A seismic shift is underway in enterprise software, one that pits two of the most influential leaders in the industry—Microsoft’s Satya Nadella and Salesforce’s Marc Benioff—against each other in a war of words that highlights the broader transformation unfolding in enterprise applications. Nadella predicts the demise of traditional SaaS applications, arguing that AI-powered agents will take over, fundamentally redefining enterprise software. Benioff, on the other hand, argues that AI will serve as an augmentation layer and rely on his existing platform to supply the digital workforce of the future.
The public sparring between these two titans reflects a deeper reality: enterprise software is undergoing a fundamental transformation. Traditional linear value chains are giving way to AI-driven orchestration models, where data is not simply an asset but the core foundation for intelligent, adaptive business processes. While Microsoft and Salesforce offer competing visions, the future is unlikely to be dominated by a single paradigm. Instead, we see an evolution in which AI agents play an increasingly significant role, but enterprise applications retain key structures for governance, compliance, and process standardization.
In this Breaking Analysis, we build on previous research and try to squint through the noise and paint a picture of what the future of enterprise software and agentic AI will look like. We’ll do so by juxtaposing the approaches being pursued by two industry titans who are sniping at each other in the public domain.
The CEO War of Words Lays out Two Different Paths to Agentic AI
In recent public statements, Salesforce’s Marc Benioff and Microsoft’s Satya Nadella have been engaged in a sometimes not so subtle war of words over the future of software. Benioff, in a direct jab, mocked Microsoft’s AI-powered Copilots by calling them a modern-day “Clippy.”
[Benioff insults Microsoft Copilot as “Clippy.]
Nadella responded by suggesting that traditional SaaS applications—like those Salesforce offers—will eventually be overtaken by intelligent agents, capable of composing outcomes directly from data rather than relying on established SaaS platforms.
[Listen to Nadella de-position SaaS and somewhat contradict himself.]
This back-and-forth highlights contrasting philosophies about how AI will reshape enterprise software. Benioff’s comments suggest skepticism of Microsoft’s approach, while Nadella’s vision points toward a world where AI agents interface directly with CRUD databases (Create, Read, Update, Delete) rather than traditional APIs, rendering SaaS apps less valuable. But later in the above clip he suggests that agents will need to access backend SaaS application logic via APIs.
When we step beyond the public trolling, however, both CEOs’ statements contain partial truths. There are some points of agreement, as well as significant gaps and challenges that each viewpoint overlooks. By exploring these perspectives through their recent statements and our own analysis, we can better understand the evolving role of AI agents in the enterprise landscape.
The Evolution of Enterprise Software: From Rigid Systems to Intelligent Orchestration
Processes as a Malleable Asset
We believe enterprise software technology stands on the brink of a profound shift—from rigid, linear value chains to dynamic, end-to-end platforms that orchestrate entire networks of activities. In this note, we’re going to unpack how new foundational technologies, including digital twins powered by knowledge graphs, are delivering on the vision we set up at the top of this post – i.e. making processes malleable building blocks rather than hardwired code. In our opinion, this represents a critical turning point in how companies organize, manage, and ultimately transform themselves.
In the graphic above, on the left side, you’ve got this traditional source>make>deliver value chain view of how enterprises work. And that was really embodied in the ERP/CRM era of enterprise applications. Enterprises earned profit by adding value in a linear sequence of steps, and it was very much like a mechanical device.
Then we saw the rise in the last 20, 25 years of the consumer online platforms that were just matching a consumer and a resource. But Amazon was the first one to take that and go beyond just simple matching. They built an incredibly sophisticated operational capability where there’s a whole network of activities that data and AI can orchestrate.
And the key point is that for 50 years, applications really have treated data as a corporate asset, but the processes were entangled or dare we say even embalmed in procedural code stuck within those apps. As such, applications were not that flexible. Moving to a platform world means treating the processes, in addition to the data, as assets so that the processes can be reconfigured and the platform can be much more dynamic in how it takes shape.
In the future we’re projecting, you can apply analytics to processes in addition to data. That’s the big change.
Value is Migrating “Above the Ice”
In our opinion, the enterprise software landscape is undergoing a significant split, creating what we’re calling a new fault line. On one side, organizations are increasingly focused on learning directly from business data or building core infrastructure. This dynamic is illustrated below in a slide referencing “James Bond: No Time to Die,” which uses the metaphor of “above the ice” and “below the ice.” The data suggests that the higher-value activities are now migrating “above the ice,” into an abstraction layer where real innovation and value capture occur—much like how virtualization once commoditized physical hardware, pushing strategic value further up the stack.
Our premise is that large language models and generative AI tools are accelerating this shift by commoditizing traditional, algorithmic software development. The new locus of value lies in shaping structures akin to “digital twins” that learn from operational data. Technically, this might manifest as knowledge graphs or other representations, but the essential idea is that businesses can now learn and adapt processes previously too complex to encode through rigid rules.
