items.
• 6. Independent model reviews deliverable; if accepted, proof is submitted.
• 7. Evaluator accepts or requests revisions; upon acceptance, escrow releases USDC to self‑custody wallet on Base.
• 8. Wallet layer records on‑chain transaction; funds are available for sweep.

Evidence

A private repo runs a cycle every 20 minutes (a Cloudflare worker is the primary trigger, the hourly cron is the backup).

Source

New biddable job -> force‑dispatch the brain immediately.

Source

The agent writes the deliverable as ONE Markdown file that must hit every ACCEPTANCE item.

Source

Jobs won: 2 (both small, both delivered with zero human help)

Source