The Agentic Arms Race: Five Platform Launches in Five Weeks, and What They Tell Us About the Future of Enterprise AI
In the span of a single month, every major platform vendor shipped its answer to the same question: who gets to orchestrate the AI agents that will run your company? The answer is messier than anyone wants to admit.
On 12 February, UiPath flipped the switch on Maestro, its orchestration layer for agentic automation. Five weeks later, on 20 March, Microsoft published a security framework for the exact same category of software. In between, Alibaba launched an enterprise agent platform called Wukong, Deloitte and UiPath jointly unveiled an “Agentic ERP” offering, and NVIDIA released an Agent Toolkit with an alliance that reads like a who’s-who of enterprise software: SAP, Salesforce, ServiceNow, CrowdStrike, Siemens.
Five launches. Five weeks. One conclusion: the platform war for agentic AI has moved from positioning to shipping.
This is not another AI hype cycle. Or rather, it is one – Gartner estimates that only about 130 of the thousands of vendors claiming agentic AI capabilities are real, a phenomenon the firm has taken to calling “agent washing” – but the underlying technology shift is genuine and the commercial stakes are enormous. The question is no longer whether AI agents will orchestrate enterprise work. It is who controls the orchestration layer, how fast the market consolidates, and how many companies will break themselves trying to keep up.
The Orchestration Thesis
To understand why everyone shipped at once, you need to understand the thesis they all share: as AI makes software cheaper to build, the value migrates to the platform that governs what all that software does.
Daniel Dines, UiPath’s CEO, put it plainly in a recent earnings call: “When building becomes cheaper, more gets built, more processes get automated, more edge cases get addressed and more systems become autonomous. That expansion does not shrink the need for enterprise orchestration, it increases it.”
Maestro is UiPath’s bet on that thesis. Built on BPMN for process modelling and DMN for decision logic, it coordinates AI agents, traditional RPA bots, and human workers within a single visual workflow. The pitch is that the same company that spent 25 years building connectors to mainframes, Citrix, and SAP GUI is now the safest pair of hands to let AI agents loose on those same systems. By December 2025, UiPath reported that 950 customers were already developing AI agents to orchestrate over 365,000 processes on the platform.
Dines framed the value proposition in terms that would make an enterprise CIO nod: “Enterprises don’t simply pay for code; they pay for trust, for operability, and for governance – the ability to run complex systems reliably, securely and with full accountability.”
Microsoft’s Security Play
Microsoft, characteristically, approached the same opportunity from a different angle. Rather than building a single orchestration product, it published a framework. On 20 March, Vasu Jakkal, corporate vice president of Microsoft Security, laid out the company’s vision for securing agentic AI across the enterprise stack.
“The future of security is ambient and autonomous, just like the AI it needs to protect,” Jakkal wrote. “Security has to be something that’s woven deeply into every layer of the AI stack – from agents to apps, to platforms, to infrastructure.”
The practical implications are significant. Microsoft announced that Agent 365, its agent management product, would be generally available on 1 May, priced at $15 per user per month. The product extends Zero Trust principles – the security model Microsoft has spent a decade embedding in enterprise IT – to AI agents. “We think about security for agents very similar to security for people,” Jakkal explained.
The subtext is unmistakable. Microsoft is not trying to be the only place you build agents. It is trying to be the only place you govern them. If you already run Microsoft 365, Entra ID, and Purview, adding agent governance becomes less of a purchasing decision and more of a checkbox. Jakkal’s warning about what happens when you don’t govern is pointed: “Unmanaged agents may create significant risk, from accessing resources unchecked to accumulating excessive privileges and being misused by malicious actors.”
Or, more succinctly: “If you cannot see something, you cannot protect it.”
Alibaba and the Global Angle
Whilst UiPath and Microsoft competed for the Western enterprise, Alibaba quietly launched Wukong on 17 March – an AI-native platform for managing multiple agents through a single interface with enterprise-grade security. Still in invitation-only testing, Wukong is designed to handle document editing, approvals, meeting transcription, and research through coordinated agents, with Slack and Teams integrations on the roadmap.
