Microsoft Signals Agent-First Shift as Enterprise Apps Move to Human-Agent Collaboration

Microsoft is positioning enterprise applications for an agent-first era, arguing that both how software is built and how it is used are changing simultaneously, with governance and operational oversight embedded from the outset.

In a conversation with Daniel Newman, CEO of technology research and advisory firm Futurum, Ryan Cunningham, corporate vice president of Microsoft Power Platform, described the current moment as one in which “the world is being turned upside down in two dimensions at the same time.” One dimension is how software is built, with agents and AI expanding what can be created. The other is who participates.

In other words, it is “the dramatic expansion in efficacy of what [we] can build and the dramatic expansion of who can participate,” Cunningham said.

Rewiring System Design Around Process Ownership

Microsoft’s thesis centers on bringing process experts directly into system design. Companies that want AI to reshape operations need HR, finance, supply chain, and service leaders shaping the workflows agents will execute. That requires higher abstraction platforms where agents assist with coding, orchestration, and automation rather than expecting business users to generate raw code.

Cunningham emphasized that adopting an agent-first model is not about removing humans from the loop but redefining their role. Agents monitor, reason, and act, while humans define intent and provide oversight. In retail scenarios such as fraud detection and refunds, Microsoft has seen agents radically compress response times and unlock “millions of dollars of value” by transforming processes rather than merely accelerating them.

Plan Designer within Power Apps reflects this architecture-led approach. Instead of generating code from a single prompt, it embeds what Cunningham described as “a digital software team,” including a requirements agent, process agent, data agent, and solution architect agent. The experience is collaborative and structured, teaching process experts to think like solution architects rather than relying on opaque automation.

Non-Negotiable Architecture: Trust, Auditability, Scale

Governance and trust remain central. Cunningham framed trust as foundational, stressing that systems must be observable, auditable, and adjustable. In regulated industries, customers are already using the platform to modernize legacy systems, including decades-old codebases, while leveraging managed governance, security, lifecycle management, and high-availability capabilities built into the broader Power Platform stack.

Security, he argued, requires a defense-in-depth strategy. “It is absolutely risky to go too fast. It is also very risky to go too slow,” he said. The cost of inaction, in a market where competitors and threat actors evolve simultaneously, may be the greatest risk.

Cunningham summarized the strategic direction succinctly: “The future of enterprise software is not a single killer agent or a clever prompt. It’s a managed environment where intent turns into action—safely, repeatedly, and at scale.”

What This Means for SAPinsiders

Agent-first development reshapes application ownership. The shift expands system design beyond IT into business units with deep process expertise. ERP and platform leaders must support collaborative architectures where domain experts and AI agents co-create workflows under structured governance.

Governance moves from afterthought to architectural layer. Managed security, lifecycle oversight, audit trails, and operational controls are positioned as core platform capabilities rather than subsequent additions. Vendors that cannot embed observability and traceability directly into agent orchestration will struggle in regulated and mission-critical environments.

Human-agent collaboration becomes the new interface standard. Traditional task-based UIs are evolving into oversight and orchestration spaces where employees supervise, intervene, and refine agent activity. Application roadmaps will increasingly focus on visibility into agent actions and structured intervention rather than static dashboards alone.