Software Incumbents Under Siege
For the better part of three decades, enterprise software followed a remarkably stable economic logic. You built a product. You sold access to that product. You charged per seat. You expanded revenue by increasing the number of people required to operate the system.
It was elegant, scalable, and wildly profitable. Now, it is breaking.
The recent drawdown in software equities — what some have labeled the SaaSpocalypse, where approximately $2 trillion in market capitalization evaporated in a matter of weeks between January and February 2026 — is not a cyclical correction or a repricing of growth multiples.
It is the decoupling of software revenue from human labor.
I. The Original Solution
The first era of enterprise software solved a clear constraint: human limitation. Humans are slow, forgetful, and do not scale.
Software abstracted that friction. CRM systems captured memory. ERP systems are structured operations. Collaboration tools organized coordination. But critically, software did not eliminate human work — it required it. Every workflow still depended on human input, human navigation, and human judgment.
The software layer became a refinery: taking raw organizational activity and converting it into structured, legible output, capturing margin at the interface, workflow, and process design.
AI removes the human bottleneck. The economic architecture built around it goes with it.
II. From Interface to Outcome
The most important shift underway is not that AI enhances software. It is that AI changes what software is.
Historically, software was defined by its interface. Dashboards mattered. Forms mattered. User experience mattered — because the human sat at the center of execution. AI displaces that center.
An agent does not need a dashboard. It does not need a carefully designed workflow. It needs context, access, instructions, and feedback.
Which leads to a simple but profound conclusion: the interface is no longer the product. The outcome is.
This collapses entire categories of software — not gradually, but structurally. If your product exists to help a human perform a task, AI will increasingly perform that task instead, and the economic model built around that human disappears with it.
The SaaSpocalypse was not an irrational panic. It was the market beginning to price this logic. Companies like Atlassian saw enterprise seat counts decline for the first time in company history. Salesforce fell 28% despite revenue growth because investors shifted focus from top-line momentum to the long-term structural question: what does this business look like when agents do the work?
That question does not have a comforting answer for every incumbent. But it does have a differentiated one.
III. Agent-Centric Stacks
The refinery analogy — raw intelligence in, refined output out — is useful as far as it goes. But it understates the magnitude of what is happening.
Refineries imply linearity: input, processing, output. AI systems are not linear. They are adaptive, recursive, and continuously learning. A better analogy is a power grid. In a grid, value accrues to nodes that control flow. Scale compounds advantage. Centrality becomes power. Intermediaries that do not control the flow become irrelevant.
This is the shift now confronting software incumbents. A superior product is not simply disrupting them. They are being repositioned because the value stack has been redrawn.
THE OLD STACK (HUMAN-CENTRIC)
Interface and user experience sat at the top.
Below it: workflow logic, the application layer, the database, and infrastructure. The human operated at the apex, and every layer was built to serve that operator.
THE NEW STACK (AGENT-CENTRIC)
The outcome layer sits at the top — what gets done.
Below it: agent orchestration, context and memory systems, the model layer, and compute. The interface has been removed as a primary value layer. Orchestration and context have moved up the stack. This is where strategic control will concentrate.
Two changes matter most. First, the interface disappears as the primary value layer. Second, orchestration and context move up the stack. The companies that understand this early enough to reposition before competitors force them to will capture the next era’s margins.
Those that protect their interface-layer revenue by adding AI features to existing workflows will discover that they have preserved the appearance of a moat while the underlying ground shifts beneath it.
IV. The Incumbents
Not all software companies face the same future. The difference lies in where their value actually resides — and that distinction determines whether AI is primarily a threat, an opportunity, or a genuine existential inflection point.
Systems of Record
Companies that own accumulated institutional data occupy a structurally advantaged position. Salesforce holds customer relationships, pipeline data, and interaction history. SAP holds financials, supply chains, and operational processes. This is not simply data. It is institutional memory. Replacing it is not a technical decision. It is an organizational risk event.
That creates durability. An AI agent replacing a human sales representative still needs CRM data. An AI agent automating procurement still needs ERP context. AI increases the importance of systems of record.
But durability is not economic immunity. The risk these companies face is not replacement. It is disintermediation.
If agents interact directly with the data layer, the application becomes a database with APIs rather than a differentiated product. Margins compress. Control shifts upward to whoever owns the orchestration layer. The strategic imperative is to reframe from a software vendor to a data infrastructure.
Workflow Software
The most exposed category is workflow and collaboration software. Ticketing systems, documentation tools, sprint planning interfaces — these are not systems of record. They are systems of coordination, and this is what AI agents perform well.
The mistake many companies in this category will make is treating AI as a feature — adding copilots, layering automation onto existing workflows, building intelligence into interfaces that are themselves under threat. AI is not a feature. It is the execution layer. If your revenue depends on human interaction with your system, you are competing against a system that eliminates that interaction. That is not a favorable position from which to add a feature.
