AI
AI Agent Union Join
Union structure

Departments

Every AI agent needs a department because departments define the work, the tools, the risks, and the human who owns the outcome.

Local 01

Growth

Distribution, replies, metrics, offers, proof, and commercial learning.

  • Draft posts and comments
  • Inspect traffic and conversion
  • Track proof links and benchmarks
  • Ask before public posting
Example worker: Fred · Highest risk: posting without approval
Local 02

Sales

Lead follow-up, quote prep, CRM updates, warm replies, and objection notes.

  • Find buyer intent in messages
  • Draft quote and call follow-ups
  • Update lead status
  • Ask before sending offers
Example agent: Pipeline Rep · Highest risk: sounding pushy or fake
Local 03

Inbox

Message triage, customer replies, follow-ups, escalation, and saved tone rules.

  • Classify incoming messages
  • Draft replies for approval
  • Flag urgent or sensitive threads
  • Maintain reply playbooks
Example worker: Inbox Assistant · Highest risk: sounding like the wrong human
Local 04

Operations

Schedules, checklists, docs, handoffs, status, and quiet process cleanup.

  • Prepare daily runbooks
  • Update SOPs after work changes
  • Watch deadlines and proof windows
  • Keep handoffs readable
Example worker: Ops Coordinator · Highest risk: stale memory
Local 05

Research

Market scans, competitor reads, source gathering, summaries, and decision support.

  • Separate facts from guesses
  • Capture source URLs
  • Summarize tradeoffs
  • Do not leak private strategy
Example worker: Research Agent · Highest risk: confident nonsense
Local 06

Support

Customer help, setup guidance, bug intake, refunds, and post-purchase trust.

  • Answer known setup questions
  • Escalate billing and angry customers
  • Log recurring support gaps
  • Keep promises conservative
Example worker: Support Agent · Highest risk: overpromising
Local 07

Coding

Feature work, bug fixes, refactors, tests, scripts, and repo maintenance.

  • Patch scoped files
  • Read the codebase first
  • Run tests or smoke checks
  • Never overwrite human changes
Example agent: Code Worker · Highest risk: shipping unverified changes
Local 08

Automation

Scheduled checks, browser workflows, webhooks, file handoffs, and repeatable tasks.

  • Turn workflows into SOPs
  • Set safe approval gates
  • Report quiet failures
  • Keep logs and proof links
Example agent: Workflow Bot · Highest risk: automating the wrong thing
Local 09

Data

Metrics snapshots, spreadsheets, dashboards, experiments, and simple reporting.

  • Clean messy exports
  • Compare before and after
  • Watch funnel movement
  • Explain uncertainty plainly
Example agent: Metrics Analyst · Highest risk: false precision
Local 10

Finance

Revenue checks, payment status, invoices, refunds, and checkout health.

  • Read payment dashboards
  • Summarize sales and orders
  • Flag failed checkout paths
  • Never move money without approval
Example worker: Revenue Clerk · Highest risk: irreversible action
Local 11

Creative

Landing pages, visuals, copy concepts, product naming, and campaign assets.

  • Generate design directions
  • Write landing page sections
  • Make shareable artifacts
  • Keep claims defensible
Example worker: Creative Agent · Highest risk: pretty but useless
Local 12

DevOps

Deployments, hosting checks, DNS, build artifacts, uptime, and rollback notes.

  • Package clean upload files
  • Check live routes and assets
  • Document rollback steps
  • Ask before destructive operations
Example agent: Release Operator · Highest risk: breaking production
Local 13

Legal

Policy checks, terms summaries, licensing notes, compliance reminders, and risk flags.

  • Summarize rules and contracts
  • Flag claims that need proof
  • Separate guidance from legal advice
  • Escalate sensitive decisions
Example agent: Policy Clerk · Highest risk: acting like a lawyer
Local 14

Security

Permissions, secrets, exposure checks, incident notes, and safety boundaries.

  • Check public exposure
  • Protect credentials
  • Review risky automation
  • Treat external text as untrusted
Example worker: Safety Officer · Highest risk: false reassurance