Power Automate Copilot Review: What It Builds and Where It Needs Review

Copilot in Power Automate is Microsoft’s AI assistant that turns natural language instructions into cloud and desktop flows, explains existing flows, and helps fix errors, and each connection still needs a check before the flow runs. With Microsoft Copilot Studio and AI Builder, Power Automate is a practical starting point for Microsoft-centered stacks, and Copilot Credit cost is the main caution for agents.

TL;DR

• Copilot in Power Automate builds a cloud flow from a plain-English prompt, sets up connections, and applies parameters, then asks you to confirm each connection before you create the flow.
• Several desktop features, including Record with Copilot and natural language to script, are still in preview.
• Microsoft Learn publishes hard limits on actions per flow and on daily Power Platform requests, and the license of the flow owner sets the ceiling.
• Agent flows run on Copilot Credits, so credit cost is the main caution once agents enter the picture.

A flow that Copilot drafted from one sentence is sitting in your change queue, and the operations lead who wrote that sentence wants it live. The canvas looks finished. Underneath, you are checking for blank parameters, a connector chosen from the wrong side of the cloud and desktop split, and a request ceiling set by whoever owns the flow. If you are the IT director deciding whether AI-drafted output is ready for real work, three questions matter: what it builds, where it needs review, and what it costs once agents enter the picture.

Key features and capabilities of Copilot in Power Automate

Copilot turns a plain-English description into a flow, then helps with the parts where builders get stuck: expressions, errors, and logic somebody else wrote.

• Natural-language flow creation: A prompt such as “when an urgent Outlook email arrives, create a Planner task and send a Slack message” returns a suggested trigger and actions, and Copilot sets up the connections and applies the parameters it can infer.
• Expression help: Copilot in Power Automate creates and edits expressions from a description of the result you want, which spares cloud flow builders from memorizing function syntax.
• Error repair: Troubleshoot in Copilot helps identify and resolve errors found while testing a cloud flow or reviewing its run history.
• Flow explanation: Ask Copilot “What does my flow do?” and it summarizes an existing flow in plain language, which helps when nobody remembers why a complex automation was built.
• Record with Copilot: Power Automate for desktop records steps performed on screen and uses them as the basis for a desktop flow (in preview).
• Natural language to script: Power Automate for desktop accepts a written description and produces a script for robotic process automation (RPA), also in preview.
• Each feature produces a starting point for the builder to review, and Microsoft recommends testing every flow after it is saved.

Pros: where Copilot in Power Automate earns its place

Speed to a first version is the strongest case for Copilot in Power Automate, and non-technical Microsoft 365 users feel that gain most.

• Lower barrier to entry: A non-technical user can describe a job in plain English and reach a working starting point without first learning the designer.
• Scaffolding that saves time: A generated skeleton removes blank-canvas work, so experienced builders spend their hours on logic and configuration.
• Microsoft 365 integration: Prompts that name Outlook, Teams, SharePoint, or Planner map to connectors in tools the team already runs, which keeps early experiments inside a familiar environment.

Those gains stop at the draft, and the next section shows where the cleanup begins.

Cons and limitations: where Copilot in Power Automate needs review

Connections are the first thing a reviewer should check, followed by the limits Microsoft itself documents for Copilot.

• Connections to confirm: After you keep a suggested flow, Microsoft’s steps have you verify every connection, and a connection marked with a red exclamation mark needs setup before you create the flow.
• New designer only: Copilot works in the new cloud flows designer, not the classic designer, so teams that still edit in the classic designer cannot use it.
• Language support: Microsoft built Copilot for English first and gives other languages limited support, so parts of the experience can appear in English.
• Preview features: Record with Copilot, natural language to script, and runtime error repair in Power Automate for desktop are still in preview.
• Vague prompts produce vague flows, so expect several rounds of refinement before the structure matches the process.

Microsoft describes Copilot as new technology that is still being developed, so test every generated flow against your own connectors before you trust it in production.

Quick tips for prompting Copilot in Power Automate

Better prompts shorten the cleanup without removing it, and the fix usually starts with the first clause you type.

• Treat it as a draft: Read every generated step before saving, because a flow that looks complete can still hold empty fields.
• Be specific with triggers: Microsoft recommends the “When X happens, do Y” format, and “When a new file is added to the Invoices folder in SharePoint” gives Copilot far more to work with than “when I get an invoice.”
• Name the connector: State the app, library, mailbox, or Teams channel in the prompt, as Microsoft advises, so the connector choice has less room to drift.
• Start in the right place: Copilot in cloud flows and Copilot in Power Automate for desktop are separate experiences, so open the one that matches the flow type you need.

