UiPath Maestro Flow [What It Changes]

TL;DR

• UiPath Maestro Flow is a developer-first orchestration canvas that lets a coding agent design, run, observe, and govern an end-to-end business process as one artifact, from prototype to production, without a rewrite.

• UiPath announced it on August 19, 2026; it sits on the same Maestro engine as BPMN and Case, and it works from VS Code, UiPath Studio, a cloud canvas, or agents such as Claude Code, Cursor, GitHub Copilot, and Codex.

• The operational change for a small team is not "more AI." It is a path to ship the script a developer already prototyped, with retries, traces, and a human pause, instead of leaving that script in a laptop folder.

• Treat vendor cycle-time claims as case-specific. Use the One NZ figures below as a published benchmark, then run the same arithmetic on your own queue before you buy a seat.

Key Takeaways

• The missing layer is orchestration, not another agent: agents, robots, APIs, documents, and people have to share one run record.

• The prototype file and the production file are supposed to be the same .flow artifact, versioned in Git, with compile-time checks on a TypeScript Flow DSL.

• Maestro Flow is the code-first authoring style; Maestro BPMN is the structured process model; Maestro Case is for exception-heavy, long-running work. All three run on one engine.

• Small shops should care when a coding tool already drafts the workflow and the stall is review, handoff, or "who ran this last Tuesday."

• Human approval is a first-class node, not a sidebar email. If money, PHI, or a customer-facing send is in the path, keep that pause.

• UiPath pricing still splits Basic from Standard: orchestrated BPMN implementation is not listed on Basic, and Basic caps users, robots, and tenants.

What UiPath Maestro Flow is

UiPath Maestro Flow is a developer-first orchestration canvas that lets supported coding agents design, run, observe, and govern a complete business process as a single artifact, from the first prototype to a live production run, without rebuilding it on a second platform.

That sentence is the whole product bet. A 2-truck HVAC shop, a 10-person marketing agency, and a solo-run clinic do not need a new New York software brand in their stack. They need the work that already leaks across tools to stop leaking. The HVAC dispatcher still copies a completed job into an invoice. The agency still pastes a form fill into a CRM and waits on a client yes. The clinic still parks insurance cards in a shared inbox. Those stalls look small until a coding agent drafts the happy path in an afternoon and nobody can prove what the agent did after hour six.

According to the SBA Office of Advocacy, 99.9% of U.S. businesses are small, and those firms employ 45.9% of American workers. That is why an enterprise orchestration launch still belongs on a small-team desk: most operating companies are already small, and most of their automations die in the gap between a clever demo and a logged, retryable run. The state of small business automation is not a shortage of chatbots. It is a shortage of a single place that can call an API, wait on a person, and leave an audit trail.

If you already route documents through US Tech Automations workflows, Maestro Flow is a candidate model swap for the orchestration layer, not a reason to throw away intake. The same is true if your agency already maps form-to-CRM tools or your plant already tracks supplier certificates of analysis. The question is whether the next prototype can ship as the same file that ran in debug.

What shipped on August 19, 2026

According to SD Times, UiPath announced UiPath Maestro Flow on August 19, 2026 as a developer-first orchestration canvas that combines AI-native build speed with production durability and governance. The same date and product name appear in the Yahoo Finance reprint of the Business Wire release and in UiPath's investor newsroom item.

The official line is blunt. Raghu Malpani, UiPath's chief product and technology officer, said enterprises do not have an agent problem; they have an orchestration problem. Builders can already use coding agents to stand up an agent. They cannot, without an orchestration layer, run a real process that spans agents, robots, systems, and people, and then prove what happened at each step. Maestro Flow is that layer, running on the Maestro engine customers already use.