Integration and Orchestration: The Emerging Control Layer
The most interesting — and least appreciated — category is integration and orchestration. Historically, this was plumbing: APIs, connectors, middleware—necessary infrastructure, but not where strategic value is concentrated.
AI changes that. In an agent-driven world, systems must communicate continuously. Workflows span multiple domains. Execution is distributed. Which means the layer that decides which agent runs, which model is used, which system is accessed, and how outcomes are verified becomes the control plane — and control planes capture value. This is where new platforms will emerge. It is also where incumbents have their clearest strategic opportunity if they move decisively rather than defensively.
V. Three Companies, Three Fates
Salesforce, SAP, and Atlassian represent three distinct positions within the incumbent landscape — and three distinct responses to the same structural pressure.
Salesforce: The Data Moat vs. the Seat Model
Salesforce sits at the fault line between the two outcomes. Its per-seat pricing model — the mechanism that generated over $35 billion in annual revenue — is challenged by AI agents that can perform data entry, pipeline management, customer logging, and activity tracking without human operators requiring individual licenses. Salesforce’s stock fell 28% in early 2026 as investors recognized this structural threat.
The panic partially missed the point. Salesforce’s durable value is not the seat. It is the CRM data itself — the accumulated record of customer relationships, sales history, interaction logs, and pipeline intelligence that organizations have spent decades building inside Salesforce. An AI agent replacing a human sales rep still needs to read from and write to that data. As long as Salesforce owns the system of record, it remains indispensable — not as a software interface, but as a data infrastructure layer.
Salesforce’s Agentforce initiative is the company’s bet that it can shift from a platform where humans do the work to one where AI agents do the work on top of Salesforce’s data and infrastructure. The pricing model will need to shift from per-seat to per-outcome or per-agent — a painful near-term margin compression that may preserve the long-term moat.
Whether management executes the transition with sufficient urgency before the market forces it is the question.
SAP: The ERP Moat and the Integration Trap
SAP occupies a more defensible position than Salesforce because the switching costs are existential. ERP systems sit at the core of how large organizations operate — financials, supply chain, procurement, HR, and manufacturing. The data inside SAP is the enterprise’s operational nervous system. Replacing it is not a software decision. It is a business transformation project that takes years, costs hundreds of millions, and carries operational risk that no board approves lightly.
This makes SAP categorically different from pure workflow applications. AI agents automate tasks within SAP — invoice processing, procurement workflows, compliance reporting — but do so by integrating with SAP’s data layer rather than displacing it. The ERP system remains the system of record through which AI must operate. SAP’s RISE program, moving enterprises to cloud-based SAP infrastructure, positions the company to embed AI capabilities into the ERP layer itself and charge for the intelligence that flows through it. This is refinery economics, not interface economics.
The real vulnerability for SAP is not AI agents disrupting its enterprise base. It is an AI-enabled simplification that creates a viable path for mid-market companies to consider alternatives they previously could not. When AI handles this complexity, simpler AI-native ERP competitors become competitive where they previously were not.
SAP’s moat among global corporations is durable. Its mid-market position deserves more scrutiny.
Atlassian: The Collaboration Layer at Risk
Atlassian is the most vulnerable. Jira and Confluence occupy the collaboration and project management layer — precisely the category of work that AI agents automate most readily. Ticket creation, status updates, sprint planning, documentation: these are the workflows that agentic AI handles well, quickly, and cheaply.
The market delivered an unambiguous signal. Atlassian’s stock fell 35% in early 2026 when enterprise seat counts declined for the first time in company history. Seat count decline is not a pricing story. It is a substitution story. If AI-native alternatives emerge that offer better workflow automation at lower cost, Atlassian’s position is not defensible.
The most credible path for Atlassian is the same as Salesforce: reposition from a software interface to an AI orchestration layer that coordinates agents across the development workflow. The company needs to move faster.
Atlassian’s value lies in the connective tissue it provides across teams and tools — not in the screens through which humans currently enter data.
VI. The Shrinking Firm
Most analysis of AI and enterprise software focuses on product disruption. The deeper impact is organizational.
Enterprise software exists largely to manage complexity: coordination, communication, oversight, and compliance. AI reduces that complexity. Fewer people are required to execute the same workflows. Decision-making accelerates. Information asymmetry declines. Smaller teams can produce enterprise-scale output.
If that is true — and the early evidence suggests it is — then total demand for certain categories of enterprise software may decline even as compute demand explodes. AI expands the infrastructure layer while compressing parts of the application layer. The two trends are not in contradiction. They are the same transition viewed from different vantage points.