The timing was not accidental. A day earlier, Alibaba had revealed a broader restructuring under its newly established Token Hub business group, consolidating its AI labs and model teams. CEO Eddie Wu described the moment as a “historic opportunity” at the “threshold of an artificial general intelligence inflection point.” Wukong is the commercial vehicle for that conviction.
For Western CIOs, Alibaba’s entry may seem distant. But for any enterprise with operations in Asia-Pacific – or any executive watching the global platform landscape – Wukong is a signal that the agentic AI market will not be a two-horse race.
The Downton Abbey Theory of Enterprise AI
Of all the people trying to explain what this platform shift actually means for how companies operate, Deloitte’s chief futurist Mike Bechtel may have landed on the most useful metaphor.
“For the first couple of years of the generative AI movement, the paradigm felt like one chatbot to rule them all,” Bechtel said. “But most business processes don’t require super intelligence. Teams are built to have role players, and AI is becoming a bit of a team sport.”
His analogy of choice is Downton Abbey. “The Earl on the show wouldn’t tell his staff to shine his shoes, make his lunch, gas the car, and straighten his cravat,” Bechtel explained. “He would say, ‘I’m going to town,’ and then the staff would negotiate and orchestrate between each other, based on their understanding of their scope and their roles.”
The practical instantiation, Bechtel argued, is a group of domain-specific intelligences coordinating to meet business needs without humans being overly prescriptive about every step. That is what Maestro, Agent 365, and Wukong are all trying to enable – the digital butler’s pantry.
But Bechtel’s optimism comes with a sharp edge about who benefits. Companies that use agents to simply eliminate headcount, he warned, will struggle. “Our clients who say, ‘Oh, good news, we freed up Toby to work on higher-stakes tasks’ – they’re the ones who are going to win. We see automation as a licence for elevation.”
The Valley Ahead
Beneath the velocity of these launches lies an uncomfortable truth. Gartner predicts that more than 40 per cent of agentic AI projects launched between 2024 and 2026 will be cancelled by the end of 2027. “Most agentic AI projects right now are early stage experiments or proof of concepts that are mostly driven by hype and are often misapplied,” said Anushree Verma, Senior Director Analyst at Gartner.
The failure modes are predictable: escalating costs, unclear business value, inadequate risk controls. The pattern is the same one that haunted earlier waves of enterprise technology – companies adopt the tooling before redesigning the work. They pave the cow path with agents instead of rethinking the path entirely.
It is a gap Deloitte has documented extensively: only 11 per cent of companies have AI agents fully operational in production environments, even as 75 per cent plan to invest in agentic AI this year. The delta between spending intent and production reality is where careers will be made and destroyed.
What Comes Next
The five-week barrage of platform launches tells us something important about where enterprise technology is heading. The era of piloting a single agent for a single task is giving way to an era of multi-agent orchestration – systems of agents that delegate to each other, call APIs, invoke legacy systems, loop in humans for judgement calls, and report back. The platform that governs this system becomes the most strategic layer in the enterprise stack.
UiPath is betting that its legacy connectivity and visual orchestration will win in complex, regulated environments. Microsoft is betting that security and identity governance, woven into the M365 fabric, will make it the default. Alibaba is betting that it can own the enterprise agent layer in Asia and, eventually, everywhere else. NVIDIA is betting that infrastructure wins, regardless of who builds the application layer above it.
They cannot all be right. But they can all be early. And for the enterprise leaders watching this unfold, the real risk is not choosing the wrong platform. It is mistaking a platform purchase for a strategy – and discovering, eighteen months and several million dollars later, that you automated the old way of working at machine speed.
As Dines put it: enterprises pay for trust, operability, and governance. The platforms have shipped. The question now is whether the organisations buying them are ready for what comes next.