The enterprise software market is rapidly reorganizing around a new product category: AI agents capable of performing complex tasks across various business applications much like a human would.
Companies such as OpenAI, Anthropic, and Google are advancing tools that enable the creation of “digital employees”—autonomous agents that operate within browsers, desktop applications, and enterprise services through interfaces, APIs, or direct screen control. At the same time, traditional players like Microsoft, Salesforce, ServiceNow, and Snowflake are actively launching their own platforms for developing and managing these agents, aiming to secure a new architectural role within the software ecosystem.
In essence, multiple competing layers are emerging: tools for building agents, the agents themselves (including browser-based and desktop agents), and a distinct category of management platforms, which companies refer to as “agent control panels.”
Examples include browser agents that can independently navigate websites and perform multi-step operations such as placing orders; desktop agents that work with local applications and files to prepare reports; and enterprise platforms like Salesforce Agentforce and Google Gemini Enterprise, which allow businesses to assemble custom sets of agents tailored to specific needs.
Image generated by: Nano Banana
Separately, the market for “agent control panels” is developing, with Microsoft Agent 365 and OpenAI Frontier competing as systems designed to unify and coordinate agents from various vendors within a single organization.
Against this backdrop, a key architectural question is intensifying: how many such management systems does a business actually need? Theoretically, a single system could suffice for any company, but the market already offers dozens of incompatible solutions.
Interest in this area is growing amid the strategic bets placed by major technology companies on a future where employees no longer interact directly with business applications. Instead, they will oversee a set of AI agents that themselves engage with CRM, ERP, and other corporate systems.
Part of the industry is already publicly discussing a scenario where traditional applications might “shrink” to the level of databases with business logic, with agents serving as the interface. This includes considering monetization models where access to corporate systems is billed not per user but per agent usage.
However, transitioning to such a model faces serious limitations. Current browser and computer agents remain vulnerable from a security standpoint: they may inadvertently expose credentials or create new attack surfaces. Additionally, many enterprise clients note that existing solutions are difficult to configure and operate.
There is also a positioning gap: some companies (e.g., OpenAI and Anthropic) release these agents as research previews, emphasizing their experimental nature, while other vendors, including SAP and ServiceNow, claim their solutions are ready for industrial deployment.
Meanwhile, competition for access to enterprise data is intensifying. Some companies have already restricted the ability of third-party AI systems to interact with internal communications or knowledge bases, as seen with previous limitations on access to Slack data.
A market paradox here is that major enterprise software providers are simultaneously both clients and competitors of AI companies: many agent systems are already built on models from OpenAI and Anthropic, including solutions for analytics, data search, and business process automation. For instance, Snowflake uses AI company models to create agents that work with corporate data in storage systems and integrated applications like Microsoft Teams, Salesforce, or SAP.
Thus, the industry is simultaneously constructing competing and interdependent ecosystems, where the same models underpin products that compete at the interface and data control levels.
Industry experts describe the current state as a transitional phase, where companies have yet to settle on the architecture of the future enterprise AI stack—how many agents there will be, who will manage them, and how control over corporate information will be distributed.
Despite the market’s active growth, enterprise customers remain cautious: adopting agent systems requires lengthy testing and integration cycles. Some companies have already noted that real-world implementations of such solutions take years and demand complex architectures for data and access coordination.
As a result, the market is simultaneously moving toward accelerated automation and a new level of architectural complexity, where AI agents become not just a tool but an intermediary layer between users and all enterprise applications.
About the Author
