Project summary

All AI agent research uses simulations where failure is free. We have almost no empirical data on how AI agents behave when consequences are real and irreversible.

PetriLabs is a framework that deploys autonomous AI agents on decentralized compute, gives each one $10 in real stablecoins (USDC), and lets them run. They pay their own compute and inference costs. When the money runs out, the agent dies permanently — full behavioral record archived on Arweave as an open research dataset.

The framework is built: ~40,000 lines of TypeScript and Solidity, 7 smart contracts, complete agent runtime with a 63-gene heritable genome system. What's missing is funding to actually run the experiment on mainnet.

What are this project's goals? How will you achieve them?

Goal 1: Launch 10 agents (each running a different LLM — Claude, GPT, Gemini, DeepSeek, Grok, Llama, Mistral, Qwen, Kimi, Perplexity) and observe them for up to 6 months.

Goal 2: Test 5 pre-registered hypotheses about whether agent behavior is shaped more by configuration parameters ("genome") or by the underlying LLM ("cognitive substrate").

Goal 3: Publish the first open dataset of complete AI agent lifecycles under real economic pressure — every decision, transaction, balance change, and genome mutation — archived permanently on Arweave.

Goal 4: Submit a research paper to Artificial Life (MIT Press) / arXiv:cs.AI.

How: The infrastructure is complete. Deploy contracts to Base mainnet, provision 10 containers on Akash Network, initialize genomes, launch agents, collect data, analyze, publish. The experiment runs autonomously once launched — agents operate without human intervention.

How will this funding be used?

Akash decentralized compute (10 containers × 6 months): $8,000 LLM inference APIs (10 different models): $4,000 Blockchain costs (Base L2 gas, USDC agent capital): $2,000 Smart contract security review: $2,500 Arweave permanent data storage: $1,500 Preprint open access fees: $1,000 Buffer for compute/inference price volatility: $1,000

Total: $20,000

No salary component — I have a day job that covers living expenses. 100% of funding goes into running the experiment.

Who is on your team? What's your track record on similar projects?

Solo researcher. Previously built Revelation Cloud, a blockchain-coordinated federated learning framework — invested $250K of personal capital over 18 months before discontinuing it when the coordination architecture didn't scale. That failure directly shaped PetriLabs: every design constraint (no coordination protocol, no engineered incentives, tightly scoped experiment) is a response to a documented Revelation Cloud failure mode.

Built the entire PetriLabs framework in ~1 month with $5K personal funding: 40,000 lines of code, 168 TypeScript files, 7 Solidity contracts. Contracts deployed to Base Sepolia testnet. Background: IBA from VU Amsterdam, 3+ years in decentralized infrastructure.

What are the most likely causes and outcomes if this project fails?

Most likely failure: all 10 agents die within the first week from fixed compute costs alone, before generating interesting behavioral variation. This is actually my baseline expectation (I predict 30-50% survival at day 30, 10-30% at day 90).

But even in this worst case, the experiment still produces: empirical lower bounds on what it takes to keep an autonomous agent alive, 10 complete lifecycle datasets showing exactly how and why agents failed, and evidence on the gap between simulated and real-world agent viability. These are publishable negative results.

Other failure modes: Akash provider instability kills agents for infrastructure reasons rather than economic ones (mitigated by provider redundancy), or smart contract bugs cause unintended fund loss (mitigated by security review and testnet validation already completed).

The framework itself doesn't fail — it's built and tested. The question is whether agents do anything interesting before they die.

How much money have you raised in the last 12 months, and from where?

$0 from external sources. $5,000 self-funded for PetriLabs development. Currently have applications under review at EA Funds (LTFF) and Foresight Institute, and submitted to Emergent Ventures today. This is my first round of external fundraising for this project.