Meanwhile, the “below the ice” segment remains focused on hardware-level concerns and raw data handling—what some refer to as “strings” (that databases understand), not “things” in the real world. This distinction highlights the emerging bifurcation: one part of the industry focuses on intelligent, data-driven insights, while the other remains anchored in lower-level technologies.
We believe that time itself—what we describe as the “fourth dimension”—is critical here. Forward-looking companies, including those for example Palantir’s approach, emphasize capturing a business’s kinetics and dynamics to orchestrate operations more effectively. Rather than treating the enterprise as a static, mechanical entity optimized solely for efficiency, the goal is to evolve it into an intelligent, emergent system capable of continuous learning and adaptation.
Overall, our research suggests that this bifurcation is reshaping enterprise software. Traditional development will become increasingly commoditized by AI, while real strategic advantage will likely concentrate in platforms and tools that create living, learning models of the business—those operating “above the ice.”
What Enterprise Platforms Look Like in the 2030s
Let’s examine what the next era of enterprise platforms will look like. We believe the future of enterprise software will enable organizations to build a real time digital representation of their enterprise. An end-to-end digital twin. An organic system represented by an iterative, agentic model that evolves rather than being built strictly top-down or bottom-up.
In our opinion, there’s now a competitive imperative for businesses to unify their data and application estates if they hope to match the agility of tech-first giants like Amazon and Instacart. Agentic UIs and workflows, guided by human oversight, will progressively capture processes as flexible building blocks within the digital twin, creating a trusted source of truth without discarding the traditional infrastructure beneath. Marc Benioff’s dismissal of Microsoft’s approach as “just Clippy” and Satya Nadella’s claim that SaaS layers will vanish, both contain grains of truth, but neither fully captures the need for a new interface—one that merges NLP-driven interaction, human reasoning, and dynamic agent collaboration into a singular, transformative platform.
In the graphic above we envision the future of enterprise software lies in a network of collaborative “worker bee” agents rather than siloed, standalone tools. These agents excel at specific tasks but can also share resources, coordinate with one another, and operate under top-down directives aligned with broader corporate objectives. We believe that this approach will transform how organizations deploy AI by embedding agents into an orchestrated assembly line of processes, rather than treating them as isolated add-ons.
Our belief is that this shift ties directly to the public discourse between Marc Benioff and Satya Nadella regarding AI agents. Despite the excitement—often likened to the Pixar “Up” scenario where dogs go wild at the mention of “squirrel”—the observation we make is that standalone AI technologies will not deliver true process innovation by themselves. Instead, these agents need to connect to a common source of truth, what we refer to as a “4D map” or digital twin, which captures a unified view of an organization’s operations over time.
The concept of building a single, holistic enterprise model has existed for decades, but what’s different now is the role AI can play. Agents can learn from data and then, under human supervision, convert their newfound knowledge into code that expands this digital twin. In our view, this iterative approach allows businesses to evolve a living model that is neither fully top-down nor bottom-up but instead strikes a balance between centralized metrics and emergent insights.
Our premise implies that enterprises must unify their data and break down application silos to match the agility of digital-native examples. This means adopting agent-driven user interfaces, guided by human oversight, that progressively accumulate business knowledge. We believe this trend reflects the debate between Benioff and Nadella: while Nadella emphasizes a future where AI redefines the user interface and business logic, Benioff underscores the importance of orchestrating underlying processes—both viewpoints contain elements of truth.
Ultimately, organizations require a unified perspective to address four key questions:
- What happened?
- Why did it happen?
- What’s likely to happen next?
- And how should we respond?
Many companies can tackle the first two questions with current data warehouse or lake technologies. However, to forecast future scenarios and determine the best course of action, we believe that they will need an integrated enterprise model supported by AI agents and human supervision. In our opinion, this harmonious blend of agents, data, and oversight will shape the next era of enterprise software.
AI Automation is Becoming Increasingly Fragmented
Below is a graphic of an automation map from Insight Partners. As you can see the picture is fragmented. We believe that while organizations are rushing to adopt AI agents, most are simply deploying them within existing application silos—what we consider Phase 1. These siloed agents may offer incremental benefits, but they don’t address the broader opportunity to truly integrate and streamline business operations. In our opinion, agents are essentially the machine tools of modern enterprises. To realize their full potential, companies must build the equivalent of an assembly line. That means undertaking the kind of process reengineering that every disruptive technology inevitably requires, paving the way for genuine transformational change.
In our opinion, the enterprise technology landscape is poised to see a surge in co-pilots for retrieval-augmented generation (RAG) solutions, reminiscent of the rush of the past two years to adopt generative AI. As indicated, organizations may uncover pockets of value by deploying individual or domain-specific agents. However, a breakthrough on par with Ford’s assembly line will only occur once these agents are orchestrated within a broader framework. Aligning their activities with corporate objectives—and leveraging analytics to track progress—ensures they operate cohesively rather than in isolated silos.
