Decision being made
Choose the operating boundary
Decide whether the experience should extend Microsoft 365 Copilot Chat, follow defined rules and paths, or reason across a longer sequence of tools and files.
Short answer
Start with the narrowest harness that fully supports the work
Use the Copilot chat harness for grounded internal answers inside Microsoft 365 Copilot Chat. Use the standard harness for defined, repeatable conversations, rules, and agent flows. Use the GitHub Copilot harness for reasoning-heavy work that spans multiple steps, tools, or files and needs to adapt as it proceeds.
Intended audience
Who this guide is for
- Business leaders comparing shared agent approaches.
- IT and AI decision-makers defining the acceptable operating boundary.
- Technical teams validating identity, channel, tool, control, and consumption requirements.
Relevant evidence
Describe the work before choosing the harness
- Identify who will use the experience and where it must appear.
- Separate knowledge retrieval from defined rules and adaptive reasoning.
- List the files, tools, channels, and actions the work actually needs.
- Identify every action requiring human review or approval.
- Confirm which user identity or trigger invokes the work.
- Estimate usage patterns before comparing billing choices.
Recommended capability
Choose by work pattern, not by feature count
| Harness | Use it first when | Operating boundary | Billing question |
|---|---|---|---|
| Copilot chat | Employees need grounded organizational answers in Microsoft 365 Copilot Chat | Internal Microsoft 365 Copilot Chat experience | Is the authenticated user covered by Microsoft 365 Copilot, or does consumption apply? |
| Standard | The experience follows defined topics, rules, paths, or structured agent flows | Internal or external shared experience with predictable behavior | Which features consume pooled Copilot Credits, and under whose identity or trigger? |
| GitHub Copilot | The work must reason through a goal, use several tools or files, and adapt across steps | Broader internal or external agentic work with a secure task sandbox | What consumes credits during creation, testing, evaluation, and use, and which controls apply? |
Why not the other solutions?
Broader capability creates a broader decision
Why not Copilot chat?
It is designed for grounded internal Copilot Chat experiences. It is not the first choice when the work must publish externally, create and edit files, or adapt across a complex tool sequence.
Why not standard?
It is strongest when the path is understood and predictable. It is not the first choice when the task must form and revise a plan, recover from problems, or coordinate several tools and files.
Why not GitHub Copilot?
Its reasoning, file, tool, skill, and memory capabilities widen cost and governance considerations. Do not choose it when grounded answers or defined rules already solve the problem.
When an alternative becomes appropriate
Change the choice when the work crosses a real boundary
- Move from Copilot chat to standard when a shared experience needs controlled paths, channels, or repeatable actions.
- Move from standard to GitHub Copilot when adaptive reasoning, longer tasks, file creation, skills, memory, or recovery are material requirements.
- Move back to a narrower harness when the broader capability does not produce evidence that justifies its cost, access, and oversight boundary.
- Compare Microsoft Foundry instead when the need is a custom AI application with model, code, infrastructure, evaluation, networking, and engineering ownership.
Risks and guardrails
Licensing inclusion is conditional
Copilot Credits are pooled across the tenant, and consumption varies with agent design and use. Microsoft 365 Copilot can include qualifying employee-facing activity, but the boundary depends on the authenticated licensed user, trigger, channel, and capability. Computer-using agents and some other activity are not included.
Budget alerts, limits, capacity allocation, and enforcement are separate controls. Confirm current behavior before relying on a limit to stop spending or preserve availability.
Compare the purchasing path separately
Prepaid Copilot Credit packs
Use tenant capacity when monthly demand is predictable enough to plan a pool. Purchased capacity is shared across the tenant, enforcement applies, and unused monthly capacity does not carry forward.
Pay-as-you-go
Use an Azure-linked meter when actual usage should be billed after the month or when demand is less predictable. Treat budgets, alerts, and limits as separate controls rather than assuming the meter is a spending cap.
Prepurchase plan
Compare a one-year commit option when eligible annual usage is predictable across Copilot Studio or other covered Microsoft agent services. Confirm the current unit name, eligible services, term, scope, and agreement before choosing it.
Microsoft documentation currently describes prepaid packs, pay-as-you-go, a Copilot Credits prepurchase option, and the broader Microsoft Agent prepurchase plan. These labels and eligibility boundaries can change. Ask Microsoft or your licensing provider which purchasing path and unit names apply to your agreement.
Questions to discuss internally
Requirements to confirm
- Is the audience internal, external, or both?
- Does the work retrieve knowledge, follow rules, or adapt a plan?
- Which identities, information, tools, and actions are required?
- Which decisions require human review and an audit trail?
- Is expected usage predictable enough for prepaid capacity or a one-year commitment, or is pay-as-you-go the clearer comparison?
- Which eligible services, billing scope, agreement terms, alerts, and limits should the responsible team validate?
- What evidence would justify moving to a broader harness?
Public Microsoft sources
Verify the current harness and billing boundaries
- Harnesses in Copilot StudioWork patterns, features, publishing, and harness comparison
- Standard harness licensingAccess paths, Copilot Credits, pay-as-you-go, prepurchase, and Microsoft 365 Copilot inclusion
- Billing rates and managementFeature rates, pooled capacity, inclusion conditions, and enforcement
- GitHub Copilot harness cost controlsCreation and runtime consumption, allocation, limits, budgets, and alerts
- Microsoft Agent prepurchase planEligible services, one-year term, commit units, purchasing scope, and agreement prerequisites
Verification date
Check the current terms before deciding
Source review completed . Availability, inclusion, consumption, and enforcement can vary by tenant, region, agreement, identity, trigger, and capability.
Continue the decision
Resolve the next product boundary
Finished this primer?Optional progress stores only this primer identifier and a completion time on this device.