Home · Practices · AI Agents

Agents built with approval gates, audit trails and human control.

Governed AI agents for regulated work — validated input, human approval, audit-grade evidence, by default. Five agents live across Banking & Finance Ops, Project Management, Legal Intake, Supplier Assurance and Energy Operations.

01 · Architecture

Every agent. Same architecture.

Every agent we build follows the same five-stage pattern. The approval gate is foregrounded for a reason.

1 item1 itemapprovedexecuted{}Inputvalidated01 · INPUTTool callsmodel · retrieval · api02 · AIHUMAN REQApproval gatehuman required03 · GATEActionexecuted04 · ACTIONAudit logevidence05 · LOGSYSTEMAGENTHUMANAGENTSYSTEM
AGENT-ARCH v1.0 · WORKFLOW VIEW
audit_log> 2026-06-07T14:22Z · agent=legal-intake · action=draft_brief · approver=K.Park · evidence=#A4621
Validated input

Every request schema-checked and scoped before the agent moves.

Human gate

High-risk actions require explicit human approval before execution.

Audit trail

Every decision logged with timestamp, owner and evidence reference.

02 · Live workflow

Here’s what one of our agents
actually looks like.

The five-stage architecture above is the conceptual model. This is the Legal Intake agent as our team sees it in build view — every node, every gate, every audit hook, in production.

LEGAL INTAKE AGENT v1.0 · PRODUCTION
17 NODES · 3 APPROVAL GATES · FULL AUDIT TRAIL
1 item 1 item 1 item 1 item approved 3 items 3 items {}Matter intake04 · webhook ?Score Routermode: rules05 · router HUMAN REQCompliance gate06a · approval Build standard06c · gate {}Escalation pkg06b · build card Send notification06d · email {}Capture decision06e · log {}Allowlist07a · projection Generate brief07 · audience-tunedx3 audiences {}Validate content08b · wording check3 items in Merge & audit log07e · evidenceretention: 7y HIGH-RISK PATH · HUMAN APPROVAL REQUIRED STANDARD PATH · LOGGED, AUTO-APPROVED
EVERY NODE LOGGED · EVERY DECISION OWNED · EVERY GATE EVIDENCED
LEGAL-INTAKE · WORKFLOW v1.3
audit_log> 2026-06-08T14:22Z · agent=legal-intake · node=06a-compliance-gate · action=approve · approver=K.Park · evidence=#A4621
03 · Live agents

Five agents. Ready to commission.

Each agent is purpose-built against a department’s actual workflow. Phase 1 demo on synthetic data in 4 weeks — no production access required.

AGENT · 01 · LIVE

Banking & Finance Ops

Transaction monitoring, reconciliation flag triage, exception routing. Built for finance teams under APRA CPS 230 obligations.

SCOPE · recon · flag triage · exception routing
AGENT · 02 · LIVE

Project Management

Status synthesis, risk surfacing, stakeholder update drafts. Reduces project office overhead while keeping humans on every decision.

SCOPE · status · risk · stakeholder updates
AGENT · 03 · LIVE

Legal Intake

Matter classification, brief drafting, conflict checks. Built for in-house legal teams and approved external counsel intake.

SCOPE · intake · brief drafting · conflict checks
AGENT · 04 · LIVE

Supplier Assurance & Onboarding

Vendor risk assessment, due-diligence packs, AI risk attestation. Built for procurement and supplier risk teams.

SCOPE · due diligence · AI attestation · onboarding
AGENT · 05 · LIVE

Energy Operations

Operational anomaly review, work-order generation, regulatory reporting drafts. Built for utilities under SOCI Act obligations.

SCOPE · anomalies · work orders · reg reporting
AGENT · 06 · IN DEVELOPMENT

Custom build — your department

If your workflow is sufficiently regulated, repeatable and decision-bounded, we can purpose-build an agent for it. Discovery is two weeks.

SCOPE · mission alignment · workflow mapping · build spec
04 · Build process

Five phases. Three sign-offs.

No off-the-shelf compromise. Every agent is built against the department’s actual workflow and released only after executive sign-off.

  1. 01

    Mission Alignment

    What the function exists to do. What the agent must never do. Written, signed, and held against the build.

  2. 02

    Workflow Mapping

    How the work actually happens — observed, not assumed. Identifies the decisions the agent will take and the gates a human must hold.

  3. 03

    Build

    The agent is built against the mission and workflow. Approval gates, audit logging, and rollback baked in from day one. Not retrofitted.

  4. 04

    Departmental Review

    The team that will use the agent reviews it on synthetic data. Sign-off required before production access is even considered.

  5. 05

    Executive Release

    Executive sign-off against documented risk posture, control set and rollback plan. Then — and only then — the agent goes live.

05 · Phase 1 demo

A working agent in four weeks. No production access required.

Phase 1 demos run on synthetic data that mirrors your workflow. Leadership sees what the agent does — and what the gates and audit trail look like — before any production access is granted.

This is the right way to show your board what’s possible before a full build is commissioned.

Discuss a Phase 1 demo
4-WEEK DEMO · SYNTHETIC DATA · NO PROD ACCESS · FIXED FEE
06 · Governance built in

Every agent. Same controls. By default.

CONTROL · 01

Named owner

Every agent has a named human owner in the department that uses it. Not a vendor, not IT.

CONTROL · 02

Kill-switch

Documented kill-switch authority and tested rollback procedures. Refreshed every six months.

CONTROL · 03

Approval gates

High-risk actions queue for human approval. Auto-allowed actions are still logged with role and time.

CONTROL · 04

Audit log

Every material decision logged with timestamp, owner, evidence reference and retention period.

CONTROL · 05

Monitoring

Drift detection, performance baselines and anomaly alerting on every production agent.

CONTROL · 06

Reporting

Quarterly agent posture report to executive and risk committee. Same format every time.

CPS 230 · CPS 234 · AS ISO/IEC 42001 · ASD ISM · NIST AI RMF · AICD · Privacy Act 1988 (ADM 2026)
06 · Start here

The first call is diagnostic.

Thirty minutes. No pitch. No proposal until it makes sense.

Book a diagnostic
30-MINUTE DIAGNOSTIC CALL · VIDEO OR MELBOURNE · BOOK DIRECTLY