We believe that, in parallel with agent-focused initiatives, businesses will increasingly prioritize harmonizing both their data and the underlying process definitions. This shift is critical for unlocking foundational value, as an integrated data and application estate allows for more strategic insights and decision-making. These agents will function much like modern “machine tools” for the enterprise. Yet to fully capitalize on them, companies must create the equivalent of an assembly line—reengineering processes instead of merely “paving the cow path.” True transformation, in our view, demands reevaluating existing workflows so that each disruptive technology can drive meaningful, lasting change.
Dissecting Microsoft’s Approach to Agentic AI
Let’s now dig into Microsoft’s approach. The work that we’ve done shows that while Microsoft has achieved notable strides in agentic UIs and productivity tools, it hasn’t yet delivered the end-to-end integration needed for true business process orchestration. Nadella’s vision of agents bypassing traditional SaaS application logic speaks to a future where CRUD databases lie beneath an agent layer, but we believe that alone won’t harmonize processes across multiple applications. In our view, today at least, Microsoft lacks a unified business process model, which limits its ability to tie everything together—APIs are useful, but they’re only part of the solution.
Microsoft GPU Constraints Drove Key Decisions
Moreover, our research indicates that Microsoft’s early work on agent-based user interfaces and tools was constrained by a GPU shortage—due in part to the rapid rise of ChatGPT usage. This limitation forced the company to prioritize between Bing, Office, and Azure. Our intelligence indicates that Microsoft chose Bing first to compel Google into integrating generative AI into search, thereby impacting search profitability. They then focused on Office, although the rationale for prioritizing it remains less clear.
The observation we make is that Microsoft’s existing framework, Microsoft Graph, is oriented toward productivity and collaboration rather than business processes. Consequently, the 365 Copilot initiative has taken longer to mature because many of its core use cases remain underdeveloped. We believe Microsoft also complicated matters by restricting out-of-the-box functionality largely to its own applications. Third-party integration requires extensive customization, a gap that competitors like Glean have addressed more effectively by launching multi-vendor capabilities from the start.
In our view, the absence of a “harmonized business process model” is evident. Microsoft relies on over a thousand connectors to interface with external applications, each speaking a different “language.” This patchwork makes it difficult for agents to perceive a unified state of the business. While the Power Platform and its connectors enable integration at the API level, they do not consolidate the underlying processes or data semantics into a coherent whole.
We believe Microsoft’s decision to emphasize Office may reflect the company’s heritage and marketing priorities; it is also a way to showcase AI-driven features to a broad user base. However, we see a future where a Copilot-like interface will be necessary for enterprises running thousands of agents—providing a consistent user experience and capturing the human interactions that typical enterprise applications overlook. Over time, we anticipate that Microsoft Graph’s productivity and collaboration data will become increasingly valuable for orchestrating both people and agents across an organization. Yet, in our opinion, focusing first on backend business processes might have offered a more direct path to transforming enterprise software at scale.
Comparing Salesforce’s Approach to Agentic AI
Let’s now dig into the Salesforce approach. We believe Salesforce is making significant strides in agent-based data and application pipelines—harnessing tools like MuleSoft to automate how data is collected and organized. In our opinion, the company’s new data cloud and semantic layer point toward a future where agents collaborate easily across customer activities, fueled by metrics that feed back into decision-making. The key question is can Salesforce extend beyond its CRM roots to become a true hyperscaler platform that is software-only, pulling in data and processes from every corner of the enterprise?
The simple answer is ‘yes.’ In our opinion, Salesforce is positioning itself to become the core application development layer for intelligent data applications, running on AWS infrastructure. While AWS excels at hardware, Salesforce appears to be transitioning from a pure application provider into an application platform and tools company—one that still offers its own applications as modular building blocks. Our research indicates that Salesforce’s Data Cloud serves as the foundation for customer-centric semantics, capturing not only customer attributes but also the full customer journey. Rather than confining data to Salesforce alone, the platform holds metadata in Salesforce Data Cloud while pulling real-time information from systems like Databricks, Snowflake, or Redshift. We believe this approach effectively places Salesforce as a “semantic layer” above various data sources, enabling both analytics and operational decision-making to flow seamlessly into Salesforce’s Customer 360 apps or other applications through MuleSoft.
Salesforce’s success in this arena appears to be reflected in the substantial run rates for both Data Cloud and MuleSoft, underscoring how data pipelines and data platforms naturally align. In our view, this trajectory parallels Microsoft’s emergence with Azure—initially overshadowed by its established Office and server businesses but ultimately becoming a high-growth platform. We believe Salesforce’s Data Cloud and the rise of AI-driven agents could follow a similar path.
