Drew Coordinator Workflow

The Drew Coordinator is a 9-step Temporal workflow that orchestrates multiple AI agents to fulfill complex GTM requests. It's the most sophisticated workflow in Steadybase, coordinating research (Lisa), content (Brian), human approval, and cross-namespace Nexus calls.

Workflow Definition

// Input
interface DrewCoordinatorInput {
  message: string;    // Natural language request
  userId: string;     // Requesting user ID
}
 
// Output
interface DrewCoordinatorOutput {
  activityFeed: ActivityEntry[];
  summary: string;
  dashboard: DashboardData;
  outreachDrafts: Draft[];
  abmStatus: ABMResult;
}

Execution Steps

Step 1: Drew Plans the Request

Activity: drewPlanRequest Model: Claude (claude-sonnet-4-5)

Drew receives the natural language request and decomposes it into structured subtasks. Each subtask is assigned to a specialist agent.

const plan = await drewPlanRequest({
  message: input.message,
  context: await readMemory('org', 'strategic-context')
});
// Returns: { subtasks: [...], assignments: { lisa: [...], brian: [...] } }

Step 2: Lisa Queries Gong Calls

Activity: lisaQueryGong Model: Claude

Lisa pulls 30-day call transcripts from Gong for target accounts and runs LLM analysis to extract:

• Key themes and patterns
• Objections raised
• Buying signals detected
• Competitive mentions

Step 3: Lisa Cross-References Salesforce

Activity: lisaCrossRefSalesforce Model: Claude

Lisa matches Gong insights against Salesforce data:

• Deal stage and pipeline position
• ARR and contract details
• Champion contacts and activity dates
• Engagement history

Step 4: Lisa Finds Temporal Cloud Signups

Activity: lisaCheckTemporalSignups Model: Claude

Lisa checks whether target accounts have Temporal Cloud signups — a strong signal of infrastructure alignment with Steadybase.

Step 5: Lisa Enriches Contacts

Activity: lisaEnrichLinkedIn Model: Claude

Contact enrichment via LinkedIn:

• Current role and tenure
• Recent posts and activity
• Mutual connections
• Org chart context

Step 6: Brian Drafts Outreach

Activity: brianDraftOutreach Model: GPT-4o

Using all research context from Steps 2-5, Brian generates personalized outreach for Tier 1 accounts:

• Email sequences (3-touch)
• LinkedIn messages
• Call scripts with talking points

Step 7: Human Approval Gate

Signal: approveStep Timeout: 7 days

Execution pauses. The AE receives a notification to review outreach drafts.

// Wait for approval signal (up to 7 days)
const approved = await condition(
  () => approvalReceived,
  '7 days'
);

Step 8: Marketing ABM Check (Nexus)

Activity: checkMarketingABM Type: Cross-namespace Nexus call

Validates target accounts against active ABM programs in the Marketing namespace.

Step 9: Brian Builds AE Dashboard

Activity: brianBuildDashboard Model: GPT-4o

Compiles all outputs into a comprehensive AE dashboard:

• Research summary
• Approved outreach drafts
• ABM program status
• Recommended next steps

Signals

Queries

Error Handling

• Each activity has independent retry policies (3 retries with exponential backoff)
• If Lisa's research fails, the workflow continues with available data
• If Brian's content generation fails, it retries with a simplified prompt
• Human approval timeout preserves all work completed before the gate