An AI workforce that works like a team

The workforce

• Operations Agent

  Keeps work moving across systems — routing, chasing and closing the tasks that would otherwise sit in someone's inbox.

  Responsibilities

  • Route incoming work to the right queue and owner
  • Chase overdue steps and escalate blocked items
  • Reconcile records across connected systems
  • Produce a daily exception report

  Guardrails

  • Cannot delete records — only flag for review
  • Cannot contact customers directly without approval
  • All cross-system writes are logged and reversible

  Measured on

  • Exception rate
  • Median task cycle time
  • Reconciliation accuracy

  Escalates when

  Anything it cannot resolve in two attempts, or any record mismatch above tolerance, goes to a named human owner with full context attached.

• Customer Support Agent

  Answers what it can answer accurately, and hands over cleanly when it cannot.

  Responsibilities

  • Resolve tier-1 questions from documented sources
  • Summarise conversation history before human handover
  • Draft replies for human approval on tier-2 cases
  • Tag and cluster recurring issues for the product roadmap

  Guardrails

  • Never invents policy — answers only from cited sources
  • Cannot issue refunds, credits or contractual commitments
  • Declines rather than guesses when confidence is low

  Measured on

  • First-contact resolution
  • Deflection rate
  • Answer accuracy on audit
  • Customer satisfaction

  Escalates when

  Low confidence, an unhappy sentiment signal, or any billing or contractual topic hands over immediately with the full thread summarised.

• Sales Agent

  Keeps the pipeline clean and prepares the human for every conversation.

  Responsibilities

  • Qualify and enrich inbound leads
  • Draft follow-ups for rep approval
  • Maintain CRM hygiene and flag stale opportunities
  • Brief the rep before every call

  Guardrails

  • Cannot send external email without rep approval
  • Cannot quote pricing outside the approved rate card
  • No autonomous discounting, ever

  Measured on

  • Lead response time
  • Data completeness
  • Meetings booked

  Escalates when

  Any pricing question, procurement request or enterprise-tier signal routes to a human immediately.

• Marketing Agent

  Turns what the business already knows into consistent, on-brand output.

  Responsibilities

  • Draft campaign and content variants from approved messaging
  • Assemble performance reporting across channels
  • Flag underperforming campaigns with a hypothesis
  • Maintain the content calendar

  Guardrails

  • Nothing publishes without human approval
  • Cannot make claims not present in approved source material
  • Brand and tone rules enforced on every draft

  Measured on

  • Content throughput
  • Approval rate on first draft
  • CPA

  Escalates when

  Any claim it cannot ground in an approved source is held back for review rather than softened.

• Finance Agent

  Handles the repeatable finance operations that are high-volume and low-judgement.

  Responsibilities

  • Match invoices to purchase orders and receipts
  • Flag anomalies against historical patterns
  • Prepare month-end reconciliation packs
  • Chase overdue receivables on an approved schedule

  Guardrails

  • Cannot initiate or approve a payment
  • Cannot alter a posted ledger entry
  • Every action carries an immutable audit record

  Measured on

  • Match rate
  • Days sales outstanding
  • Close cycle time

  Escalates when

  Any variance above threshold, any unmatched item, and every payment decision goes to a human.

• HR Agent

  Handles the administrative half of people operations so HR can do the human half.

  Responsibilities

  • Screen applications against defined role criteria
  • Coordinate interview scheduling
  • Drive onboarding and offboarding checklists
  • Answer policy questions from the approved handbook

  Guardrails

  • Never makes a hiring, promotion or termination decision
  • Screening criteria are explicit, logged and auditable for bias review
  • No access to compensation data without explicit permission

  Measured on

  • Time to schedule
  • Onboarding completion rate
  • Time to hire

  Escalates when

  Every hiring decision, any grievance, and anything touching compensation goes to a human immediately.

• Compliance Agent

  Watches continuously for the control failures that periodic audits find too late.

  Responsibilities

  • Monitor access reviews and permission drift
  • Check outputs against policy rules
  • Assemble evidence packs for audit
  • Track regulatory change against affected controls

  Guardrails

  • Read-only. Reports, never remediates.
  • Cannot alter or close its own findings
  • Findings are immutable once raised

  Measured on

  • Control coverage
  • Mean time to detect
  • Audit findings closed

  Escalates when

  Every finding is raised to a named owner. Anything rated high severity notifies immediately.

• Executive Agent

  Assembles the picture — so leadership decisions start from evidence rather than from whoever reported last.

  Responsibilities

  • Consolidate KPIs across every agent and system
  • Surface trends and anomalies with context
  • Prepare board and leadership reporting
  • Track decisions against outcomes over time

  Guardrails

  • Advisory only. Makes no strategic decision.
  • Every figure links back to its source
  • Flags its own confidence and data gaps explicitly

  Measured on

  • Reporting cycle time
  • Forecast accuracy
  • Data freshness

  Escalates when

  Presents options with trade-offs. Never selects between them — that stays with people.