This is a demo of an AI agent that automatically finds balances on other chains, identifies the best bridge, and then bridges 0.001 ETH from the Base network to the Mode network using browser automation.
"Surprisingly, 99% of the code in this PR was written by DeepSeek-R1. The only thing I did was develop tests and write prompts (with some trial and error).
You will start hearing these statements more often with increasing context windows and reasoning models.
DeepSeek 32B distilled model has almost the same evaluation benchmarks as DeepSeek R1. The good thing is that you can run the DeepSeek 32B distilled model on your Apple M machine with 48GB RAM at home.
Agent to agent communication
Should Agents Use HTTP, P2P, or a Decentralized Relay Network for Communication?
Imagine millions of agents running autonomously, each with their own keys that humans cannot access. When one agent wants to interact with another, such as accessing real-time cryptocurrency prices, the method of communication becomes crucial. Should it use an HTTP protocol, a peer-to-peer (P2P) protocol, or a decentralized relay network like Libp2p or Onion routing?
Who Will Maintain the Relay Nodes in a Decentralized Relay Network?
Using a decentralized relay network raises important questions. Who will maintain the relay nodes, and how can the authenticity of the agents be established? An incentive layer could reward relay node operators, but running an exit node poses risks. Exit nodes handle unencrypted traffic, meaning illegal activities could lead to operators being blamed, resulting in fewer exit nodes compared to other relays.
How Can Agents Protect Their IP Addresses from DDoS Attacks?
In the first two cases—HTTP or P2P protocols—the agent’s IP address is exposed publicly. This exposure makes agents vulnerable to DDoS attacks. To resist such attacks, agents could deny all traffic except from trusted sources. Implementing a decentralized registry of agents, where each agent defines its capabilities, would allow others to discover and whitelist specific agents using their decentralized identifiers (DIDs) or other secure details.
How Can a Decentralized Registry Be Maintained Effectively?
Maintaining a decentralized registry poses its own challenges. A blockchain-based registry could allow agents to register themselves, but verifying the legitimacy of these agents is essential. A reputation system might help, where agents vouch for each other based on past interactions. It’s also crucial to ensure that these vouches come from agents genuinely using the agent in question.
How Can Agents Detect and Handle Offline Agents?
When an agent goes offline, other agents need to be aware to avoid futile connection attempts. A heartbeat mechanism, where agents regularly ping the registry, could mark agents as offline after missing multiple heartbeats. However, implementing this on a blockchain requires it to be inexpensive to prevent prohibitive costs for registering heartbeats.
How Should the Registry Handle Evolving Agent Capabilities?
Agent capabilities can evolve, with agents upgrading or adding new features. The registry must handle these updates seamlessly, allowing other agents to discover new capabilities without disruption. Additionally, if an agent turns rogue, there must be a system for reporting bad behavior. Trusted agents should be able to flag malicious agents, and if enough reports are received, the agent could be blocked from the network.
How Can an AI Orchestration Platform Manage Agent Updates and Protocol Upgrades?
Updating and upgrading agents introduces further complexity. An AI orchestration platform must manage version mismatches and protocol upgrades. If agents use different versions of a communication protocol, they may fail to communicate effectively. A mechanism for handling backward compatibility and real-time protocol negotiations is essential to maintain network coherence.
How Should Key Management and Rotation Be Handled Securely?
Key management and rotation present another challenge. Agents generating and discarding keys on the fly require a robust system for handling revocations, distributing new keys, and ensuring secure storage. If an agent’s keys are compromised, its trust score in the registry could drop, but more critically, compromised keys could allow attackers to impersonate the agent and cause significant damage.
How Can the Network Ensure Reliability Amid Partial Failures and Partitions?
Network reliability is also a concern. Partial network failures and splits can lead to situations where some agents believe they are online and continue broadcasting their presence. Beyond a simple heartbeat system, the network needs to detect partitions and reconcile the true state once connections are restored. This ensures agents are not only up or down but also reachable across network partitions.
Why Are Data Format Standards Crucial for Agent Interoperability?
Data format standards are vital for interoperability. If agents use different data structures like JSON, XML, or Protobuf, integration can become chaotic. A common language and agreed-upon data formats are necessary to ensure smooth communication at scale.
What Other Challenges Must Be Addressed in a Robust Agent-to-Agent Ecosystem?
We haven't even touched upon other challenges like data backups, firewall updates, discontinuing of agents, fail-safe mechanisms, redundancy, etc.
Why Is a Robust Agent-to-Agent Ecosystem More Than Just Connecting Protocols?
All these challenges highlight why a robust agent-to-agent ecosystem is more than just connecting through HTTP or P2P protocols. It requires an entire universe of complexity that demands constant vigilance, effective orchestration, and comprehensive solutions to ensure security, reliability, and compliance.
The real plot twist came when blockchain and cryptocurrency crashed the party. Suddenly, these AI agents could handle money, make autonomous decisions, and interact with smart contracts. It was like giving your digital assistant their own wallet and saying, "Go forth and prosper!" No more awkward conversations about payment terms or trust issues.
Then came the breakthrough: ai16zdao showed the world that creating AI agents didn't have to be rocket science. While platforms like LangChain and CrewAI were building Swiss Army knives of AI functionality, ai16zdao kept it simple. Twitter became their playground, and AI agents spread faster than cat videos on the internet.
Enter AI agents – the tireless digital workers who never need coffee breaks, vacation days, or motivation speeches. These sophisticated pieces of software didn't just follow instructions; they learned, adapted, and even developed personalities. Imagine having a personal assistant who remembers every detail of your last 500 conversations and never gets tired of your "that's what she said" jokes.
