AIBS / Power Platform + HVE / Practical guide
Use HVE Squad to decide which Power Platform features a client needs and what schema supports them. Explain the evidence, alternatives and first-release tradeoffs before implementation.
Practical Guide, source slide 1. The 25-slide source narrative is retained with current-contract corrections and 13 course bridges. Arrow keys navigate; printing includes all 38 slides.
Choose who owns the loop
Source 2. Lab 01: select the ownership model around the first working service.
The starting estate
Source 3. Exercise 01 maps these handoffs before proposing another agent.
From a human hub to a platform
One natural-language entry point in the flow of work.
Bounded assistance and orchestration across the process.
Authorised data and business operations remain anchored at their source.
Source 4. Preserve the system of record while reducing the manual joins.
The false shortcut
Source 5. Lab 01: build an owned service, not another disconnected demo.
The missing connection
Source 6. Its headline percentages are not independently established here and are not repeated as workshop evidence. Use the client's measured baseline.
Where organisations get stuck
Source 7. The curve is qualitative; autonomy is not a compulsory destination.
What it takes
Find today's evidence and the outcome that matters.
Give process experts a governed way to make changes.
Run, observe, correct and repeat.
Source 8. The exercises move from discovery to controls to independent reuse.
Who can build
Source 9. Lab 03 challenges the process diagnosis; lab 06 retains human-owned agent decisions.
Components of an agentic business process
Natural-language experience, shared context and work inside Microsoft 365 apps.
Structured work, role-specific experiences and human oversight at key decisions.
Assistance grounded in authorised business context, including Work IQ where supported, paired with deterministic, traceable actions.
Source 10. Lab 02: extraction, deterministic validation and human acceptance are distinct.
The operating stack
Managed Environments, data policies, environment groups, solutions, pipelines, evaluations and telemetry. Prove that a configured constraint held.
Nine stages, four principles, a named decision owner and maximum decision wait. Decide what must be true and when to stop.
Sixteen roles dispatched into customer-owned project state. Carry the applicable constraint into each request and require acknowledgment.
Source 11. Shared delivery workflow: advisory work, actual implementation and client acceptance have separate evidence.
Frontier CoE framework
Source 12. Exercise 05 records a blocking dependency, not just an average readiness score.
The operating model
Source 13. Lab 01 launches the operating loop around an employee service.
What the CoE actually ships
Source 14. Every lab carries constraints into design and tests; lab 10 verifies independent second-client delivery.
The review lens
A working agent, flow or app in a governed environment each loop.
Fails as: a loop ending in a status deck.
One named business decider, a written maximum wait and escalation.
Fails as: business involvement only at steering meetings.
Investigate real estate, permissions and documents before designing.
Fails as: assuming what is running or published.
The receiving owner runs the harness on their own machine.
Fails as: a wiki and a client still calling you.
The source uses four equal columns because all four are reviewed together. A strong demo cannot compensate for absent ownership.
Source 15. Unticked means the next conversation, not a box to fill with invented evidence.
The instrument
Decision owner, maximum decision wait, next working demo date and an outcome stated as something the client can run, deploy or test.
One observable action, one named person, one due date inside 48 hours and agreed evidence of done.
Source 16. Use the lab receipt. Record stage evidence, not fictitious completion of nine stages in one session.
The source's concrete implementation
Selected project, human handoffs and faithful output persistence
The server Coordinator owns routing, workers and stage order.
squad_status / squad_historysquad_respond is not operator squad_approve.
hve-project.json canonical server-file projection stakeholder libraryActual provider/drive/folder identity. No parallel legacy scaffold.Source 17, corrected against the locally supplied current Cowork plugin contract, October 2026. Follow the shared workflow; verify the exact installed release before teaching it.
Who makes it true
coordinator, federation-coordinator
Classify, resolve, dispatch, enforce gatesintake-validator, analyst, researcher, designer
Turn evidence into testable requirementsexperimenter, data-scientist, architect, lead
Challenge shape and sequencedeveloper, tester, product-owner
Build, prove quality, maintain the backlogscribe, technical-writer, presenter
One state writer, durable memory, business narrativeSource 18 role vocabulary. The current Cowork bridge does not choose these roles; the server Coordinator owns dispatch. Advisory role labels do not prove code execution.