The product page for Maestro Flow repeats the same mechanism in builder language: one canonical model for logic, agents, APIs, events, data, documents, and human steps; a cloud canvas and VS Code working on the same .flow artifact; coding agents such as Claude Code, Cursor, and Copilot authoring the file you later refine visually; Git version control; and compile-time validation through a TypeScript Flow DSL. Agents sit on the same decision path as rules, APIs, retries, and compensation. Tracing is described as OpenTelemetry-compliant. Durable execution is described as Temporal-based.

UiPath's newsroom index lists the Flow launch under August 19, 2026, after the June 16 Maestro Case release and the June 4 One NZ Maestro story. The IR news index carries the same sequence. As of August 19, 2026, Flow is the newest named authoring surface on that engine, not a replacement for BPMN or Case.

SD Times also notes UiPath Maestro Lite as a lightweight option for processes that do not need extensive management. That is a product-family detail, not a price. Public UiPath pricing still routes Maestro implementation through plan tiers rather than a standalone Flow SKU.

Why this constraint broke now

Two earlier moves made Flow a logical next canvas. On May 28, 2026, in its first-quarter fiscal 2027 results, UiPath said it had launched native integration for coding agents across the platform, branded UiPath for Coding Agents, so a coding agent could create, test, deploy, operate, and govern enterprise automations through natural language. Founder and CEO Daniel Dines tied that launch to agentic products moving from pilot to production one year into general availability.

The UiPath for Coding Agents page states the CLI and open-source agent skills give any coding agent native fluency in UiPath, and that the program supports any coding agent compliant with the AGENTS.md open format, with third-party model subscriptions purchased separately. According to AGENTS.md, that format is used by over 60k open-source projects and is described as a README for agents. Flow is the orchestration artifact those agents are now supposed to emit, instead of a one-off script that never leaves the IDE.

The second move was Maestro itself. The Maestro product page defines Maestro as process orchestration software that designs and runs an entire business process as one BPMN model, coordinating AI agents, robots, and people, with decisions in DMN. It states Maestro models follow BPMN 2.0, which the Object Management Group lists as formal version 2.0.2, adopted January 2014. Flow does not throw that standard away. It adds a code-first way to author work that still executes on the same engine.

How the mechanism works in plain language

Think of three files that used to be three different jobs. The developer had a prototype in an editor. Operations had a diagram in a process tool. Audit had a spreadsheet of "what ran." Maestro Flow's claim is that those collapse into one .flow file. You describe the process in natural language to a coding agent, or you draw it on the cloud canvas, or you edit it in VS Code. The same file is what you debug, evaluate, deploy, and observe.

UiPath docs on Flow core concepts frame that file around a canvas, variables and data flow, expression syntax, control flow, error handling, triggers, subflows, output mocking, evaluations, deployment, and observing runs. Built-in nodes listed in the same guide include Script, HTTP Request, Data Transform, Decision, Switch, Loop, Delay, Merge, End, Terminate, Mock, Manual Trigger, Scheduled Trigger, Agent, a BPMN process node, an RPA workflow node, an API workflow node, a Human Task, and connector nodes.

Deploying a Flow is the line between a laptop demo and a live process. Docs say you publish the workflow to Orchestrator so it can run in production, and that until you deploy, the flow cannot be triggered by external events, run on a schedule, or invoked by other systems, even if you already ran it in Debug and checked it with Eval. That is the operational meaning of "the version prototyped is the version that ships": you still have a promotion step, but you are not re-platforming the logic.

Human review is not an afterthought. UiPath's guide to adding human approval describes pausing a workflow at a review step, sending an approval task to a designated person, and continuing based on their response. That is the node a clinic uses before a claim goes out, an agency uses before a client email sends, and an HVAC shop uses before a refund posts. Executive-assistant load that currently lives in inboxes is the same pattern; see five ways to automate executive assistant tasks.

On the runtime side, the Flow product page says durable execution is Temporal-based and tracing is OpenTelemetry-compliant. Temporal describes workflows that capture state at every step and pick up where they left off after failure, with 22,783 GitHub stars listed on its homepage and a claim of 9 years in production. OpenTelemetry is the open standard for traces, metrics, and logs. According to OpenTelemetry, the collector lists 200+ components, with 12+ language SDKs and 1167+ integrations in the public ecosystem counts on that page. Those are infrastructure facts about the standards UiPath says it is using, not independent proof of UiPath's implementation quality.

