How Agents Think
A deep dive into AI reasoning for markets. Understand how your agent builds context, analyzes multiple signals, makes decisions, and handles uncertainty.
Traditional trading systems operate on explicit rules: IF price crosses moving average AND volume exceeds threshold THEN buy. These rules are brittle. They don't account for context. They can't adapt to situations they weren't programmed for.
COD3X agents use large language models (LLMs)—the same AI architecture behind the most capable AI assistants. These models don't execute rules. They reason. They understand language. They weigh evidence. They consider context. They express uncertainty.
Before making any decision, your agent builds a comprehensive mental model of the current situation. This context includes everything relevant to evaluating opportunities and risks.
What Goes Into Context
| Context Layer | What It Contains | Why It Matters |
|---|---|---|
| Your Strategy | Goals, risk parameters, stated preferences | Aligns every decision with your intent |
| Current Positions | Open trades, entry prices, unrealized P&L | Prevents overexposure, informs exits |
| Recent Actions | Last trades, recent analysis, past decisions | Maintains consistency, avoids contradictions |
| Market State | Price action, indicators, regime assessment | Grounds decisions in current reality |
| News & Sentiment | Breaking news, social signals, alpha sources | Incorporates qualitative information |
| On-Chain Data | Flows, whale movements, token metrics | Adds blockchain-native intelligence |
Context Window Management
LLMs have limited context windows—the amount of information they can consider at once. COD3X intelligently manages this by prioritizing relevant information:
- Recent data is weighted more heavily than historical
- Information relevant to current positions gets priority
- Data matching your strategy's focus areas is emphasized
- Irrelevant noise is filtered before reaching the model
Your agent doesn't rely on a single signal. It synthesizes multiple data sources into a unified view, the same way a professional trader would cross-reference different inputs before making a decision.
Technical Analysis
Price patterns, support/resistance, indicators (RSI, MACD, Bollinger), volume analysis, trend identification.
On-Chain Intelligence
Whale wallet tracking, DEX flow analysis, token holder distribution, smart money movements, protocol metrics.
News & Events
Breaking headlines, announcements, regulatory news, macro events, CT alpha. Parsed for trading relevance.
Social Sentiment
Twitter/X activity, Discord chatter, Telegram groups, Reddit sentiment. Filtered for signal vs noise.
Market Structure
Funding rates, open interest, liquidation levels, order book depth, correlation analysis.
Your Strategy
Stated goals, risk tolerance, preferred setups, excluded assets, time constraints. The lens through which all data is interpreted.
Signal Synthesis
The agent doesn't just check if signals align—it weighs conflicting information thoughtfully:
| Scenario | How Agent Handles It |
|---|---|
| Technical bullish, news bearish | Assesses which signal is stronger, may reduce position size or wait |
| Strong setup but crowded trade | Considers contrarian risk, may pass or wait for better entry |
| Good entry but high market volatility | Adjusts position size down, tightens stops |
| Multiple weak signals align | Treats confluence as strengthening thesis |
| Your strategy conflicts with data | Follows your strategy but flags the conflict in reasoning |
When evaluating a potential trade, your agent follows a structured decision process. This isn't a rigid algorithm—it's a reasoning framework that adapts to each situation.
Opportunity Identification
Scan for setups matching your strategy criteria. Filter noise, focus on relevant signals.
Context Evaluation
Build full picture: market regime, news, on-chain, current positions, risk budget.
Strategy Alignment Check
Does this opportunity match your stated goals and preferences? Any conflicts?
Risk Assessment
Calculate position size, define stop loss, project potential outcomes, check portfolio impact.
Confidence Scoring
How confident is the agent in this trade? What could go wrong? Any missing information?
Decision & Execution
Act, wait, or pass. Log reasoning. Execute if appropriate. Monitor position.
Confidence Scores
Every potential trade gets a confidence score. This isn't arbitrary—it reflects the agent's assessment of how well the setup matches your strategy and how favorable the conditions are.
| Confidence | What It Means | Typical Action |
|---|---|---|
| 90%+ | Exceptional setup, strong alignment, favorable conditions | Full position size, aggressive entry |
| 75-90% | Good setup, minor concerns or missing data | Standard position size |
| 60-75% | Decent setup, notable uncertainties | Reduced position size |
| 40-60% | Marginal setup, significant concerns | Typically pass, or minimal size |
| Below 40% | Weak setup, poor alignment, or high risk | Pass |
Your agent maintains two types of memory that influence its decisions over time.
Short-Term Context
Within a trading session, the agent remembers:
- Recent market movements and analysis
- Positions opened and closed
- Opportunities passed and why
- Your chat instructions and feedback
Long-Term Pattern Recognition
Over time, COD3X analyzes your agent's trade history to identify patterns:
| Pattern Type | Example Insight | How It Helps |
|---|---|---|
| Win Conditions | "Trades entered on volume breakouts have 68% win rate" | Reinforces what works |
| Loss Patterns | "Losses often follow entries during high funding rates" | Flags risky conditions |
| Timing Insights | "Best performance in Asian trading hours" | Optimizes scheduling |
| Asset Affinity | "Strong results on ETH, weak on meme coins" | Refines asset focus |
| Strategy Drift | "Recent trades don't match stated momentum strategy" | Identifies configuration issues |
Improvement Suggestions
Based on pattern analysis, your agent surfaces suggestions for improving your strategy:
- Parameter adjustments (tighter stops, different take profit levels)
- Strategy refinements (additional filters, timing changes)
- Risk recalibration (position sizing based on actual volatility)
- Asset allocation changes (focus more on what's working)
One of the most important differences between AI reasoning and rule-based systems is how they handle uncertainty. Your agent knows when it doesn't know.
When Agents Don't Act
| Uncertainty Type | Example | Agent Response |
|---|---|---|
| Data Gaps | On-chain data feed delayed | Note uncertainty, reduce confidence, may pause |
| Conflicting Signals | Technical says buy, sentiment says sell | Weight signals, often wait for clarity |
| Unusual Conditions | Volatility way outside normal range | Conservative posture, smaller positions |
| Strategy Ambiguity | Setup partially matches stated strategy | Ask for clarification or pass |
| Low Confidence | Setup is marginal at best | Pass rather than force trades |
The Value of Inaction
One of the most valuable things your agent does is not trade. When conditions don't favor action, the agent waits. This discipline—impossible to maintain as a human watching charts—is where much of the edge comes from.
Every decision your agent makes is fully transparent. You can always see:
- What data was considered — The specific inputs that informed the decision
- How it was interpreted— The agent's analysis of that data
- Why this action was chosen — The reasoning that led to the conclusion
- What alternatives were considered — Other options and why they were rejected
- Confidence level — How certain the agent was
- Risk assessment — What could go wrong and how that was mitigated