Stated limits, not promised roadmap
squad_run holds. A separately authorised human releases the gate.
The described tools return text deliverables. Tenant deployment is a separate, authorised operation.
The customer folder is authoritative. A copied folder gets a new project ID, not a silent merge.
A disabled tool is blocked. Never record output from a tool that did not run.
Source 19. All labs separate advisory output, human decisions and verified implementation.
What HVE should help you specify and review
Ask HVE which business decision, baseline and owner make a first slice worth developing.
Supply relevant records and ask for facts, assumptions, missing context and verified capability claims.
Challenge alternatives, scope and constraints before accepting component choices.
Ask what a bounded experiment could establish, and what would disprove the recommendation.
Obtain exact behavior, data/action contracts, ordered tasks and dependencies. A separate implementer executes them.
Bring feedback against accepted tests. Separate supplied scenarios, actual observations and unexecuted recommendations.
Ask HVE to reason from failure evidence, preserve uncertainty and define the next safe change.
Let another person find accepted project evidence and request the next bounded development task.
Extract a named question set, contract or test pattern. Replace client facts and revalidate assumptions.
The full HVE method includes implementation. This course focuses on upstream feature/schema decisions; coding capability is not the core task. Five teaching requests do not certify all nine stages or prove a platform operation.
Source 20, with capability caveats. Inventory scope · Agent evaluation. Each lab receipt records observed, failed, blocked and not-run separately.
What makers actually build
Pixel-level design and Power Fx.
Quarantine supported. Per-app Conditional Access requires Managed Environments.Data-first UX from the Dataverse model.
Those specific quarantine/per-app controls are unsupported, not all security controls.Code-first web apps in a platform host.
Quarantine/per-app controls supported. Native Power Platform Git integration is not; ordinary Git is a separate route.Plans, data and apps on one surface.
Export can be edited externally; PAC redeployment creates a new app disconnected from the original plan.Connector-based automation. Agent use needs the appropriate trigger, response, same-environment solution and supported execution pattern.
A subordinate capability, not a third peer flow type. No GA date is shown in the checked documentation.
Attended or unattended desktop automation. Hosted RPA uses Microsoft-hosted infrastructure; not every desktop flow runs on a hosted Azure VM.
Record machine, identity, capacity and attended/unattended requirements separately from a cloud-flow action.
Source 21 with product-specific caveats. Recheck tenant/release support; preview features are not required by this course. Code apps · Canvas lab 04 · Code-app lab 08.
Building a Copilot Studio agent, the HVE way
Answer one real question in a governed environment this loop.
Not only a topic map or demo script.Name the person who accepts the agent's actual answers.
Not only a steering summary.Use sources that exist under the permissions that actually apply.
Not what somebody believes is published.The receiving owner reruns the set and ships a small change while you watch.
Not a handover deck.May it answer beyond the approved sources?
Which sources and actions may it reach, as which identity?
What may run without review, and who may interact with it?
What tests define acceptance, and who reruns them?
The upper row describes delivery behaviour; the lower row defines retained human authority. They must not collapse into one agent instruction.
Source 22. Lab 06 tests the human decisions and action boundary. Lab 10 proves independent adaptation.
Success criteria
Six months later, the second engagement took the same effort as the first, and nothing from the first was reused.
One disconfirming sentence sits above two evidence columns. Test the delivery method with the same discipline used to test the product.
Source 23, paraphrased falsification test. Lab 09 separates suspected value from financial evidence; lab 10 measures actual adaptation and reuse.
Put it in the calendar
Confirm delivery shape and owner. Create project state, inventory registered and unregistered assets, record baseline and verify applicable attribution.
Exit: sourced inventory + field sheetStage maturity, assign authority to constraints, verify environment groups and data policies, prepare the governed pilot.
Exit: owned controls + pilot evidenceTest and promote only with evidence. Verify each agent's owner and applicable identity; watch the second solution being built.
Exit: measured pilot + acceptance recordRun the harness on a cadence, measure usage, reassess maturity and reuse assets across teams.
Exit: client-run change + dated reuseSource 24 with scope corrections. PAL applies to eligible Power Platform resources, not a universal day-one release gate. New Copilot Studio agents use Agent IDs; legacy agents can use app registrations and Agent Builder differs. Name an accountable owner in every case.
