Going from two to three: how enterprise apps are making space for AI

(© canva.com) For decades, business software has been built around a relatively simple relationship. A person has a job to do, and an application provides the data, workflows, rules, and transactions needed to get it done. Artificial Intelligence (AI) does not replace either side of that relationship, despite much of the discussion about autonomous agents suggesting that it might. Instead, it introduces a third participant with a different role to play. Traditional enterprise applications are designed around defined data, rules, and workflows. AI introduces a different capability — it can interpret intent, reason across information, choose among possible actions, and adapt as circumstances change. That changes not only what enterprise applications can do, but also the role people play in directing and governing them. People bring judgment, intent, and, crucially, accountability. Applications provide trusted business context — transactions, permissions, policies, workflows, and controls. AI can work across that context — reasoning about it, determining next-best actions, co-ordinating work, and, when permitted, taking action. Governance defines the division of labor Businesses need to decide what an AI agent can do autonomously, what requires human approval, and what it should never do. Its actions should be visible and auditable, with people remaining accountable for the decisions and outcomes that matter. A customer recently described the division of labor as the company automates portions of revenue reporting that previously required manual preparation. Sloan Session, CFO of Dura Software, described how AI can gather and organize information, while people remain responsible for the judgment that follows, saying: > The agents handle the pull. The humans handle the judgment and the personal touch. For example, a grants manager can ask AI to identify expenses that may be billed against a specific government grant. The application provides the relevant transactions, grant terms, and approval history — AI evaluates that context and presents the eligible expenses for the manager to review and decide how to proceed. This division of responsibility creates a useful test for whether an application is genuinely AI-native. Simply attaching a chatbot to existing software does not fundamentally change how work happens. If users have to leave their workflow, explain the business context to an AI tool, evaluate its answer, return to the application, and manually execute the recommendation, then the work has not materially changed despite the technology. AI-native apps bring the three roles together The more interesting model emerges when the roles of people, applications, and AI are designed together from the outset. Each has a clearly defined role to play, and work can move naturally among them. The user defines the outcome and handles decisions that require judgment. The application supplies the trusted operational context and enforces business controls. The agent interprets, reasons, coordinates, and acts within the boundaries the business has defined. The result is less effort spent moving information between systems and more time focused on the work that requires judgment. Peter Bernat, enterprise systems leader at KELTEC, explains how employees at the industrial filtration and component solutions provider have benefited from AI embedded in the company’s NetSuite ERP system: > The best AI is the one that’s already in your workflow. NetSuite Next puts AI at the center and makes it easy to use from day one, which is a really big deal. It didn’t feel like we had to spend weeks getting it ready. We could start seeing value almost immediately, AI gives users more power to shape the software There is another implication that may prove just as significant — AI can change not only how people use an application, but how they shape it. Historically, organizations have had to adapt their processes to the structure of business software or rely on administrators and developers to configure and extend that software. AI creates the possibility of shifting more of that relationship in the other direction. A user might describe a desired outcome in natural language and have AI help configure a workflow, change a preference, create a report, or build an extension. Instead of first learning how the software needs to be configured, the user can increasingly start by describing what the business needs to accomplish. For example, a user can ask AI to set up email-sorting rules, so messages automatically move into the right folders. The application supplies the user’s existing folders, senders, and rules — AI translates the request into proposed settings, and the user reviews and applies them. The same rules still apply. AI-generated changes should operate within appropriate guardrails, and material changes to business processes should occur with the user’s knowledge and consent. The goal is software that can adapt more readily to the organization while maintaining the controls enterprise systems require. This is why the arrival of agents should prompt a broader rethink of enterprise application design. Much of the current discussion about AI focuses on what it will replace — applications, interfaces, or even users. That framing misses the more immediate change taking place. The more useful question is how people, applications, and AI can work together, with each taking on the responsibilities it is best suited to handle. The applications that get that relationship right will make it easier to move from intent to action without losing the accountability, governance, and trust that businesses require.