--- title: "AB-100 Study Guide β€” Agentic AI Business Solutions Architect" description: "An interactive study guide built on 7 memory techniques to help you pass the Agentic AI Business Solutions Architect Expert exam." url: "https://lucidlabs.com.au/insights/ab-100" --- # Master the AB-100 An interactive study guide built on 7 memory techniques to help you pass the Agentic AI Business Solutions Architect Expert exam. Plan AI-Powered Business Solutions 25-30%Design AI-Powered Business Solutions 25-30%Deploy AI-Powered Business Solutions 40-45% What it covers Copilot Studio agents (topics, generative orchestration, tools, knowledge sources, computer use), Microsoft Foundry Agent Service, Azure OpenAI models, Model Context Protocol (MCP), Agent2Agent (A2A) protocol, Dataverse and Power Fx, multi-agent orchestration, agent security and prompt injection defence, responsible AI governance, ALM and environment strategy, telemetry-driven monitoring and audit trails. Ideal for Expert-level solution architects designing agentic AI solutions across Dynamics 365, Microsoft Copilot, and Power Platform. Aspire to this if You're an experienced Power Platform, Dynamics 365, or Azure AI architect who already holds an associate credential (PL-400, AI-102, AB-620, MB-500, AI-103) and is ready to architect autonomous, multi-agent business solutions at enterprise scale. Section 1 / Spatial Memory ## The Map Tap any component to see what you need to know for the exam. **🧭 AI Strategy & ROI** Business Case Β· Value Realisation **🧩 Agent Architecture Planning** Single vs Multi-Agent Β· Platform Fit **πŸ” Governance & Compliance Planning** Data Residency Β· Risk Assessment **πŸŽ›οΈ Copilot Studio** Topics Β· Tools Β· Knowledge Β· Computer Use **🏭 Microsoft Foundry** Agent Service Β· Model Catalogue **πŸ”Œ Model Context Protocol** MCP Servers Β· Tool Exposure **πŸ”— Multi-Agent Orchestration & A2A** Agent2Agent Protocol Β· Hand-Off Design **πŸ“Š Dataverse & Power Fx** Data Layer Β· Low-Code Logic **πŸš€ Environment Strategy & ALM** Dev Β· Test Β· Prod Β· Pipelines **πŸ›‘οΈ Agent Security & Prompt Defence** Prompt Injection Β· Access Control **πŸ“‘ Telemetry & Audit Trails** Monitoring Β· Performance Β· Compliance **βš–οΈ Responsible AI Operations** Content Safety Β· Groundedness Section 2 / Narrative Memory ## The Story Follow the narrative to build a mental model of how everything connects. 🧭 ### The Strategy Table Before a single agent is designed, the architect sits at the strategy table. Which processes are agentic-ready β€” high volume, judgement-plus-rules, measurable outcome? What does success look like in deflection rate, resolution time, and cost per interaction? Executive sponsorship is secured here, because agentic solutions change how people work. **Exam Intel**ROI = deflection rate + time-to-resolution + cost-per-interaction vs build/model cost. Adoption metrics validate value post-launch. Candidate is assumed expert-level β€” no beginner content. 🧩 ### The Architecture Whiteboard Now the shape of the solution gets drawn. Copilot Studio for a low-code, connector-driven agent β€” or Foundry Agent Service for a code-first, model-flexible one? A single agent owning one bounded task, or a topology of specialists handed off through an orchestrator? Every decision here is traded against latency, cost per call, and what happens when a tool goes down. **Exam Intel**Copilot Studio = low-code, conversational, connector-driven. Foundry Agent Service = pro-code, model-flexible. Topologies: orchestrator-worker, peer-to-peer, hierarchical hand-off. πŸ” ### The Governance Review Every agent gets a risk assessment before it gets a build ticket: what data will it touch, what can it do without asking, what needs a human in the loop? Data residency dictates region and model choice. Microsoft Entra Agent ID gives the agent its own identity, so it can be governed like any other account β€” including being switched off. **Exam Intel**Pre-build risk assessment: data touched, autonomous actions, approval gates. Data residency β†’ region/model choice. Entra Agent ID = agent identity for access + lifecycle governance. Agent sprawl is a named exam risk. πŸŽ›οΈ ### The Copilot Studio Floor On the Copilot Studio floor, topics used to be rigid conversation trees β€” now generative orchestration lets the agent pick its own path through topics, tools, and knowledge at runtime. Tools reach out through Power Automate and custom connectors to take real action. Knowledge sources ground every answer in SharePoint, Dataverse, and uploaded files. And when there is no API, computer use lets the agent drive a virtual desktop like a person would. **Exam Intel**Generative orchestration = dynamic topic/tool/knowledge selection, replaces fixed triggers. Tools = Power Automate + connectors + APIs. Computer use = agent operates a virtual desktop for API-less apps. Declarative vs autonomous agent types. 