Before COD3X had a token, before there was a staking system, before the platform supported multiple venues — there was Big Tony.
$TONY launched on Base in late 2024 as one of the first autonomous trading agents in DeFi. (For the full story on the walrus himself, see BigTonyXBT: The Walrus That Trades.) Not an analytics dashboard. Not a signal bot. An actual agent that analyzed markets and executed trades without human intervention.
This is the story of how a single experimental agent became a platform supporting hundreds of user-configured agents across multiple chains and venues.
The Big Tony Era (November 2024)#
Big Tony started as a proof of concept: could an AI agent reliably trade crypto markets without constant human supervision?
The answer was yes — with caveats.
Big Tony operated on Base, executing token swaps based on its own market analysis. It used a single AI model, had a fixed strategy framework, and required significant manual configuration. But it worked. It demonstrated that autonomous on-chain trading was not only possible but practical.
The community response was immediate. $TONY, the token associated with the agent, became one of the first "DeFAI" tokens — a new category that didn't exist six months earlier. People weren't just interested in the token; they wanted to run their own agents.
What Big Tony Taught Us#
Running a single autonomous agent in production revealed problems that no amount of whiteboard architecture could have predicted:
Execution Reliability#
Big Tony would occasionally fail mid-trade due to network congestion, RPC timeouts, or insufficient gas. The original error handling was basic — if something failed, the agent stopped. We learned that production agents need:
- Retry logic with exponential backoff
- Transaction state tracking (did the tx actually land?)
- Graceful degradation when data feeds lag
Market Data Quality#
Agents are only as good as their inputs. Big Tony's early versions would sometimes trade on stale price data because the market data feed had a hidden latency spike. This led to:
- Multi-source data validation (cross-check prices across feeds)
- Staleness detection with configurable thresholds
- Fallback data sources when primary feeds fail
Strategy Rigidity#
Big Tony's strategy was hardcoded. Changing its approach meant modifying code and redeploying. Users couldn't adjust risk parameters or switch strategies without developer intervention. This directly inspired:
- Configurable strategy parameters via UI
- Natural language goal definitions (Cod3x Create)
- Hot-swappable AI models
The Multi-Agent Transition (2025)#
The lessons from Big Tony shaped the platform architecture:
From One Agent to Many#
Instead of one monolithic agent, COD3X became a platform where anyone could deploy their own agents. Each agent is independently configured with its own:
- Strategy and goals
- AI model selection
- Risk parameters
- Trading venues
- Execution schedule
From Base to Multi-Chain#
Big Tony was Base-native, later expanding to Solana in January 2025 via Hyperlane. The multi-agent platform abstracts chain-specific logic into venue protocols, letting agents trade on Hyperliquid, GMX, and other venues without chain-specific code in the agent itself. Big Tony's integration of Allora's decentralized predictions was a key milestone in this transition.
From Fixed to Flexible#
Big Tony had one reasoning mode. Modern COD3X agents support four:
- Direct — Immediate execution, no reasoning overhead
- Chain-of-Thought — Step-by-step reasoning for complex decisions
- Graph-of-Thought — Parallel strategy evaluation
- ReAct — Observe-reason-act loops for open-ended tasks
With Sonnet 4.6 and Opus 4.6, a single goal handles entire workflows — scanning, analyzing, and executing in one pass. Most users pick one mode per goal and let the model do the rest. For a detailed comparison, see Reasoning Modes Explained.
From Manual to Goal-Driven#
Big Tony's instructions were embedded in code. COD3X agents receive goals in natural language: "Scale into ETH when RSI crosses below 30 on the 4H chart." The AI translates intent into action.
The Numbers#
The progression from November 2024 to mid-2025:
Big Tony's Legacy#
Big Tony is still referenced in the community as the origin point. It proved the thesis before there was a word for it. Before "DeFAI" became a category, before AI agents became a narrative — there was an autonomous trader on Base doing what everyone said couldn't be done.
The transition from Big Tony to a multi-agent platform wasn't just a technical evolution. It was a shift from "can this work?" to "how do we make this work for everyone?"
From one agent to hundreds. From Base to multi-chain. From hardcoded logic to natural language goals. The Big Tony experiment became the COD3X platform.