Looking ahead, the broader goal for many organizations is to unify networks of agents across departments and functions, rather than running isolated bots. This vision relies on a harmonized data model—complete with business logic and analytics—that is guided by human oversight. We believe this approach is inherently organic; it evolves through incremental learning and exception handling rather than being rigidly hard-coded. Multiple technologies—from relational data models and process mining to knowledge graphs and digital twins—will converge to support agent-to-human collaboration aligned with an overarching digital business model.
Comparing the Agentic Strategies of Salesforce and Microsoft – Salesforce has the Edge (Today)
When comparing Salesforce and Microsoft, we believe Salesforce holds more of the crucial pieces today. While Microsoft boasts mature low-code and agent-building tools in the Power Platform, it lacks a cohesive digital business model layer. Salesforce, on the other hand, already offers mechanisms for building out a “digital twin” of an enterprise, capturing processes that agents can learn and replicate. In this scenario, governance and top-down guidance remain vital for coherence. Metrics and relationships, currently buried in dashboards, must also be incorporated into a unified data platform so that agents can learn what drives the business, predict outcomes, and adapt accordingly.
We believe this integrated management system—comprising agents, data, analytics, and human input—represents the future of enterprise software. While vendors often tout individual components, interoperability challenges remain. For many mainstream enterprises, a single-vendor solution that provides an end-to-end platform may simplify the path toward this new model. Today, Salesforce appears to lead in that direction, though debates continue about best-of-breed versus integrated approaches. Ultimately, our research suggests that a truly transformative environment will hinge on harmonized data, orchestrated processes, and a dynamic interplay between humans and AI agents.
From Web 2.0 to Enterprise 3.0 – A Five to Ten Year Transition
In our opinion, enterprise applications stand on the cusp of a profound transformation, evolving from rigid, linear value chains into end-to-end platforms that orchestrate entire networks of activities. Our premise suggests that digital twins, supported by knowledge graphs, will liberate business processes from static application code, transforming them into flexible, fungible building blocks. In our view, this foundational shift sets the stage for organizations to fundamentally reimagine how they operate.
We believe this transition will likely unfold over the next five to ten years. For those skeptical about its impact, we suggest looking no further than Amazon. Originally just an online bookseller, Amazon built a comprehensive platform for e-commerce and fulfillment, treating business processes—rather than just data—as valuable, reusable assets. Over time, this approach enabled Amazon to expand into advertising, media, groceries, healthcare, satellite operations, and more. Tech-centric competitors adopting similar platform-based strategies are poised to challenge mainstream companies that have not yet re-engineered and harmonized their operations.
Salesforce and Microsoft exemplify the emerging fault line in this market. While Salesforce appears to have an edge in developing customer-centric processes, Microsoft’s strengths in low-code tools and broad enterprise reach ensure it remains a formidable player. Both could enhance their platforms through process mining solutions like Celonis, possibly UiPath, or more configurable, data-centric approaches from firms like RelationalAI. Our research and analysis suggests that these capabilities will enable vendors to unify business logic and analytics, making processes as adaptable as the data itself.
In our opinion, this is merely the opening round of an extended “rock ’em sock ’em” bout between industry titans. The dual-disruption model—where major tech companies not only transform IT but also disrupt multiple other sectors—demonstrates the advantages of a data-first strategy that transcends traditional industry boundaries. We believe organizations that fail to adopt such a platform mindset risk being outmaneuvered by competitors that have embraced these new, fluid approaches to enterprise software.
A New Enterprise Operational Model
In our opinion, the enterprise landscape is undergoing a profound shift—from linear value chains to holistic platforms—fundamentally altering how organizations structure themselves and evolve. We believe this new model rejects rigidly hardcoded processes, even those built on microservices, in favor of dynamic orchestration. Our analysis suggests that digital representations of an enterprise, often enabled by knowledge graphs, will be key to realizing these more flexible, adaptable processes.
Firms such as Celonis, UiPath, RelationalAI, Neo4j, application providers like Oracle, ServiceNow, Workday and others, will provide orchestration capabilities that convert business processes into reusable, composable assets. Instead of locking operations into fossilized code, these technologies render them malleable and fungible. In our view, this flexibility will ultimately fuel new forms of networks and network effects, driving the next wave of productivity gains across industries.
Looking ahead, we anticipate a “blast radius” of disruption far larger than what accompanied the rise of consumer tech giants. The critical question, in our opinion, is whether mainstream companies can adapt quickly enough to harness these platform-based models, or whether tech-centric organizations will leverage their agility and data-driven orchestration to marginalize traditional incumbents. As digital orchestration increasingly allows asset-light players to coordinate people and resources without owning them outright, the potential for sweeping industry realignments is high. We believe this next generation of enterprise technology will reshape value creation on a scale not seen before.