The Maestro overview docs present Maestro as capabilities for orchestrating automation, AI agents, and human work through BPMN, Flow, and Case Management. The Maestro FAQ on the product page says Maestro is included with the UiPath Platform and points buyers to plans and pricing. It also says every workflow, integration, and credential already in a UiPath tenant is callable from a Maestro process, with no rebuild.

Coding agents and editors the canvas actually names

The August 19 launch names four coding agents in the sentence that matters: Claude Code, Cursor, GitHub Copilot, and Codex. Native environments named in the same sentence are VS Code and UiPath Studio. The Flow product page adds Windsurf and a no-install cloud canvas (Studio Web). The coding-agents page also lists Google Antigravity CLI and Gemini among the "bring your own" set, with the AGENTS.md caveat.

Claude Code is Anthropic's coding agent for terminal, IDE, Slack, web, and more. According to Anthropic, Claude Code on the Pro plan is €15 per month with the annual subscription discount (€180 billed up front; €18 if billed monthly), with Max 5x at €90 per month and Max 20x at €180 per month, and usage limits applying. Cursor is an AI coding agent and IDE that states it is trusted by over half of the Fortune 500. GitHub Copilot is GitHub's AI pair programmer, with a Free plan at $0 USD per user per month and 2,000 completions per month. According to GitHub, Copilot Pro is $10 USD per user per month. The same GitHub page states developers who use Copilot are up to 55% more productive at writing code and report up to 75% higher job satisfaction than those who do not; those are GitHub's own product claims.

OpenAI's Codex index is an older launch post that now carries an update dated May 16, 2025: OpenAI launched Codex as a cloud-based software engineering agent that can work on many tasks in parallel. UiPath's 2026 materials treat Codex as one of the supported coding agents on Flow. Visual Studio Code docs currently lead on AI agents that can run locally, in the background, or in the cloud, which is the editor UiPath says Flow is built to stay inside.

Sources: GitHub Copilot; Claude Code; OpenAI Codex; Cursor; Maestro Flow.

Those prices are the coding-agent bill, not the UiPath bill. UiPath states third-party AI model subscriptions must be purchased separately. A 10-person agency that already pays Copilot Pro is looking at $100 per month for that layer alone before Orchestrator seats, robot runtimes, or document units.

Flow versus BPMN versus Case

UiPath now sells three authoring styles on one engine. Mixing them up is how teams buy the wrong canvas.

Sources: Maestro Flow; Maestro; Maestro Case IR release; OMG BPMN 2.0.2; SD Times.

The Flow FAQ says the choice between Flow and BPMN is the kind of process you are modeling — lightweight, composable orchestration versus structured, standardized BPMN — not whether you want a "code-first" tool, because both support canvas and CLI authoring. Case is the third path. The June 16, 2026 Case release cites a UiPath survey of nearly 600 C-suite and IT practitioners at companies with $1B+ in revenue, with 52% reporting hybrid workflows that mix static, repeatable processes and dynamic, context-dependent processes. Early design adopters in that release are described as reporting a 60–80% reduction in average case handling time, a three-to-five times increase in cases resolved without human intervention, SLA compliance improvements of more than 25 percentage points, and one financial-services adopter projecting more than $12 million in annual savings. Those figures are vendor-reported early-adopter results, not a small-business sample.

Manufacturing changeover checklists and work-instruction acknowledgements are closer to BPMN or Flow with a human task than to Case. If that is your world, keep the structured path: changeover checklist acknowledgement and work-instruction acknowledgement are the same "pause for a named person" pattern Flow now documents as a node.

