Match the architecture to the work: introducing Maestro Case

A CFO asked me last quarter where her AI investment was showing up in the profit and loss (P&L). It was a fair question. MIT's NANDA project had just reported that 95% of enterprise generative AI initiatives had produced no measurable revenue impact. Her board had seen the number. So had everyone else's.

The honest answer is that the bottleneck was never the models. It's what the models walked into.

The average enterprise runs on more than 175 applications, six overlapping automation tools, and a thousand quiet handoffs no agent can see. Pour intelligence into that and you do not get outcomes. You get more shadow case management in spreadsheets, more pilots that stall on the third handoff, more "automated" processes that are not.

Coding agents have changed who can build and how fast. A prototype that took a sprint now takes an afternoon, and new builders who couldn't ship a year ago are shipping now. But the conversation inside our customers has moved past authoring speed. The question now is: if my team can produce a working prototype in an afternoon, why does it still take months to run reliably in production?

Because generating code is no longer the hard part. Running it is. Organizations still struggle with how that work moves across systems, people, and processes. It goes like this: the prototype works on a laptop. Production needs it to coordinate with seven downstream systems, route exceptions to the right person, comply with audit requirements, recover from failures at 3:00 am, and do all of that at enterprise scale. That’s the gap, and there is no magic combination of better prompts, MCPs, evaluators, or sandboxes capable of closing it, because the bottleneck isn't the agent.

It's the platform underneath.

What orchestration actually does

The platform underneath has to do what the agent and the model cannot do on their own: coordinate work across all of that surface area. Orchestration changes the picture. A single operating layer, coordinating AI agents, robots, and people across the full lifecycle of complex work, with governance built in. UiPath Maestro is how we deliver it. Architecture, coupled with intelligence and a new definition of the builder persona.

The architectural choices operations leaders make over the next 24 months will decide whether their AI investment compounds or evaporates.

The mistake is treating all enterprise work like the same orchestration problem. It isn’t. Some work breaks when variation enters the system. Other work breaks when you try to force variation out of it.

Two kinds of work

Some of the most important work in your enterprise is a defined path, repeated reliably. An invoice arrives, gets matched, gets routed, gets posted, and gets paid. A loan moves from credit pull to AUS to conditions to closing. Compliance asks: did we follow the steps?

BPMN is the right modeling approach here. It is how you get predictable cycle times, throughput per full-time equivalent (FTE), and audit by construction.

Other work looks nothing like that. A complex commercial claim. An anti-money laundering (AML) investigation. A care coordination case. The goal is defined. The path is not. Complicated by exceptions, fast-changing conditions, and wider compliance and oversight concerns.

Compliance asks a different question here: what evidence did you consider, and why did you decide this way?

This is where orchestration stops being about sequencing steps and starts being about coordinating evolving context.

For this work, the right tool is Maestro Case—the new UiPath agentic case management capability. The case becomes a living business entity carrying its own data, participants, and timeline. A case manager agent governs the lifecycle. Stage manager agents drive each phase to completion.

The framing I use with customers: BPMN is for flow complexity. Agentic case management is for context complexity. Path versus goal. The question is not which one to pick. The question is whether you can run both, and the hybrids in between, on the same operating layer.

When BPMN is the right model, standardization is the value. Lake Michigan Credit Union pulled ten days out of their consumer lending cycle and now runs 15% more loan volume on the same operation. SunExpress moved three high-volume processes onto orchestrated flows and took $200,000 in savings and two months of backlog out in a quarter.

When agentic case management is the right model, outcomes are the value. One of the largest Florida-based credit unions runs check fraud as cases and now reviews ten times more checks than before, with $2.7 million in fraud losses prevented. A leading secure access service edge (SASE) platform projects 40% of inbound IT tickets resolved end to end without a human handoff, because the case carries its full context to whoever is closest to resolution. These customers are not just examples of the difference between BPMN and agentic case management; they are the evidence that you can, and should, run both.

Where they converge

The dominant reality of enterprise work is hybrid. Forcing a top-level choice between flow and context is what produces the brittle BPMN diagrams ops teams quietly route around, and the open-ended cases where standardization gets silently abandoned.

Procure-to-pay is hybrid. Accounts payable (AP) invoice processing belongs in BPMN. A supplier dispute opens as a case. Strategic sourcing runs as a case with BPMN sub-flows for request for proposal (RFP) issuance and contract approval.

Commercial claims are hybrid. The claim itself is a case: months of evolving evidence, adjusters, forensic accountants, engineers, legal. Inside that case, payment issuance and reserve-change approvals run as BPMN sub-processes with segregation of duties.

Financial crime investigations are hybrid. AML investigation runs as a case. Once the suspicious activity report (SAR) decision is made, SAR filing is a BPMN flow with deterministic controls. Same operating layer. Right model at each altitude.

UiPath Maestro is built for that reality. BPMN, agentic case management, and developer-first flow modeling run on the same event-sourced execution backbone, share the same governance fabric, the same Data Fabric semantic layer, the same process intelligence.

Hybrid work breaks when orchestration fragments. Compliance ends up with multiple audit stories. Operations ends up with multiple control surfaces. Teams route around the gaps manually.

The platform layer matters because it keeps flow work, context work, and the hybrids in between operating on the same coordination and governance layer. AIUC-1 certification and ISO/IEC 42001:2023 alignment matter for the same reason: governance only works when it applies consistently across the operation.

What I’d ask every AI leader to look at this quarter

The bottleneck was never intelligence. The work intelligence must walk into has always been the harder problem.

If your platform is forcing a choice between flow and context, the enterprise is paying for it: in shadow systems, in missed service-level agreements (SLAs), in customers who do not call back for the right reasons.

Match the architecture to the work. Run them on the same operating layer. The gap between pilot and outcome closes. Earlier this week, our CEO Daniel Dines wrote that the execution layer compounds with every model release, and the orchestration layer compounds with every new artifact built.

I’d frame the next phase this way: the enterprises that win will be the ones capable of running deterministic flows, agentic cases, and the hybrids in between on the same operational fabric.

That has been the work of the UiPath Platform for a decade. Maestro extends it: one operating layer for structured flows, judgment-heavy cases, and the systems in between. With the launch of Maestro Case, your most strategic, decision-heavy work runs at scale on the same platform your team already trusts.

Code stays easy. So does orchestrating it.