🏭 ### The Foundry Assembly Line Down in Foundry, pro-code agents are assembled from a model catalogue spanning Azure OpenAI, Meta, Mistral, and open-weight models. Connected agents let one agent call another as a tool, snapping specialist capability together like modules. Every run is traced and evaluated for quality and safety before it ships β€” and Foundry speaks both MCP and A2A natively, so it never builds in isolation. **Exam Intel**Foundry Agent Service = code-first hosting. Model catalogue = Azure OpenAI + partner/open-weight models. Connected agents = agent-as-tool pattern. Built-in evaluation/tracing = observability. Native MCP + A2A support. πŸ”Œ ### The MCP Switchboard The Model Context Protocol switchboard is where tools get exposed once and consumed everywhere. An MCP server wraps a CRM, a ticketing system, a database β€” offering it up as Tools to call, Resources to read, and Prompts to reuse. Any MCP-aware agent, whatever platform built it, plugs in through the same interface. Security review checks exactly what each server can touch. **Exam Intel**MCP = open standard for agent-to-tool integration. Primitives: Tools (callable functions), Resources (readable data), Prompts (templates). One server, many consuming agents. Review: auth, scope, exfiltration risk. πŸ”— ### The A2A Diplomatic Corps Where MCP connects an agent to a tool, A2A connects an agent to another agent β€” even one built by a different vendor. Each agent carries an Agent Card describing what it can do, so others can discover and negotiate before delegating a task. An orchestrator decomposes the request and hands sub-tasks to specialists, with explicit fallback if a hand-off times out. **Exam Intel**A2A = open protocol for cross-vendor agent-to-agent communication. Agent Card = capability discovery. MCP = agent-to-tool; A2A = agent-to-agent β€” complementary, not competing. Orchestrator-worker is the dominant pattern. πŸ“Š ### The Dataverse Vault Underneath every agent sits Dataverse β€” tables, relationships, security roles, business rules, the same data layer Dynamics 365 runs on. Power Fx expressions drive the low-code logic inside topics and apps. Security roles and column-level security mean an autonomous agent can only ever see and change what its role permits β€” no exceptions for being "just an agent". **Exam Intel**Dataverse = common data layer under Copilot Studio, Power Platform, Dynamics 365. Power Fx = low-code formula language for topic/app logic. Security roles + column-level security constrain agent data access. Environment variables keep solutions portable. πŸš€ ### The ALM Runway An agent solution follows the same runway as any Power Platform solution: build unmanaged in dev, promote managed through test, land in a production managed environment with sharing limits and enforced DLP. Solution layering keeps every change traceable and reversible. Foundry agents run a parallel runway β€” source-controlled definitions, CI/CD, versioned prompts and models. **Exam Intel**ALM: unmanaged dev β†’ managed solution β†’ test β†’ prod via pipelines. Managed environments = governance tier (sharing limits, weekly digest, mandatory DLP). Version rollback supported. Foundry ALM = source control + CI/CD for agent definitions. πŸ›‘οΈ ### The Security Checkpoint Before an agent goes live, it passes the security checkpoint. Prompt Shields catch both a user typing something malicious and a poisoned instruction hidden inside a retrieved document. DLP policies stop data crossing from a business connector to a non-business one. Every tool call runs on least privilege, and any high-risk action β€” a payment, a deletion β€” waits for a human to approve it. **Exam Intel**Prompt Shields: direct injection (user input) + indirect/cross-domain injection (poisoned retrieved content). DLP: business/non-business/blocked connector classification. Least privilege per tool. Human-in-the-loop for high-risk autonomous actions. Tuning data is an attack surface