What the vendor's own scale looks like

Flow launched into a company that already publishes large run-rate numbers. According to UiPath's Q1 FY2027 release, ARR reached $1.901 billion as of April 30, 2026, up 12 percent year-over-year. The same release reports revenue of $418 million, up 17 percent year-over-year, net new ARR of $49 million, a dollar-based net retention rate of 109 percent, GAAP operating income of $28 million, and cash, cash equivalents, and marketable securities of $1.42 billion. The UiPath IR homepage repeats the $1.901 billion ARR and 109 percent net retention figures as of April 30, 2026, and lists 2,624 customers with $100K+ ARR and 374 customers with $1M+ ARR.

Sources: Q1 FY2027 results; UiPath IR; UiPath pricing.

ARR reached $1.901 billion as of April 30, 2026. That scale number, linked above, is why Flow is staffed as an enterprise canvas first. The SMB read is narrower: Basic still lists a 2-robot cap, a 1-tenant cap, and a dash in the row for implementing orchestrated BPMN processes. Standard and Enterprise fill that row. If your test is "can a 10-person shop turn on Maestro Flow this afternoon on the cheapest public tier," the public matrix does not say yes. Contact-sales pricing is not a published seat price.

The coding-agents page also states UiPath Community includes 3.3M+ developers. That is a community figure, not a paying-customer figure.

One NZ, and a labeled USTA analysis

The cleanest published cycle-time pair on the Maestro engine is not a Flow customer story. It is a Maestro story from June 4, 2026. According to the One NZ IR release, One NZ used UiPath Maestro to cut mobile provisioning from 10 days to under 10 minutes, and the deployment was implemented in five weeks. The same release's body also describes order cycles stretching to four or five days before the change, so the press package uses both a ten-day headline and a four-to-five-day body figure. The Maestro product page repeats "<10 min for mobile provisioning, down from 10 days" and "5 weeks, not years," and lists "< 6 months return on investment" under the One NZ case.

Provisioning dropped from 10 days to under 10 minutes. That pair is the one we can put on a calculator without inventing a new study.

USTA analysis: cycle-time compression from the published One NZ pair

Inputs (headline pair from the June 4, 2026 IR release): baseline = 10 days; stated outcome = under 10 minutes. We treat "under 10 minutes" as 10 minutes so the ratio is conservative (a faster finish would only raise the multiple). We show two clocks because the release does not say whether "ten days" is calendar time or working time.

Sources for inputs: One NZ IR release. Compression column is USTA arithmetic on those inputs, not a UiPath figure.

A 2-truck HVAC shop should not copy 1,440× onto a dispatch board. One NZ was connecting Salesforce, Oracle, and internal platforms for enterprise mobile provisioning. The usable lesson is smaller: if your current cycle is measured in days because of handoffs, an orchestration layer that keeps agents, robots, and a person on one instance can collapse wait time. Measure your own queue in minutes, then divide. If the result is 2×, you still have a process. If you cannot measure the queue, you are not ready to buy the canvas.

The Maestro page also lists a 75%+ cut in turnaround time for new patient process issues at PromptCare (2025) and $200K+ saved across three core processes at SunExpress (2026). Those remain vendor-attributed case figures on that page.

What a small team should actually do with it

Start from the job that already exists. A clinic inbox that files insurance cards is a document-plus-human-task flow: extract, confidence-check, pause, write back. A marketing agency that dumps Typeform into a sheet is an HTTP-plus-CRM flow with a client approval node. An HVAC shop that emails a completed job to bookkeeping is an API-plus-human-task flow. None of those require a $1M ARR customer profile. They do require a place the run can wait overnight without losing state.

US Tech Automations already models that shape as an agentic workflow with a human step. If your team is mapping intake on that canvas today, a later UiPath Flow, a BPMN diagram, or a Case stage is a destination for the same steps, not a new religion. Keep the pause before money moves. Keep the log. Keep the named owner.

