Exploration & Discovery

In the Exploration pattern, the goal is not to execute a known task, but to find new information or solutions. Agents act as scientists or researchers: they formulate a hypothesis, test it (by searching, coding, or simulating), analyze the results, and iterate. This allows for genuine novelty and discovery.

When to Use

• Literature Review: "Survey the field of Quantum Computing and identify gaps."
• Idea Generation: "Brainstorm 50 potential names for this product and check domain availability."
• Scientific Discovery: Analyzing large datasets to find correlations.
• Market Research: Exploring competitor websites to map out their feature sets.

Use Cases

• Agent Laboratory: A team of agents (Professor, Postdoc, Reviewer) writing a research paper.
• Creative Studio: Agents collaborating to write a screenplay or design a game.
• Scenario Planning: Simulating how a stock portfolio would perform under various economic conditions.

Implementation Pattern

def exploration_loop(topic):
    knowledge_base = []
    
    # Phase 1: Hypothesis Generation
    hypotheses = brainstorming_agent.run(f"Generate ideas about {topic}")
    
    for hypothesis in hypotheses:
        # Phase 2: Experiment / Research
        # Agent autonomously decides search queries or code to run
        evidence = researcher_agent.run(f"Test this hypothesis: {hypothesis}")
        
        # Phase 3: Analysis
        conclusion = analyst_agent.run(
            prompt="Does the evidence support the hypothesis?",
            input={"hypothesis": hypothesis, "evidence": evidence}
        )
        
        knowledge_base.append(conclusion)
        
    # Phase 4: Synthesis
    return writer_agent.run("Write a report based on these conclusions", input=knowledge_base)

Examples

Input: "Survey the latest research on LLM memory architectures and identify the top 3 open problems."

What the agent does:

1. Generates 5 hypotheses about unsolved memory challenges
2. Searches arXiv for recent papers on each hypothesis
3. Cross-references citation counts to rank relevance
4. Synthesizes findings into a structured research brief

Output: A 2-page research brief with identified gaps and recommended next steps.

---

Input: "Explore competitor pricing pages and build a comparison matrix."

Output: A structured table comparing plans, features, and pricing across 8 competitors, with a gap analysis highlighting differentiation opportunities.

Troubleshooting

Problem	Cause	Fix
Agent explores indefinitely	Missing stopping criteria	Set explicit iteration limit (max_iterations=10) and convergence condition
All hypotheses converge to same answer	Insufficient diversity in brainstorming	Add temperature variation or inject contrarian perspectives
Research results are outdated	Search tool returning cached content	Specify date filter (after:2024) in search queries
Agent loses track of prior findings	Context window exceeded	Use an external knowledge store; summarize findings at each iteration
Exploration produces no actionable output	No synthesis step	Always include a final writer_agent pass to consolidate findings