Copilot Studio vs Agentforce: a data-gravity decision
Microsoft Copilot Studio and Salesforce Agentforce get compared as if this were a capability shootout. It almost never is. The suite that already holds your system of record, your identity and your governance estate usually wins — and the licensing model decides the bill.
- Neutral comparison
- Updated August 2026
- Vendor-published figures
Summary
Key takeaway: Microsoft Copilot Studio is a low-code studio built on Power Platform that grounds agents in Microsoft 365 data and publishes them into Teams, SharePoint and voice, metered in Copilot Credits. Salesforce Agentforce builds agents in Agentforce Builder on CRM records and Data 360, metered in Flex Credits at 20 credits per action or $2 per conversation. As of August 2026 each is strongest inside its own estate, so the choice is usually settled by which suite holds your system of record and identity — then by which licensing shape fits your volume.
If you are comparing Microsoft Copilot Studio and Salesforce Agentforce feature by feature, you are probably solving the wrong problem. Both are suite-native agent layers, and suite-native products are bought for adjacency, not for the feature list.
The honest core of this decision is data gravity and licensing. An agent is only as useful as the records it can read and write, and moving records between estates is the expensive part of any agent project. Whichever suite already holds the system of record for the work you are automating starts months ahead — not because its builder is better, but because the integration, the permissions model and the audit trail already exist. Your directory, your single sign-on and the people who administer them do not move for an agent project either.
Concretely, as of August 2026: Microsoft documents Copilot Studio as a graphical, low-code studio for building and managing AI-powered agents and workflows, built on Power Platform, grounded in Microsoft Graph, SharePoint and Dataverse, published into Teams, Microsoft 365 Copilot, websites, mobile and voice channels, and metered in Copilot Credits. Salesforce builds agents in Agentforce Builder on top of CRM records and Data 360, turning existing Flows, Apex, prompt templates and MuleSoft APIs into agent actions, reasoning through its Atlas engine, and metering in Flex Credits — 20 credits per action, $500 per 100,000 credits — or a flat $2 per conversation.
This page compares the two the way a buyer mid-evaluation needs them compared: two mental models, a dimension-by-dimension table, the case for and against each, how to price two different licensing shapes against the same workload, and a verdict that includes the third answer nobody selling you a suite will mention. Every figure here was read from the vendors’ own pricing and documentation pages in August 2026, and both ship frequently — verify before you commit.
Where we stand
AI Agentics makes an AI agent platform, so we have a horse in this race. We have tried to describe every option here as its own team would — including the cases where another platform is the better buy. Vendors ship fast and pricing changes; verify current details on each vendor’s own site before you commit.
An agent inside your work estate vs one inside your record estate
Stop thinking about builders and start thinking about which stack the agent is born inside.
The suite-native bet. Both products make the same argument, pointed at different estates: the agent should live where the data, the identity and the administrators already are, so nothing has to be moved, re-permissioned or separately audited. It is a strong argument. Most failed agent projects die on integration and governance, not on prompt quality, and a suite-native layer removes both — inside its own boundary. That last clause is the whole comparison.
Name the boundary and it stops being abstract. For Copilot Studio it is Entra ID, Microsoft Graph, SharePoint, Dataverse and the Power Platform connector catalogue, with Purview and Sentinel collecting the logs. For Agentforce it is CRM objects and sharing rules, Flows, Apex, knowledge articles and Data 360, with the Trust Layer and Audit Trail wrapped around them. Both lists are genuinely impressive. Neither extends past its own edge.
So the question is not “which builder is better?” It is “which boundary is my work already inside?” Answer that honestly and the shortlist usually collapses to one, or to neither. For the patterns underneath either choice, see agent orchestration.
Four forces settle most of these evaluations
Only one of them shows up on a feature comparison, which is why feature comparisons keep producing the wrong answer.
The records. List every system the agent must read from and write to, then mark which one is authoritative. If a single system answers “what is true right now” for most steps, the suite that owns it has largely won. If three systems share that role, no suite owns your work.
The identity estate. Agents act as someone, with that person’s permissions, and every action has to land in a log somebody already reads. Whoever runs your directory controls how cheap that is. It is also where independent platforms have to work hardest — see how we handle SSO, audit logs and SOC 2 for the shape of that conversation.
The builders. Suite-native tools are strongest where the suite’s administrators already are. Salesforce admins and Microsoft admins are not interchangeable, and the ones you have constrain what you can still maintain in eighteen months.
The licensing shape. Both meter consumption, in different units. Microsoft charges Copilot Credits per answer, per action and per minute of voice, sold as $200 monthly capacity packs of 25,000 credits, pay-as-you-go against an Azure subscription, or prepaid credit units. Salesforce charges 20 Flex Credits per action at $500 per 100,000 credits, or a flat $2 per conversation, with seat and flat-fee employee options alongside. Similar capability, very different invoices.
