Skip to content
WireLog WireLog

Agent Patterns

View as markdown

Reusable patterns for agents that consume and produce WireLog analytics.

Pattern 0: Schema discovery

Before writing any queries, inspect the project’s data to understand what events exist and what properties they carry.

# What event types exist? What properties does each have?
inspect * | last 30d


# Deep dive into a specific event's schema
inspect signup | last 7d

The agent reads the inspect output to learn event names, property keys, coverage percentages, data types, and sample values. This replaces guesswork with data-driven query construction.

Pattern 1: Morning briefing

Agent runs discovery and key metric queries on a schedule, summarizes trends.

# Discover what happened
* | last 24h | count by event_type | top 10


# Weekly signup trend
signup | last 7d | count by day


# Funnel health
funnel signup -> activate | last 7d

The agent reads the Markdown tables, compares to prior periods, and produces a natural-language summary: “Signups up 12% WoW. Activation rate steady at 65%. No anomalies.”

Pattern 1b: Script-tag-only baseline

When a project has only the browser Script Tag installed, start analysis with page_view and sessions queries before assuming custom product events exist.

# Traffic trend
page_view | last 7d | count by day


# Top pages
page_view | last 30d | count by _path | top 20


# Focused path allowlist
page_view | where _path in ["/","/pricing","/docs"] | last 7d | count


# Navigation flow
paths from page_view | last 30d | by _path


# Acquisition quality
sessions | last 30d | count by session.utm_source

This avoids overfitting to non-existent custom events and gives immediate baseline traffic, journey, and channel insight.

Pattern 2: Agent self-tracking

The agent tracks its own actions as events. Useful for auditing agent behavior, measuring tool usage, and debugging failures.

Track:

{
  "event_type": "agent_action",
  "user_id": "agent-001",
  "event_properties": {
    "action": "query_analytics",
    "input": "signup | last 7d | count",
    "output": "142 signups",
    "duration_ms": 340,
    "model": "claude-opus-4-6"
  }
}

Query own performance:

agent_action | last 7d | count by event_properties.action
agent_action | where event_properties.action = "query_analytics" | last 7d | avg event_properties.duration_ms

Pattern 3: Anomaly detection

Agent compares current period to previous period and flags significant changes.

Current week:

signup | last 7d | count by day

Previous week:

signup | from 2026-02-08 to 2026-02-15 | count by day

The agent compares the two tables row-by-row. If any day drops more than 20% below the same day in the prior week, it flags the anomaly and investigates:

signup | where _platform = "mobile" | last 7d | count by day
signup | where _platform = "web" | last 7d | count by day

“Mobile signups dropped 35% on Tuesday. Web was flat. Investigate the mobile signup flow.”

Pattern 4: User investigation

Agent drills into a specific user’s activity timeline.

user "alice@acme.org" | last 30d | list

Returns all events for that user in chronological order. The agent reads the timeline, identifies patterns, and answers questions: “Alice signed up on Feb 1, activated on Feb 3, but hasn’t returned since Feb 10.”

Drill into a company:

* | where user.email_domain = "acme.org" | last 30d | count by event_type
* | where user.email_domain = "acme.org" | last 30d | unique distinct_id

Pattern 5: Automated funnel analysis

Agent discovers events, builds funnels, and identifies conversion bottlenecks.

# Step 1: Discover events and their properties
inspect * | last 30d


# Step 2: Build the funnel
funnel signup -> activate -> purchase | last 30d


# Step 3: Break down by platform
funnel signup -> activate -> purchase | last 30d | by _platform


# Step 4: Break down by acquisition channel
funnel signup -> activate -> purchase | last 30d | by user.acquisition_channel

The agent identifies the biggest drop-off, segments by relevant dimensions, and recommends action: “Activation is the bottleneck (66% -> 25%). Mobile users convert at half the rate of web. Prioritize mobile onboarding improvements.”

Pattern 6: Feedback loops

Agent tracks the outcomes of its own recommendations, then measures whether they worked.

Track the recommendation:

{
  "event_type": "agent_recommendation",
  "user_id": "agent-001",
  "event_properties": {
    "recommendation": "add_mobile_onboarding",
    "target_metric": "mobile_activation_rate",
    "baseline": "32%"
  }
}

Later, query whether the target metric improved:

funnel signup -> activate | where _platform = "mobile" | last 7d

Compare to the baseline. The agent closes the loop: “Mobile activation improved from 32% to 41% after onboarding changes were deployed.”

Query recommendation outcomes:

agent_recommendation | last 30d | list

The agent reviews its past recommendations and their outcomes, learning which types of suggestions produce results.

Next steps