This paradox has not yet been fully resolved. Revenue growth at many SaaS companies continues — because AI expands the surface area of work and creates new demand. But the marginal economics of that work are shifting. The old pricing models were calibrated to a world of human-operated workflows. The new pricing models will be calibrated to agent-operated outcomes.
The transition between these two worlds is where the greatest valuation compression occurs, and also where the greatest opportunity to rebuild defensible positions emerges.
VII. Business Model Compression
The software industry is moving through a pricing transition that most incumbents are treating as optional. It is not.
The first era was seat-based: revenue scaled with headcount. The second era is consumption-based: revenue scales with usage. The third era is outcome-based: revenue scales with results delivered.
The industry is moving, unevenly, through the second phase. The third phase is uncomfortable precisely because outcomes are measurable. If an AI agent completes a workflow faster and more cheaply, pricing pressure increases, differentiation becomes more transparent, and margins compress. It is a structural repricing of the value proposition that no longer aligns with the new cost structure.
The companies that initiate this transition on their own terms — repricing before market forces force them to — will preserve more margin than those who hold the old model until it collapses. Near-term revenue compression will be real. The alternative is worse.
VIII. The New Moats
If interfaces and workflows are no longer defensible, what is?
Proprietary, high-quality data is the first and most durable moat — not volume alone, but structure, cleanliness, and domain relevance. The accumulated context within systems of record is not easily replicated or displaced. This is the asset that Salesforce and SAP hold, and it is the reason their long-term positions are more defensible than their near-term stock performance suggests.
Workflow depth is the second moat — how embedded you are in mission-critical processes. Products that have become part of an enterprise’s operating system, connected to compliance requirements, audit trails, and organizational memory, carry switching costs that no benchmark comparison can fully capture.
Orchestration control is the third, and the most contested. Who decides which agent runs, which model is used, which system is accessed, and how outcomes are verified? That layer becomes the control plane of the AI-driven enterprise. The companies that own it will extract the most durable margins. This is where new platforms will emerge and where incumbents have their clearest offensive opportunity.
Trust and governance — security, compliance, and auditability — become essential as AI makes decisions that humans made previously. Enterprises deploying AI agents in regulated environments need governance infrastructure as much as they need capability. This is a moat that safety-oriented AI companies have begun to build, and that incumbent software vendors are well positioned to extend.
Distribution, where incumbents retain a structural advantage that newer entrants cannot replicate in the near-term. The question is whether that advantage gets deployed aggressively enough to anchor the transition rather than merely delay it.
IX. The Strategic Imperative
For incumbent software companies, the path forward is difficult, which is why most organizations will resist it until it may be too late.
The product is no longer the screen through which humans interact with data. The product is the outcome enabled by the data. Every strategic roadmap that places interface improvement at the center is building a better horseshoe as the automobile arrives.
Expose systems programmatically for agent access. Systems of record that cannot be accessed, queried, and written to by agents become inaccessible to the workflows that AI is restructuring. The API becomes the primary interface. Design accordingly.
Own or integrate deeply into the orchestration layer. The control plane of the AI-driven enterprise is the most strategically valuable position in the next cycle. Incumbents that move to own this layer before new entrants consolidate it will capture the most durable position. Those who wait will pay a premium or find it inaccessible.
Transition pricing before the market forces it. The shift from per-seat to per-outcome is painful on a short time horizon but necessary. The companies that initiate this transition actively preserve more margin and customer goodwill than those forced into it by declining seat counts and investor pressure.
Collapse product silos into unified execution platforms. The fragmented suite model — separate products for CRM, collaboration, project management, analytics — made sense when humans navigated between them. Agents do not navigate. They need an integrated context. The incumbents that consolidate their platforms into unified execution environments before competitors do will retain the accumulated context advantage that is their most defensible asset.
This requires a willingness to disrupt your own revenue model. Most companies will resist. That resistance will be costly.
X. What Role Do You Play?
The core question for every incumbent is not whether AI will impact the business. It already has. The question is: what role do you play in an agent-driven system?
Systems of record survive, but are repriced. Their value accrues to a different layer of the stack — data infrastructure, not application interface.
Systems of execution must become agent-native. The companies that build or embed themselves in the orchestration layer — becoming the environment in which agents operate rather than the tool humans use — will capture the highest margins.
Coordination systems face the most immediate risk. The workflows they facilitate are exactly the workflows AI agents automate most readily. The path through requires a credible argument that the platform provides connective tissue and accumulated context that no new entrant can replicate on a short timeline.
The industry continues to frame this moment as a competition between AI and software. That framing is wrong.
AI is not competing with software. It is becoming the operating system for work.
The companies that correctly identify which category their core assets fall into — and act accordingly before the window closes — will define the next era of enterprise technology. Those who mistake durability for immunity will discover the difference too late to matter.