Copilot Credits per pack
Microsoft's published $200/pack/month capacity pack
Flex Credits per action
Salesforce: $500 per 100,000 credits, about $0.10 each
Gravity wells
records and identity — neither moves for one project
A quick test that usually decides it
Write down the ten steps of your highest-value agent workflow, and beside each step name the system it touches. If eight or more name the same suite, buy that suite’s agent layer and stop evaluating. If they are spread across four systems and two vendors, no suite-native option will feel native, and you should be comparing independent platforms instead.
Copilot Studio vs Agentforce across nine dimensions
Nine dimensions, taken from each vendor's own published pages in August 2026, plus the column that exists because a lot of work belongs to neither estate.
| Dimension | Microsoft Copilot Studio | Salesforce Agentforce | Independent platform |
|---|---|---|---|
| What it is | Low-code studio on Power Platform | Agent layer on the Salesforce platform | Standalone agent platform |
| Build experience | Natural language, topics, or a reasoning harness | Builder with doc, canvas and Agent Script views | Visual builder plus optional SDKs |
| Grounding data | Microsoft Graph, SharePoint, Dataverse | CRM objects, Data 360, Intelligent Context | 200+ integrations, SQL and REST |
| Governance surface | Entra ID, data policies, Purview, Sentinel | Trust Layer, Guardrails, Audit Trail | SSO/SAML, guardrails, audit logs, SOC 2 |
| Licensing shape | Copilot Credits consumption meter | Flex Credits, per conversation, or per user | Flat subscription tiers |
| Published figures | $200/pack/month = 25,000 credits | $500 per 100k credits; $2 per conversation | Free tier; Pro $49/mo |
| Free entry point | Free maker licence, after a paid tenant pack | Salesforce Foundations, a $0 add-on | 500 runs/month, no credit card |
| Deployment | Microsoft-hosted SaaS, no on-prem documented | Salesforce-hosted, no self-host option | Fully hosted SaaS, no self-hosting |
| Best fit | Gravity already sits Microsoft-side | Gravity already sits Salesforce-side | Gravity sits between the two, or neither |
How to read the pricing row
Neither published figure is a total. Microsoft publishes the $200 capacity pack and its per-action credit rates, but not the pay-as-you-go price of a single Copilot Credit — the Azure pricing page shows a dash and asks you to sign in or request a quote. Salesforce publishes $500 per 100,000 Flex Credits, then footnotes every worked example to say it excludes Data 360 credits and other consumption services. Both exclude the integration work for systems outside the chosen estate, which is usually the larger number.
How to compare two licensing models honestly
You cannot compare enterprise agent licensing at list price. You compare it by running your own workload through each meter.
Agent products are metered in three broad shapes, often in combination. Per-user seats track headcount: predictable, but you pay for people who never trigger an agent. Consumption units — conversations, messages, actions, credits — track usage: efficient designs cost less, but nothing is forecastable until you have run for a month. Bundled entitlements arrive inside a licence you already hold, which makes the marginal cost of a pilot genuinely zero until you cross a line nobody mentioned.
That third shape is what distorts evaluations, and both vendors sell one. On Microsoft’s published billing table, employee-facing usage by a Microsoft 365 Copilot licensed user is marked no charge across the main meters, within stated fair-usage limits — though Microsoft’s own pricing and licensing pages describe the exact scope of that inclusion slightly differently, so confirm it in the licensing guide before you model on it. Salesforce sells the same idea as a flat fee: $125 per user per month ($150 for industries) buys unmetered Agentforce usage for employees, while customer-facing agents stay on the credit or conversation meter. A bundled allowance makes the incumbent look free during a pilot and expensive at scale, and finding the crossover is the whole job.
Copilot Studio: one currency, three ways to buy
Copilot Credits replaced per-message billing on 1 September 2025. Buy them as $200/month capacity packs of 25,000 credits (a tenant licence, no per-user assignment), pay-as-you-go against an Azure subscription, or as a one-year pool of prepaid credit units. Published rates: 1 credit a classic answer, 2 a generative answer, 5 an agent action, 10 for tenant graph grounding, and 10 to 75 a minute for voice depending on tier. Unused credits do not carry over, and at 125% of prepaid capacity custom agents are disabled rather than throttled.
