For years, the enterprise software industry sold digital transformation through dashboards, workflows, and chatbots. The chatbot answered questions. The workflow routed approvals. The dashboard showed what had already happened.
That era is now being challenged by a more ambitious model: AI agents.
An AI agent is not merely a conversational assistant. It is software designed to pursue a goal, reason through steps, use tools, access enterprise data, and complete tasks on behalf of a user or team. Google Cloud describes AI agents as systems that use reasoning, planning, memory, and autonomy to complete tasks for users. OpenAI’s developer documentation similarly frames agents as applications that can plan, call tools, collaborate across specialist agents, and maintain enough state to complete multi-step work.
The enterprise battle is no longer about who has the smartest chatbot. It is about who controls the agent layer between people, data, workflows, and business outcomes.
This is why the next software war has begun.
Microsoft, Salesforce, Google, ServiceNow, UiPath, Oracle, OpenAI, Anthropic, and a growing field of startups are racing to define how work gets done when software no longer waits for clicks. The old SaaS model asked users to log in, search, fill forms, move tickets, update records, and trigger workflows. The agentic model asks a different question: why should humans operate software when software can operate software?
Microsoft has been explicit about this shift. At Build 2025, the company declared that “the age of AI agents” had arrived, pointing to advances in reasoning and memory and highlighting agentic capabilities across GitHub Copilot and Microsoft’s broader developer ecosystem. Microsoft has also positioned Copilot Studio as a way for business and technical users to build custom agents connected to enterprise data and workflows.
Salesforce is pushing the same battlefield from the CRM side. Its Agentforce platform is marketed as an enterprise agentic platform to build, deploy, and manage autonomous agents across business functions. Salesforce says Agentforce agents can retrieve data, create action plans, and execute tasks within organizational guardrails.
ServiceNow is turning the fight toward enterprise workflows. In January 2025, it announced AI Agent Orchestrator and AI Agent Studio for the ServiceNow Platform, followed by its Yokohama release, which added preconfigured AI agents across CRM, HR, IT, and other workflows.
Google is building from the cloud and productivity angle. Its Gemini Enterprise Agent Platform is described as a platform for developers to build, scale, govern, and optimize enterprise-ready agents, while Gemini Enterprise is positioned as a secure environment where employees can discover, create, share, and run agents for workflows.
UiPath, once strongly associated with robotic process automation, is also repositioning around agentic automation. In April 2025, it launched a platform designed to unify AI agents, robots, and people on one intelligent system. Its Agent Builder focuses on creating and deploying agents for complex processes such as invoice dispute resolution.
The chatbot answered. The copilot assisted. The agent acts. That single difference is why every major enterprise software company is now rewriting its product strategy.
The market signals explain the urgency. Gartner predicted that 40% of enterprise applications would feature task-specific AI agents by 2026, up from less than 5% in 2025. In another forecast, Gartner said 33% of enterprise software applications would include agentic AI by 2028, up from less than 1% in 2024, and that at least 15% of day-to-day work decisions would be made autonomously through agentic AI by 2028.
The numbers are even more striking in specialized markets. Gartner forecast that supply chain management software with agentic AI capabilities would grow from less than $2 billion in 2025 to $53 billion in spending by 2030.
This is not just another productivity feature cycle. It is a structural shift in enterprise software economics.
In the traditional SaaS model, value was measured by seats, modules, licenses, and user adoption. In the AI agent model, value may increasingly be measured by completed work: tickets resolved, claims processed, invoices reconciled, code reviewed, leads qualified, procurement requests completed, compliance checks performed, and customer issues closed without human handoffs.
That threatens and strengthens SaaS at the same time.
It threatens SaaS because agents can sit above applications and reduce the need for users to manually navigate dozens of systems. If an employee can ask an agent to “prepare the vendor risk summary, check contract exposure, update the procurement record, and notify legal,” the employee may not care which screens or modules the agent touched.
But it strengthens SaaS vendors that already own enterprise data, permissions, workflows, and customer trust. Salesforce has CRM data. ServiceNow has workflow data. Microsoft has productivity, identity, and collaboration data. Google has cloud, Workspace, and search-like enterprise discovery. Oracle and SAP have deep systems of record. UiPath has automation and process execution heritage. The winners will be the companies that can safely connect AI reasoning with enterprise action.
In enterprise AI, the model is only the engine. The real moat is permissioned data, workflow context, governance, integration, observability, and trust.
That is why interoperability has become a central front in the war. The Model Context Protocol, originally introduced by Anthropic, is gaining attention as a way for agents to connect with external tools and systems. Microsoft has supported MCP as part of its vision for an “open agentic web,” and Zendesk recently announced MCP adoption to connect AI agents with support infrastructure such as tickets, knowledge bases, and customer information.
OpenAI has also moved from chat toward action. In March 2025, it introduced tools for building agents, including the Responses API and Agents SDK. In July 2025, it introduced ChatGPT agent, describing it as a system that can think and act using tools to complete tasks with user guidance.
The enterprise implications are massive. A customer service agent can do more than suggest a response; it can check account history, validate entitlement, issue a refund within policy, update the CRM, and close the ticket. A finance agent can inspect invoices, match purchase orders, detect anomalies, ask for missing documents, and route exceptions. An HR agent can answer policy questions, initiate onboarding steps, provision access, and escalate sensitive cases. An IT agent can diagnose incidents, search logs, create remediation plans, and coordinate with automation tools.
But the excitement comes with serious caution.
Gartner has also warned that more than 40% of agentic AI projects may be canceled by the end of 2027 due to factors such as escalating costs, unclear business value, or inadequate risk controls. That warning matters because agents introduce a different class of risk from chatbots: they do not merely generate text; they can initiate actions.
Microsoft has highlighted the governance challenge directly, arguing that every agent needs an identity and that organizations must know who is accountable for its behavior. This is a crucial enterprise point. If an AI agent approves a refund, changes a customer record, sends an email, modifies code, or triggers a payment workflow, the business must know who authorized it, what data it used, what policy it followed, and how to reverse or audit the action.
The winning enterprise agent will not be the one that sounds most human. It will be the one that behaves most responsibly inside real business constraints.
That is where the next layer of competition will intensify: agent identity, access control, audit trails, human-in-the-loop approvals, simulation, rollback, monitoring, data boundaries, and cost management. Enterprises will not deploy autonomous agents broadly simply because they are impressive in demos. They will deploy them when they are measurable, controllable, secure, and accountable.
The first phase of generative AI was dominated by prompts and chat interfaces. The second phase was copilots embedded into productivity and developer tools. The third phase, now emerging, is agentic enterprise software: AI systems that can coordinate work across applications, departments, and business processes.
For software vendors, this is a land grab. For enterprises, it is a strategic architecture decision. Choosing an AI agent platform may become as consequential as choosing an ERP, CRM, cloud provider, or workflow platform. It will determine where business logic lives, how data moves, how work is automated, and which vendor becomes the operating layer for digital labor.
The war has begun because the prize is enormous: control over the future interface of work.
The companies that win will not merely offer better chat. They will offer trusted execution. They will connect fragmented enterprise systems, understand business context, enforce governance, and complete meaningful work with measurable outcomes.
The companies that lose may still have good software, but their applications could become invisible plumbing beneath someone else’s agent layer.
The next enterprise software giant may not be the company with the most screens. It may be the company whose agents get the most work done.