Course bridge / HVE-DECISIONS-1.0
Canvas offline/photos/checklists, agent knowledge/actions, portal visibility and BI lineage imply different feature and data needs.
Inspect alternatives, challenge unsupported scope and explain every entity/field through a selected feature.
Development cases are fictional teaching inputs, not your HVE outputs or live platform tests.
The five main learner requests
Use interviews and examples. Separate confirmed need, future request and assumption.
01 Understand the capabilityChoose first-release features, then compare the data structures they need.
04 Compare inspection modelsExplain how a new feature changes fields, relationships, ownership/history and scope.
08 Add skills or reassignment?Walk records, review alternatives and let another person explain the choices.
07 Model safe operation10 Decide variation50 main prompts across ten scenarios. They are not a manual specialist dispatch sequence; the server owns routing.
No implementation prerequisite
Attach the guide, policy, examples and HVE-DECISIONS-1.0 case. Do not attach a starter schema as the answer.
Decide included/deferred/rejected/unverified with a reason.
Ask what every field/relationship serves and what a smaller option loses.
Use design feedback honestly. Let a colleague explain what is needed and why.
HVE orientation · Cowork working method. Live connection rehearsal and learner timing remain separate gates.
HVE skill 01 / Capability and service scope
Choose minimum CoE capabilities and first-service features, then the data linking use cases, decisions, service activity and assets.
One portfolio row versus relationships/history only where needed.
Central versus team-owned capability, first release and later variation.
AI CoE Launch + First Value Stream.
HVE skill 02 / Order features and data
Choose extraction, correction, exception, clarification and acceptance features.
Source/order/line/issue/review concepts against a simpler bounded pilot.
Multi-line scope, unknown quantity and preservation of original/corrected data.
Order Intake Accelerator.
HVE skill 03 / Insight and intervention choices
Choose insight, current-work, follow-up and trial-comparison features.
Historical events, current state and follow-up history have different meaning.
Rota versus reminder versus observation, and the minimum data needed to evaluate it.
Process-to-Value Sprint.
HVE skill 04 / Canvas feature and schema choices
Decide capture, offline, photo, checklist-version and remediation features from the interview.
Flat defect slots versus repeatable findings; simple question snapshots versus versioned templates.
First release and justified entities/fields/relationships. No preset table/control blueprint.
Field Quality Accelerator.
Canvas decision exercise. No formula debugging.
HVE skill 05 / Supplier identity and history
Choose self-service, correction, status and private-review features.
Contact-owned versus organization-associated applications and submission/review history.
Who sees what, which originals persist and whether delegation is needed now.
Supplier Onboarding Accelerator.
HVE skill 06 / Agent capability and data
Choose explanation, preparation, confirmation, submission and later status features.
Transient conversation state versus persistent request/identity/history needs.
What the agent may do, what it must remember and what it should not retain.
Transactional Service Agent Pilot.
HVE skill 07 / Operating features and data
Choose queue, reconciliation, assignment and attempt/history features.
Job intent, attempt outcome, target observation and resolution are different concepts.
Minimum data for accountable operating decisions and later reviewer/target variation.
Legacy Workflow Bridge.
Exercise 07. No code/debugging task.
HVE skill 08 / Workbench feature/model choices
Choose the comparisons, availability explanations and accountability a dispatcher needs.
Custom versus simpler app experience; single skill/assignment fields versus associations/history.
What multi-skill and reassignment require now versus later.
Dispatch Workbench Accelerator.
HVE skill 09 / Insight and casework data
Choose diagnostic, lineage/freshness and managed investigation features.
Work versus line grain; source facts, measures, dispositions and recovery evidence.
Case identity/history for partial invoices and repeated observations.
Revenue Leakage Diagnostic + Resolution.
Exercise 09. No report-authoring prerequisite.
HVE skill 10 / Standard and client variation
Separate standard features from Nora-specific needs and inherited assumptions.
Fixed values, managed references, proportionate configuration and true variants.
Region/reviewer schema consequences and what the new client must approve afresh.
Repeatable Accelerator Delivery.
From the deck to the next 48 hours
Source 25 closing, connected to the commitment and transfer tests in sources 16, 22 and 23.