Agentforce: pick one meter per org
Flex Credits cost $500 per 100,000, and a standard action costs 20 credits — about $0.10 — while a voice action costs 30. The alternative is $2 per conversation, and Salesforce states the two cannot run in the same org. Seats sit alongside: $5 per user per month (which still requires Flex Credits), $125 or $150 per user per month for unmetered employee usage, and Agentforce 1 Editions from $550 per user per month including 2.5M Flex Credits per org per year. Credits do not roll over, but there is no overage penalty.
Define one real workload
Not a category — a specific agent. Users served, runs or conversations per month at steady state, and the peak month. Vague inputs produce vague quotes.
Read both meters as published today
Go to each vendor's current licensing page yourself. Note the unit, what counts as one, what your existing licences already include, and what happens on overage. The figures here were read in August 2026 and will drift; treat them as a starting shape, not a quote.
Model the same volume through both
Run steady-state and peak through each meter, then run both again at 3x, because successful agents grow. The shape of the curve matters more than the entry price.
Get the real number, then add what it excludes
Ask each account team for the figure that would appear on your invoice. Then add integration work for systems outside the chosen estate, the admin skills you must hire or retrain, and the second platform you will buy if this one cannot reach half your data.
What each builder actually is, and what it costs you
Both lists below come from each vendor's own published documentation and pricing pages, read in August 2026.
Copilot Studio
A browser studio built on Power Platform, with three building blocks: agents, workflows and agent flows. The structural choice inside it is the harness — Microsoft’s word for the runtime that decides when to call the model, what to send it and which tools to invoke. The standard harness matches a request to topics you design, with generative answers filling the gaps. The GitHub Copilot harness is a reasoning runtime that takes a goal, breaks it into steps and adjusts when one fails. A third harness extends Microsoft 365 Copilot Chat and publishes to internal teams only.
The case for
- Entra ID, Microsoft Graph, SharePoint, Dataverse and the Power Platform connectors are first-party, not integrations you build.
- If you hold Microsoft 365 Copilot licences, employee-facing agent usage is marked no charge across the main meters within fair-usage limits.
- Governance runs deep: Power Platform data policies, maker audit logs in Purview, agent activity in Sentinel, customer-managed keys.
- MCP servers connect through an onboarding wizard, and tool changes made server-side are reflected automatically.
- Unusually broad channels for a low-code tool — Teams, Microsoft 365 Copilot, SharePoint, web, mobile and voice, with published per-minute rates.
The case against
- The pay-as-you-go price of a single Copilot Credit is not published; the Azure pricing page shows a dash and asks for a quote.
- Getting started is tenant-level work: an Entra tenant plus a Power Platform environment with a Dataverse database.
- One interaction can hit several meters — Microsoft's own example spends 12 credits answering a single grounded prompt.
- Prepaid credits do not carry over, and at 125% of capacity custom agents are disabled rather than throttled.
- On the GitHub Copilot harness you are charged from the moment you start building, not from publish.
Agentforce
Agentforce Builder, and it is deliberately text-first. You describe the agent in plain language, the assistant generates subagents, instructions and actions, and you tighten them in whichever of three synchronised views suits you: a doc-like editor, a low-code canvas, or Agent Script, a human-readable JSON expression language with conditionals, loops, variables and hand-offs. Reasoning runs through the Atlas Reasoning Engine, whose hybrid mode is configurable so teams can trade model creativity for structured business logic. Every agent compiles to a portable JSON file.
The case for
- Existing Flows, Apex, prompt templates, APIs and knowledge articles become agent actions without a rebuild.
- Agent Script makes behaviour steerable by more than prompt text — conditionals, loops, variables and explicit hand-offs.
- Agents compile to a portable JSON file, so versioning and promotion between environments are real artefacts.
- The Trust Layer predates the agent product: dynamic grounding, zero data retention, PII masking, toxicity filtering, audit trail.
- Observability ships as a product — session-level tracing, clustering by intent, quality scores and per-agent cost.
The case against
- Only the $0 Foundations tier is self-serve; every paid tier routes through an account executive.
- Flex Credits and per-conversation pricing cannot run in the same org, so a mixed estate picks one or splits orgs.
- Unused Flex Credits do not roll over into the next subscription term.
- The unmetered add-ons cover employees only; customer-facing agents stay on the credit or conversation meter.
- Data 360 credits are excluded from every worked pricing example, so the published maths is not the whole bill.
A two-week bake-off beats a capability matrix
Build the same agent twice, under a fixed deadline, and let the result carry the argument into the room.
- Name the authoritative system for every step of the workflow.
- Agree the success threshold before either build starts.
- Use one evaluation set, graded by the same person, for both.
- Time-box to two weeks and ship whatever exists at the end.
- Name the team that will own it in eighteen months.
- Price steady-state and peak volume under both meters, then again at 3x.
- Check SSO, audit logging and data handling against your own policy.
