AI agent workforce · LendEasy Docs

Run AI as bounded, attributable workers with signed context, least-privilege tools, expiring sessions, action-level policy, human oversight, and one causal audit trail.

LendEasy treats an AI agent as a worker identity—not an invisible feature inside a screen. It receives a task, opens a bounded session, acts only through declared tools, produces attributable output, and ends in an explicit state. It never borrows a human’s identity or permission. The roster itself — Nova, Cookie, Docgest, and Arya — is introduced in Meet the agents; this page is the contract they all work under.

Control stack

tenant policy and suppression
          ↓
task eligibility + autonomy ceiling
          ↓
signed work packet + least-privilege tools
          ↓
AI session budget and model/prompt version
          ↓
tool-level authorization + execution-time compliance
          ↓
output review / governed action gateway
          ↓
quality measurement + evidence graph

Every layer can narrow authority. No layer can widen the ceiling established above it.

The public API provisions human users and AI workers, exposes retained AI outputs and review, and supports scoped suppression. Session creation itself runs at an internal runtime boundary; the /ai-sessions contract below is its client surface — see API availability for the current boundary.

An agent exists as a registration before it can be dispatched. POST /v1/servicing/workers declares its adapter contract, queue and task-type eligibility, capacity, and the three grant lists — factGroupGrants, toolGrants, and eligibleActionCodes — that set the ceiling every layer below can only narrow. Workforce & worker registration covers that contract and the human-user equivalent.

Session lifecycle

LifecycleAI work session

REQUESTEDtask and objective identified

GATEDeligibility, suppression, and packet verified

ACTIVEbounded facts, tools, time, and budget

OUTPUT PRODUCEDcitations and confidence stored

REVIEWEDaccepted, edited, rejected, or escalated

CLOSEDtool ledger and outcome frozen

GATED→DECLINEDnot eligible, suppressed, or outside scope

ACTIVE→MISSING FACTSrequired grounding unavailable

ACTIVE→STOPPEDkill switch, human takeover, or budget

ACTIVE→FAILEDruntime failure with no silent completion

Requesting a session

curl -X POST "$BASE/v1/servicing/ai-sessions" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ai-session-case-hbr-col-51-next-action" \
  -d '{
    "caseId": "case_HBR_COL_51",
    "agentProfile": "collections_assistant",
    "objective": "Summarize the case and recommend the next compliant step.",
    "autonomyLevel": "A2"
  }'

The response pins workPacketVersion, session expiry, and effective autonomy. The requested level may be reduced by task, channel, sensitivity, jurisdiction, model qualification, or current quality policy.

Tool boundary

Reading a masked balance, drafting a message, preparing a promise, submitting an interaction, and changing a loan are different tools. Each has its own input schema, permission, allowed session profile, compliance gate, idempotency behavior, and evidence.

An agent cannot call a raw database, provider SDK, or hidden account command. Financial and contract changes go through the same public typed action path as authorized humans.

Output attribution

AI output retains agent profile, model and prompt versions, work-packet digest, citations, tool ledger, confidence, produced time, and any reviewer edits. Human approval does not relabel AI-authored content as human-authored; the graph shows both contributions.

Failure behavior

The agent returns missing_facts instead of estimating an authoritative balance, declined instead of acting outside scope, and stopped when suppression arrives. Partial text and tool calls remain evidence. A failed session never marks its task complete automatically.

Chained agents

Triage, summarization, recommendation, and drafting may form a chain, but the chain’s effective action space is the intersection of its controls. Splitting a prohibited action across agents does not make it permitted, and one agent cannot approve another agent’s high-impact action where human approval is required.

Explore Work packets & grounding, the Autonomy ladder, and Human review & kill switch.