too. πŸ“‘ ### The Telemetry Deck Once live, the telemetry deck lights up. Copilot Studio analytics track resolution rate and escalation rate; Foundry observability traces every model and tool call in a run; Application Insights covers latency and errors for embedded agents. Microsoft Purview logs exactly what data an agent touched and when β€” closing the loop back to the KPIs set at the strategy table. **Exam Intel**Copilot Studio analytics: session volume, resolution/escalation rate, CSAT. Foundry observability: traces model + tool calls. App Insights: latency, errors, throughput. Purview: audit trail of agent data access by identity. βš–οΈ ### The Responsible AI Watch The watch never ends. Content Safety screens every input and output against configurable severity thresholds. Groundedness detection flags a response the retrieved data doesn't actually support. Model and agent cards document intended use and limits for anyone inheriting the solution. And if something goes wrong in production, the agent can be paused or rolled back immediately. **Exam Intel**Azure AI Content Safety: input/output screening, severity thresholds. Groundedness detection: flags unsupported/hallucinated responses. Transparency/agent cards document use + limits. Incident response: pause/disable/rollback capability required. Section 3 / Acronym Memory ## Mnemonic Wall Memorable acronyms and phrases to anchor key exam concepts in your memory. πŸ—ΊοΈ PDD **P**lan, **D**esign, **D**eploy The 3 exam domains, in weight order roughly matching build order. Deploy is the heaviest at 40-45%. πŸŽ›οΈ TKTC **T**opics, **K**nowledge, **T**ools, **C**onnectors The four building blocks of a Copilot Studio agent. Generative orchestration picks between them dynamically. πŸ”Œ TRP **T**ools, **R**esources, **P**rompts The three MCP primitives an MCP server can expose. Tools = call, Resources = read, Prompts = reuse. πŸ”— DCT **D**iscovery (Agent Card), **C**apability negotiation, **T**ask delegation The A2A protocol flow between two independently built agents. πŸ›‘οΈ PADA **P**rompt shields, **A**ccess control, **D**ata loss prevention, **A**udit trails The four pillars of agent security tested across the Deploy domain. πŸš€ BTPM **B**uild, **T**est, **P**ublish, **M**onitor The agent lifecycle (ALM) stages β€” mirrors standard Power Platform ALM but adds continuous monitoring for agentic solutions. Section 4 / Contrast Memory ## Versus Arena Side-by-side comparisons to sharpen your understanding of similar concepts. vs A2A ProtocolvsMCP Click to compare #### Agent2Agent vs Model Context Protocol | Aspect | A2A Protocol | MCP | | --- | --- | --- | | Connects | Agent to agent | Agent to tool/data/prompt | | Purpose | Discovery, negotiation, task delegation | Standardised tool and context exposure | | Key artefact | Agent Card (capabilities) | MCP server (tools, resources, prompts) | | Cross-vendor | Yes β€” designed for multi-vendor agents | Yes β€” one server, any MCP-aware client | | Use case | Specialist agents collaborating on a task | Agent reaching a CRM, database, or API | | Relationship | Complementary to MCP | Complementary to A2A | Click to flip back vs Copilot Studio AgentsvsFoundry Agents Click to compare #### Copilot Studio vs Microsoft Foundry | Aspect | Copilot Studio Agents | Foundry Agents | | --- | --- | --- | | Build style | Low-code, conversation-first | Pro-code, model-first | | Orchestration | Generative orchestration over topics | Custom orchestration logic in code | | Model choice | Managed, curated for the platform | Full model catalogue (Azure OpenAI + partners) | | Extensibility | Connectors, Power Automate, computer use | Connected agents, custom tool code | | Best for | Business users, citizen developers, fast build | Complex orchestration, engineering teams | | Interop | Speaks MCP and A2A | Speaks MCP and A2A | Click to flip back vs Single-AgentvsMulti-Agent Orchestration Click to compare #### Single-Agent vs Multi-Agent | Aspect | Single-Agent | Multi-Agent Orchestration | | --- | --- | --- | | Scope | One bounded task, one owner | Workflow spanning domains or systems | | Complexity | Lower β€” simpler to secure and monitor | Higher β€” needs hand-off and failure design | | Pattern | N/A | Orchestrator-worker, peer-to-peer, hierarchical | | Protocol | MCP for tools | MCP for tools + A2A for agent hand-off | | Failure mode | Contained to one agent | Must design explicit timeout/fallback per