- Confirm overage behaviour: agents disabled at 125% of prepaid capacity on one side, no penalty on the other.
A bake-off beats a matrix because it surfaces the costs a matrix hides: which platform your team can actually maintain, how much integration work sits outside the chosen estate, and whether the agent clears your quality bar at all — which, for a meaningful share of workflows, it will not yet on either.
Grade both builds on four things and nothing else: task success against a fixed evaluation set, time to a working version, who could maintain it without the person who built it, and modelled cost at 3x expected volume. Everything else is preference. Agent evaluation covers the method and agent observability covers what you need to see once it is live. Incumbency will still weigh on the result, and that is fine — an estate you already govern is a genuine advantage. You just want it counted deliberately.
Which should you choose?
Two of these three answers are the suite you already own. The third exists because a lot of real work does not live in either one.
Choose Copilot Studio when…
Identity, documents and collaboration already run on Entra ID, SharePoint and Teams; the agent's work is mostly internal; and you hold Microsoft 365 Copilot licences, which mark employee-facing usage as no charge within fair-usage limits. Data policies, Purview and Sentinel then come as governance you already run rather than governance you build. It is also the option to look at for voice and for US Government Community Cloud.
Learn moreChoose Agentforce when…
The work is customer work and the records that matter are already CRM records. Existing Flows, Apex and knowledge articles become agent actions directly, Data 360 supplies grounding, and the Trust Layer's masking, zero retention and audit trail are in place before you start. Voice sits in front of Amazon Connect, Five9, Genesys, NiCE or Vonage. Try it on the $0 Foundations tier before you commit to a meter.
Learn moreWhen neither is the right answer
Your records are split across a warehouse, several line-of-business apps and both suites; or your team owns neither admin skill set; or you want model choice and a bill not attached to a suite renewal. Neither vendor's adjacency argument pays outside its own estate, and you would be buying it anyway. Then an independent platform is the honest recommendation.
Learn moreIf you remember one line, make it this: whichever suite already holds your system of record and your identity estate usually wins, and the licensing model decides the bill. Builder ergonomics, template libraries and the current state of feature parity all move faster than your procurement cycle.
The third card deserves an explanation rather than a pitch. Both suite-native options are strongest inside their own boundary and weakest outside it, so the teams they serve least well are the ones whose work sits between the two. If an agent has to read from a warehouse, update a CRM record, post to a chat channel and file a ticket in a fourth system, no suite boundary contains that workflow, and buying one anyway means paying for adjacency you will not use.
That is the gap AI Agentics is built for: a no-code visual builder with optional TypeScript and Python SDKs, 200+ integrations including Slack, Gmail, GitHub, Notion, SQL and REST, LLM-agnostic model routing with your own keys, multi-agent teams with handoffs, and one-click deploy to a REST endpoint with a scoped key. Guardrails, human-in-the-loop approvals, audit logs, SOC 2 Type II and SSO/SAML are in the platform, with a free tier of 500 runs a month. One caveat, stated plainly: it is fully hosted SaaS — no self-hosting, no VPC edition, no on-prem — so if the platform must run inside your own network, rule us out early and save the call.
Already narrowed to one side and want the field around it? See Copilot Studio alternatives and Agentforce alternatives. If the build-surface question is still open, no-code vs code agents is the other axis worth settling first.
Verify current capabilities and pricing before you commit
Every figure here was read from Microsoft’s and Salesforce’s own pricing and documentation pages in August 2026. They are US list prices, they exclude regional variation, and they exclude whatever an enterprise agreement does to them. Some numbers are not published at all — Microsoft shows no pay-as-you-go rate for a single Copilot Credit, and Salesforce puts its detailed rate card behind a download and an account executive. Both vendors revise licensing on a cadence measured in weeks. Confirm current capability, licensing units, included allowances and overage terms on each vendor’s own pages — and in a written quote — before committing.
Copilot Studio vs Agentforce, answered
Both are the agent layer of a large business suite, so the honest difference is which estate you are already inside. Microsoft describes Copilot Studio as a graphical, low-code studio for building agents, workflows and agent flows, built on Power Platform, grounded in Microsoft Graph, SharePoint and Dataverse, and published into Teams, Microsoft 365 Copilot, websites and voice channels. Salesforce builds agents in Agentforce Builder on CRM records and Data 360, turning existing Flows, Apex, prompt templates and APIs into agent actions. Capability overlaps heavily; the data underneath does not.
Go deeper on agent platforms and evaluation
Agents for work that lives in more than one estate
If your workflow crosses a warehouse, a CRM, a chat channel and a ticketing system, build it somewhere that treats all four the same way. Free to start — no credit card required.