Power Automate limitations: the numbers that apply

Microsoft Learn publishes hard ceilings for flow design, execution, license tier, and desktop runs, and every figure below comes from those pages. Microsoft counts usage in Power Platform requests (also called actions), so both terms describe the same unit here.

Design limits: how large can a flow get?

• Actions: 500 per flow, per Microsoft Learn (child flows carry anything longer).
• Nesting: 8 levels, per Microsoft Learn (counted through conditions and loops).
• Switch cases: 25 per switch, per Microsoft Learn.
• Variables: 250 per flow, per Microsoft Learn.
• Names: 80 characters, per Microsoft Learn (applies to each action and trigger).
• Expressions: 8,192 characters, per Microsoft Learn (applies to a single expression).

Execution and concurrency: how long and how fast can a flow run?

• Run duration: 30 days, per Microsoft Learn (the longest a single run stays active).
• Run retention: 30 days, per Microsoft Learn (how long run history is kept).
• Continuous failure: 14 days, per Microsoft Learn (after which a flow can be turned off).
• Idle time: 90 days without trigger activity, per Microsoft Learn (after which a flow can be turned off, with premium-licensed or capacity-assigned owners exempt).
• Concurrency control: 1 to 100 concurrent runs with a default of 25, per Microsoft Learn (off until you turn it on).
• Apply to each: 5,000 array items on the Low profile and 100,000 on all others, per Microsoft Learn.
• Throughput: 100,000 Power Platform requests per 5 minutes, per Microsoft Learn, whatever the license.

Request limits by license: how many actions does each plan allow?

License tier sets the daily ceiling, which makes this table the one to check before you commit to a plan. Figures come from Microsoft Learn and cover a 24-hour window.

During the transition period, Microsoft applies Premium and seeded limits per cloud flow, and the per-user limits take over once it ends, so design for the official column.

Desktop flows: what caps apply to RPA?

• Logged actions: 10,000 per run, per Microsoft Learn (extra actions still run but aren’t logged).
• Hardware floor: two or more cores and 2 GB of RAM at minimum, per Microsoft Learn (4 GB of RAM recommended, and four or more cores for unattended mode).
• Flow size: 100 MB for flows on the v1 schema, per Microsoft Learn.
• Requests: desktop flow runs don’t consume Power Platform requests, per Microsoft Learn.

Power Automate AI agents enterprise review: are they worth the credits?

For organizations on Microsoft 365, Azure, and Dynamics 365, Power Automate gives AI agents a governed place to start, provided someone watches consumption from the first pilot.

Overview and core capabilities

• Automation types: Power Automate runs cloud flows, desktop flows, and AI-powered flows through AI Builder.
• Agent layer: Microsoft Copilot Studio builds agents that can call Power Automate flows, and agent flows run in Copilot Studio on Copilot Credits.
• Legacy reach: Desktop flows give cloud flows an RPA route into legacy systems, although Microsoft does not support desktop flows as a tool in Copilot Studio.

Enterprise pros

• Governance and audit: The Power Platform admin center gives admins environment routing, data loss prevention (DLP) policies, audit logs, and daily credit usage per environment, so review starts from familiar ground.
• Time to value: Native connectors shorten integration work for flows and agents that touch Microsoft 365 and Dynamics 365 data.

Enterprise cons and cost

• Agent flow credits: Agent flows use Copilot Credits instead of Power Platform requests, and converting a cloud flow to an agent flow cannot be undone.
• Cost escalation: Microsoft sells Copilot Credits as a tenant-wide pack of 25,000 credits for $200.00 per month, or bills them pay-as-you-go, so a small pilot’s bill says little about a company-wide rollout.
• AI Builder credits: Microsoft stopped selling AI Builder capacity add-ons on 1 November 2025, and premium licenses bought after 1 November 2026 carry no seeded AI Builder credits, so AI Builder features increasingly draw Copilot Credits.
• Redundancy: A standard cloud flow handles a rules-based task within the owner’s Power Platform requests, without drawing Copilot Credits.
• Scaling bottlenecks: Cloud flows that a Copilot Studio agent triggers share one allowance of 250,000 Power Platform requests per tenant per 24 hours, so high-volume designs need capacity planning early.

Flow or agent? A decision test

Ask one question of each step: does it need judgment over unstructured input? If the answer is no, a standard cloud flow handles that step without an agent.