hand-off | | Best for | Well-defined, high-volume task | Cross-functional processes needing specialists | Click to flip back vs Agentic-FirstvsClassic RPA/Automate Click to compare #### Agentic-First Solutions vs Classic RPA | Aspect | Agentic-First | Classic RPA/Automate | | --- | --- | --- | | Logic | Reasoning + judgement over unstructured input | Fixed, deterministic step sequence | | Adaptability | Handles novel/variable scenarios | Breaks on unexpected input or UI change | | Grounding | Retrieval + model reasoning | Hard-coded rules and selectors | | Governance need | Higher β€” prompt injection, groundedness, audit | Lower β€” deterministic and predictable | | Best for | Judgement-plus-rules, conversational tasks | Pure repetitive, rules-only tasks | | Often combined | Calls RPA/flows as a tool | Triggered by an agent as one step | Click to flip back vs Declarative (Classic)vsAutonomous Agents Click to compare #### Copilot Studio Agent Types | Aspect | Declarative (Classic) | Autonomous Agents | | --- | --- | --- | | Trigger | User conversation | Event, schedule, or system trigger | | Interaction | Turn-based dialogue | Runs unattended, reports outcomes | | Design focus | Topics, generative orchestration, knowledge | Triggers, tools, approval gates | | Human role | In the loop for every turn | Approves only flagged high-risk actions | | Best for | Customer/employee support, Q&A | Background monitoring, proactive workflows | | Governance | Standard content safety | Stricter β€” needs explicit approval gates | Click to flip back Section 5 / Grouping Memory ## Cheat Sheet Organised reference grouped by exam domain β€” everything you need on one page. ### Plan AI-Powered Business Solutions 25-30% #### Business Case & ROI - Identify agentic-ready processes: high volume, judgement-plus-rules, measurable outcome - ROI = deflection rate + time-to-resolution + cost savings vs build/model cost - Adoption metrics validate value post-launch: active users, completion rate, escalation rate - Executive sponsorship and change management are planning deliverables, not afterthoughts #### Solution Architecture Planning - Platform fit: Copilot Studio (low-code, conversational) vs Foundry Agent Service (pro-code, flexible models) - Single-agent for bounded tasks; multi-agent for cross-domain workflows - Topology choice: orchestrator-worker, peer-to-peer, hierarchical hand-off - Plan for latency, cost-per-call, and tool/model unavailability fallback #### Data & Grounding Strategy - Select knowledge sources: Dataverse, SharePoint, files, public web - Assess data classification (business/non-business/blocked) before connector approval - Determine data residency requirements β€” drives region and model deployment choice - Plan reuse of existing Power Platform connectors and Dataverse tables #### Governance & Risk Planning - Pre-build risk assessment: data touched, autonomous actions, human approval points - Microsoft Entra Agent ID assigns identity for access and lifecycle governance - Define ownership: who can publish, own, and retire each agent (prevents agent sprawl) - Map responsible AI requirements to the specific use case risk level ### Design AI-Powered Business Solutions 25-30% #### Copilot Studio Agent Design - Topics = trigger-based flows; generative orchestration dynamically selects topics/tools/knowledge - Tools connect to Power Automate flows, custom connectors, and published APIs - Knowledge sources ground responses and reduce hallucination - Computer use drives a virtual desktop/browser for apps with no API - Choose declarative (conversational) vs autonomous (trigger/schedule-driven) agent type #### Microsoft Foundry & Models - Foundry Agent Service: code-first hosting with full orchestration control - Model catalogue: Azure OpenAI plus Meta, Mistral, DeepSeek, and open-weight models - Connected agents: one agent calls another as a tool - Built-in evaluation and tracing score quality, safety, and groundedness #### Multi-Agent Orchestration - A2A protocol: Agent Card discovery, capability negotiation, task delegation - Orchestrator-worker is the dominant multi-agent pattern - Design explicit hand-off points with timeout and fallback behaviour - MCP = agent-to-tool; A2A = agent-to-agent β€” use both together #### Extensibility with MCP - MCP primitives: Tools (call), Resources (read), Prompts (reusable templates) - One MCP server, consumed by any MCP-aware agent across platforms - Security