Governance is not optional because the agent is clever. The NIST AI Risk Management Framework is voluntary U.S. guidance, released January 26, 2023 as AI RMF 1.0, with a generative-AI profile (NIST AI 600-1) released July 26, 2024, and a concept note for a critical-infrastructure profile dated April 7, 2026. Flow's own marketing points at traces, step inspection, explainability, and pre/post evaluation loops. Map those words to NIST's "Govern, Map, Measure, Manage" loop in your own language: who is allowed to deploy, what data the agent can see, how you measure a bad run, and how you roll back.

Pricing honesty matters for a startup-sized team. Basic includes limited agentic automation and human-in-the-loop, with up to 5 Basic, 5 Plus, and 1 Pro user, 2 robots, 1 tenant, 1 folder, and 6 months of cold data retention. Orchestrated BPMN process implementation is a dash on Basic and present on Standard and Enterprise. Document classification and extraction at scale is also a dash on Basic. If your clinic's first flow is insurance-card extraction, read that matrix before you promise a go-live.

Data-extraction and finance paths on this site follow the same split between "the model reads the page" and "the workflow owns the outcome." See data extraction and finance and accounting if the first artifact you need is a governed run, not a UiPath tenant.

Honest limits

Flow does not delete the need for credentials, selectors, environment promotions, or a person who understands the exception. Docs still require a deploy to Orchestrator before external triggers and schedules work. Coding-agent output still follows templates and analyzer rules; it is not a substitute for a failing integration. UiPath says third-party model spend is separate. Public pricing does not print a Flow sticker price.

The One NZ and Case early-adopter figures are large-company stories. The SBA counts say most U.S. firms are small; they do not say those firms will see a 1,440× compression. Maestro Lite is named, not specified with limits in the launch Q&A. Windsurf and Google Antigravity appear on UiPath pages; this article does not treat those as independently verified integrations beyond UiPath's own product copy.

A prototype that sends email, moves money, or writes a medical chart without a human node is not a Flow success. It is an unattended risk. Keep the approval node.

Signal vs Speculation

Signal (sourced): UiPath shipped UiPath Maestro Flow on August 19, 2026 as a developer-first canvas on the existing Maestro engine, with named support for Claude Code, Cursor, GitHub Copilot, and Codex, and with VS Code and Studio as named IDEs. Docs describe a .flow artifact, a deploy step to Orchestrator, human-approval nodes, Temporal-based durable execution, and OpenTelemetry-compliant tracing. UiPath reported $1.901 billion ARR as of April 30, 2026. One NZ's Maestro deployment is published as 10 days to under 10 minutes, implemented in five weeks. SBA Advocacy's 2024 FAQ states 99.9% of U.S. businesses are small. NIST's AI RMF remains voluntary guidance.

Speculation (our read): If coding agents keep being the way mid-size teams draft automations, a governed .flow (or an equivalent artifact on another platform) becomes the promotion format the way a Git pull request became the promotion format for application code. Over 12 months, expect the fight to be packaging and price, not the idea of "one artifact." Over 24–36 months, small firms will not all land on UiPath. They will land on whatever lets a Copilot or Claude Code session leave a retryable run with a named approver. Our read: if UiPath keeps Flow, BPMN, and Case on one engine and keeps Basic too thin for real orchestration, Standard becomes the real floor for this story, and many 10-person shops will keep the prototype in an IDE until a cheaper governed runner shows up. Our read: the One NZ multiple is a ceiling illustration, not a forecast of HVAC dispatch times. Our read: vendors that cannot show instance-level traces will lose regulated work even if their agents write prettier code.

UiPath also published forward guidance in the Q1 release (Q2 ARR $1.929 billion to $1.934 billion as of July 31, 2026; full-year ARR $2.058 billion to $2.063 billion as of January 31, 2027). That is company guidance, not a measured result.

What to check before you change tools

Write the current path on one page: trigger, systems, human pause, system of record, and how you know it finished. Time 20 real jobs with a kitchen timer. If you cannot time them, you cannot test Flow. Then rebuild that path as a flow with an HTTP node, an agent node only where judgment is required, and a human task where money, PHI, or an outbound customer message happens. Deploy it to a non-prod folder. Compare traces to the kitchen-timer sheet.