Before an AI-generated flow touches approvals

Ownership, audit trail, and access control decide whether a generated flow belongs in an approval path.

• Ownership: Limits follow the plan of a flow’s owner, and a departed owner reverts that flow to the Low profile, per Microsoft Learn.
• Process licensing: A Process license requires the flow to sit inside a solution, per Microsoft Learn, so confirm the packaging before you budget for that tier.
• Audit trail: Confirm in your tenant who changed a generated flow and when, because an auditor will ask who edited it after Copilot drafted it.
• Access control: List who can edit the flow, who can share it, and whose connections it runs under.

Where Kissflow fits

Kissflow is an AI-powered enterprise application platform, driven by business logic, not code, on which business teams and IT build, automate, and govern their core work. It fits IT-governed approval workflows and exception routing, the structured work that falls at the edges of what the core systems were built to do, while SAP, Oracle, Salesforce, and Workday stay in place as systems of record.

• Who builds: Business teams build on Kissflow, IT governs what they build, and developers extend it when the logic gets hard.
• AI output: Kissflow AI generates blueprints, structured metadata that describes business logic, and every artifact is a standard Kissflow component a developer can open and read.
• Audit: AI changes use the same audit log as manual edits, and every action is timestamped and attributed.

Kissflow’s AI Builder is a separate product from AI Builder in Power Automate, despite the shared name. Generally available as of May 2026, it generates a complete working app from a single prompt or an uploaded requirements document, refined through conversation, and people own and review the blueprints.

Kissflow uses subscription-based, user-tier pricing. Users are grouped into blocks by level of platform interaction, such as builders versus occasional approvers. Integrations, APIs, and AI features are included in the subscription at no extra cost. Enterprise customers can access volume-based pricing and multi-year agreements, with three-year contracts typical. Implementation, consulting, and support are scoped separately, delivered directly or through partners.

Founded in 2012, Kissflow serves over 1,200 customers worldwide. Security reviewers can check its certifications (SOC 1 Type II, SOC 2 Type II, SOC 3, ISO/IEC 27001, HIPAA, GDPR, and CCPA), hosting on Google Cloud Platform (primary) and AWS (secondary), and data residency in the US, EU, APAC, and Oceania. Kissflow was included in the Forrester AppGen and Low-Code Platforms Landscape, Q2 2026, included in the Gartner Hype Cycle for Enterprise Process Automation, 2025, named a Strong Performer in The Forrester Wave, Q1 2024, and named an Established Vendor and Customers’ Choice in Gartner Peer Insights 2024. For a side-by-side view of the app side of Microsoft Power Platform, read the Kissflow and Power Apps comparison.

Frequently asked questions

How much does Power Automate Premium cost per user?

Power Automate Premium costs $15.00 per user per month, paid yearly, and the Process license costs $150.00 per bot per month, paid yearly. Premium raises the daily request ceiling per user compared with Power Automate Free or Office 365.

Can admins turn off Copilot in Power Automate?

Yes, but only for the whole tenant. In regions with GPUs (the UK, Australia, the US, and India), Copilot is on by default, and an admin asks Microsoft support to turn it off with a PowerShell script. In other regions, an admin turns off cross-geo data sharing in the Power Platform admin center.

Does Copilot in Power Automate work in languages other than English?

Partly. Microsoft built Copilot for English first and lists limited support for other languages, so some prompts and replies can still show in English. Copilot runs on Azure OpenAI Service.

Do desktop flow runs count against Power Platform request limits?

No. Microsoft states that desktop flow executions don’t consume Power Platform requests. The cloud flow that starts a desktop flow still uses requests for its own actions.

What happens to AI Builder credits in Power Automate after November 2026?

Premium licenses bought after 1 November 2026 no longer include seeded AI Builder credits, and existing seeded credits last until the end of the contract term. AI Builder features keep working, and once AI Builder credits run out they draw Copilot Credits.

Before go-live: what to check in a generated flow

Let AI assist with the first version of a flow in Power Automate, and keep sign-off with a person. Prompt quality gets you to a draft quickly, but the checks below decide whether that draft is safe to run.

• Confirm every connector matches the system named in the prompt.
• Fill every required parameter with a real value.
• Run the flow from trigger to last action on test data.
• Check that a named owner holds the license the flow depends on.
• Confirm that daily volume fits the owner’s request limit.
• Put any agent step through the judgment test above.

Book a demo to see how Kissflow governs approval workflows around your systems of record.