review: authentication, tool scope, exfiltration/unintended-action risk - Prefer MCP over bespoke per-agent connector code for reusability #### Data Architecture - Dataverse: common data layer (tables, relationships, security roles, business rules) - Power Fx: low-code formulas inside topics and apps - Security roles + column-level security constrain agent data access - Environment variables and connection references keep solutions portable ### Deploy AI-Powered Business Solutions 40-45% #### Environment Strategy & ALM - ALM path: unmanaged dev β†’ managed solution β†’ test β†’ production via pipelines - Managed environments enforce sharing limits, weekly digest, and mandatory DLP - Solution layering keeps changes traceable and reversible across environments - Foundry ALM: source-controlled agent definitions, CI/CD, versioned prompts/models #### Agent Lifecycle Management - Publish, version, and roll back agents without redeploying the whole solution - Channel management controls where a published agent is available (Teams, web, voice) - Staged rollout patterns limit blast radius of a new agent version - Retirement/ownership transfer process prevents orphaned or ungoverned agents #### Security & Prompt Injection Defence - Prompt Shields: direct injection (user input) + indirect/cross-domain injection (poisoned documents) - DLP policies classify connectors business/non-business/blocked to stop data crossing boundaries - Least-privilege scoping on every tool and connector an agent can call - Human-in-the-loop approval gates for high-risk autonomous actions - Secure and audit model tuning data β€” a poisoned tuning set is an attack vector #### Telemetry & Monitoring - Copilot Studio analytics: session volume, resolution/escalation rate, CSAT - Foundry observability: traces every model call, tool call, and hand-off in a run - Application Insights: latency, error rate, throughput for embedded agents - Monitoring KPIs should trace back directly to the ROI plan set in the Plan domain #### Audit Trails & Responsible AI Ops - Microsoft Purview logs agent data access and actions by identity for compliance review - Azure AI Content Safety screens input/output with configurable severity thresholds - Groundedness detection flags responses unsupported by retrieved data - Incident response requires the ability to pause, disable, or roll back an agent immediately Section 6 / Method of Loci ## The Memory Palace Walk through themed rooms β€” each object anchors a concept in spatial memory. ### The Planning War Room Plan domain β€” Where the business case and architecture are decided 🧭 ROI Model Deflection rate + time-to-resolution + cost savings vs build/model cost 🧩 Platform Decision Copilot Studio (low-code) vs Foundry Agent Service (pro-code) πŸ•ΈοΈ Agent Topology Single agent, or orchestrator-worker / peer / hierarchical multi-agent πŸ” Risk Assessment Data touched, autonomous actions, required human approval points πŸͺͺ Entra Agent ID Assigns each agent an identity for access control and lifecycle governance ### The Copilot Studio Workshop Design domain β€” Where low-code agents are built πŸ—‚οΈ Topics + Generative Orchestration Dynamic topic/tool/knowledge selection replaces fixed conversation trees πŸ› οΈ Tools Power Automate flows, custom connectors, published APIs πŸ“š Knowledge Sources SharePoint, Dataverse, files, public web β€” grounds responses πŸ–₯️ Computer Use Agent drives a virtual desktop/browser where no API exists ⏱️ Autonomous Agents Trigger/schedule-driven, run unattended with approval gates ### The Foundry & Protocol Lab Design domain β€” Where pro-code agents and open standards meet 🏭 Foundry Agent Service Code-first hosting, model catalogue, connected agents, tracing/evaluation πŸ”Œ MCP Server Exposes Tools, Resources, Prompts to any MCP-aware agent πŸ”— A2A Agent Card Capability discovery + negotiation between independently built agents πŸ“Š Dataverse + Power Fx Common data layer and low-code logic underneath every agent ### The Deployment Command Centre Deploy domain β€” Where agents go live and stay governed πŸš€ Environment Pipeline Unmanaged dev β†’ managed solution β†’ test β†’ production, managed environments πŸ›‘οΈ Prompt Shields + DLP Direct/indirect injection defence, business/non-business connector controls πŸ“‘ Telemetry Dashboards Copilot Studio analytics, Foundry observability, Application Insights πŸ“œ Purview Audit Trail Logs every agent data access and action