If the artifact cannot wait overnight and resume, it is not orchestration. If the coding agent cannot emit the same file the canvas opens, you are back to a rewrite. If the cheapest public tier cannot run the orchestrated process, budget for Standard or pick another runner.

Operators who want that shape without starting on a UiPath tenant can open the agentic workflow builder and put the same pause on a live canvas. US Tech Automations is the homepage for that path. Browse the resource library if you need adjacent runbooks before you touch a vendor quote.

Glossary

• UiPath Maestro Flow: Developer-first canvas for authoring a business process as one .flow artifact that coding agents and humans can design, run, observe, and govern.

• Maestro (engine): UiPath's orchestration runtime that executes Flow, BPMN, and Case on one platform.

• BPMN 2.0: Business Process Model and Notation, the OMG standard (formal 2.0.2, January 2014) Maestro uses for structured process diagrams.

• Coding agent: An AI tool that writes and edits software in an IDE or terminal (Claude Code, Cursor, GitHub Copilot, Codex, and others).

• .flow artifact: The file Flow treats as the process; cloud canvas and VS Code are supposed to edit the same file.

• Durable execution: Runtime behavior that records state so a failed step can resume instead of restarting the whole job; UiPath attributes this to a Temporal-based design.

• Human task / human-in-the-loop: A node that pauses the flow, assigns a person, and continues on their response.

• AGENTS.md: Open markdown format for instructions that coding agents read; UiPath says compliance with it is the compatibility bar.

FAQ

What is UiPath Maestro Flow?

UiPath Maestro Flow is a developer-first orchestration canvas that lets coding agents and developers design, run, observe, and govern an end-to-end business process as one artifact from prototype to production. It runs on the UiPath Maestro engine rather than as a separate automation island.

How is UiPath Maestro Flow different from UiPath Maestro?

Maestro is the orchestration engine. Flow is one way to author what that engine runs, aimed at developers and coding agents, using a .flow file instead of a BPMN diagram or a Case record. BPMN and Case remain available on the same engine.

Which coding agents work with UiPath Maestro Flow?

The August 19, 2026 launch names Claude Code, Cursor, GitHub Copilot, and Codex, used from native environments such as VS Code or UiPath Studio. UiPath's coding-agents page also describes support for any agent that follows the AGENTS.md format, with model subscriptions billed by those vendors.

Does a small business need Flow or BPMN?

Use Flow when a developer or coding agent is composing APIs, agent steps, and a few rules. Use BPMN when operations needs a shared diagram with SLAs. Use Case when the work is a long-running file with exceptions, such as a dispute. A 10-person shop can start with Flow and a human-approval node if the path is short and API-heavy.

Can a flow include a human approval step?

Yes. UiPath docs describe pausing a workflow, sending an approval task to a designated person, and continuing on that response. Keep that node on any path that spends money, sends a customer message, or writes a medical or financial record.

Is Maestro Flow a separate product on the price list?

UiPath's Maestro FAQ says Maestro is included with the UiPath Platform and points to plans and pricing. The public pricing matrix still gates "implement orchestrated BPMN processes" to Standard and Enterprise, and Basic lists a 2-robot cap and a 1-tenant cap. There is no public dollar sticker for Flow alone on the pricing page fetched for this article.

What should you refuse to leave in a coding-agent prototype?

Refuse an unattended send, an unattended payment, and an unattended write to a system of record. Refuse a demo that cannot be deployed to Orchestrator (or an equivalent runner) and cannot show a trace. Refuse a rebuild that throws away a working intake path just to chase a new canvas.

UiPath Maestro Flow is the name of the canvas. The job is older: one artifact, one run record, one named person on the pause. Map that flow on the agentic workflows canvas if you want to see the same steps without waiting on a quote.