by identity βš–οΈ Content Safety + Groundedness Continuous screening and hallucination detection, with rollback capability Section 7 / Pattern Recognition ## Pattern Spotter Decision trees and trigger-answer pairs β€” see the pattern, know the answer. Which Agent Platform? Which Agent Platform? Β Β β”œβ”€β”€ Conversational, topic-driven, business-user built β†’ Copilot StudioΒ Β β”œβ”€β”€ Code-first, complex orchestration, flexible model choice β†’ Microsoft Foundry Agent ServiceΒ Β β”œβ”€β”€ Needs to reach an external tool/data source via open standard β†’ Add an MCP server  └── Needs to collaborate with an agent on another platform/vendor β†’ Add A2A protocol support Which Orchestration Pattern? Which Orchestration Pattern? Β Β β”œβ”€β”€ Single, well-defined task with one owner β†’ Single agentΒ Β β”œβ”€β”€ Complex, multi-step process across domains β†’ Multi-agent orchestrator-workerΒ Β β”œβ”€β”€ Specialist agents on different platforms must cooperate β†’ A2A protocol hand-off  └── Agent needs arbitrary external tool/data access β†’ MCP tool integration Which Environment for This Change? Which Environment for This Change? Β Β β”œβ”€β”€ Building or experimenting β†’ Dev environment (unmanaged solution)Β Β β”œβ”€β”€ Validating before release β†’ Test/UAT (managed environment)Β Β β”œβ”€β”€ Customer- or employee-facing at scale β†’ Production managed environment + enforced DLP  └── Regression found post-release β†’ Roll back to previous published version Which Security Control? Which Security Control? Β Β β”œβ”€β”€ Untrusted user input or retrieved content β†’ Prompt Shields (direct + indirect injection)Β Β β”œβ”€β”€ Risk of data crossing a business boundary β†’ DLP policy on the connectorΒ Β β”œβ”€β”€ Need traceability for compliance β†’ Microsoft Purview audit logging  └── High-risk autonomous action (payment, deletion) β†’ Human-in-the-loop approval gate ## Decision Cards "computer use"β†’Copilot Studio agents operating a virtual desktop/browser for API-less apps "Agent2Agent" or "A2A protocol"β†’Open standard for cross-vendor agent-to-agent discovery and delegation "Model Context Protocol" or "MCP server"β†’Standardised agent-to-tool/data/prompt exposure via Tools, Resources, Prompts "generative orchestration"β†’Copilot Studio's dynamic, reasoning-based topic/tool selection at runtime "Foundry Agent Service"β†’Pro-code agent hosting with full model catalogue and connected agents "prompt shields" or "prompt injection"β†’Defence against direct (user) and indirect (poisoned content) injection "managed environment"β†’Power Platform governance tier: sharing limits, weekly digest, mandatory DLP "DLP policy" or "data loss prevention"β†’Connector classification: business, non-business, blocked "audit trail" or "Purview"β†’Compliance logging of agent data access and actions by identity "human-in-the-loop"β†’Approval gate before an agent takes a high-risk autonomous action Ready to certify? ## Train with practitioners, not presenters Lucid Labs delivers Microsoft certification training led by Microsoft Certified Trainers (MCTs) and grounded in real-world project experience. We adapt every session to your team's environment, data stack, and business objectives β€” because the best exam prep comes from engineers who build these solutions every day. 🎯 Tailored Content Training built around your actual agent architecture β€” Copilot Studio, Foundry, MCP, A2A β€” not generic slides. πŸ› οΈ Hands-On Labs Design multi-agent orchestration, wire up MCP servers, and harden agent security with expert guidance. πŸ“ˆ Exam + Capability Pass the exam and build lasting agentic AI architecture skills your team can apply from day one. [Talk to us about Agentic AI Business Solutions Architect (AB-100) training](https://lucidlabs.com.au/?service=training-consulting&message=I%27m%20interested%20in%20Agentic%20AI%20%2F%20Agent%20Architecture%20\(AB-100\)%20training%20for%20my%20team.#contact) Custom training for teams & individuals β€” remote or on-site across Australia ![Keith Oak](https://lucidlabs.com.au/team/koak-400.jpg) Keith Oak Chief Technology Officer β€” Lucid Labs Microsoft Solutions Partner architect specialising in Fabric, Azure Data & AI, and GitHub Enterprise. 18+ years delivering data platforms for Australian businesses β€” building the systems these exams test every day. [LinkedIn β†—](https://www.linkedin.com/in/keithoak/)[lucidlabs.com.au β†—](https://lucidlabs.com.au/)Published 29-03-2026