Enterprise software management expands from apps to capabilities

Enterprise software management has long centered on applications. Organizations inventory them, assign owners, manage access, track costs, monitor integrations and handle contracts and renewals. That work still matters.

Embedded AI adds another layer.

An enterprise might know which ERP, HR, CRM or collaboration systems it uses without knowing which AI capabilities are active inside them. It might also lack visibility into who can use those capabilities, what data they can reach, what actions they can take or what they cost.

The application might not be new. The feature might be.

That is why enterprise software management increasingly must manage both applications and the capabilities within them.

One application can contain many AI capabilities

AI no longer arrives only as a separate product that goes through a recognizable purchase, implementation and deployment process.

AI assistants, copilots and agents, along with features for summarization, search, recommendations and automation, increasingly arrive inside applications organizations already use.

One SaaS application can contain multiple embedded AI features with different data-access patterns and behaviors. That limits what an application inventory can tell the enterprise.

Knowing that a company uses an HR system says little about whether an embedded recruiting assistant is active.

The software is visible. The individual capability might not be.

Knowing that a collaboration platform is approved does not tell leaders whether meeting transcription, AI summaries or action-item generation are enabled.

Knowing which CRM the company uses does not show which autonomous customer-service capabilities are in production.

The software is visible. The individual capability might not be.

SaaS releases make the problem continuous

Continuous SaaS releases were already turning application management into an ongoing process rather than an occasional upgrade project.

Vendors ship releases. Enterprises must absorb them.

Routine SaaS releases can change workflows, reports, permissions and user experience.

The vendor can ship the release, but the enterprise must decide what that release means.

That work includes determining which changes matter to the business, which need testing, who should evaluate them and whether a capability should be enabled, delayed or restricted.

AI complicates that familiar process because a routine release can introduce a capability with new data access, automation, cost or governance requirements.

A separate AI purchase creates an obvious review point.

A new AI feature within an already-approved application might appear as one item among many in the release notes.

That does not make the capability unimportant. It makes it easier to overlook.

Administrators should review and test consequential updates before broad production use when practical, but modern SaaS environments move too quickly to turn every release into a project.

That makes it more important to identify which changes require attention and which can remain routine.

Not every new button matters. Some new capabilities do.

The application inventory is no longer enough

Traditional application inventories are still necessary for contracts, ownership, security, integration and lifecycle management. Application portfolio management gives enterprises a way to inventory, evaluate and govern software at the application level.

Embedded AI, however, creates a second, capability-level layer.

An AI feature inventory can track each capability, including who can use it, what data it can access and what actions it is permitted to take. It can also capture the platform and vendor, affected workflows, ownership, human-review requirements, cost, usage, risk and expected outcomes.

At the individual capability level, organizations also need visibility into questions such as:

Is the feature enabled? Who can use it? What data can it reach? Does it only generate or recommend something? Can it take action on its own? Can it update records or trigger actions in another system? Does a person approve the result first? How is the capability priced? When was it last reviewed?

Those details cannot always be inferred from the application record.

That means enterprises increasingly need to track important AI features individually rather than assuming the application record tells them everything they need to know.

Capabilities can differ sharply inside the same application

The application-level view can also hide differences in risk. Two AI features from the same vendor can behave very differently even when they are inside the same product and covered by the same contract.

One might summarize text. Another might make recommendations. A third might act on behalf of a user. A fourth might interact with multiple systems.

That is why deciding which embedded AI features should be turned on increasingly depends on operational effect and risk rather than merely whether the vendor has made the feature available.

Risk is not the only difference that can be hidden at the application level. Individual AI capabilities can also have different costs and pricing models.

One capability might be included in the existing subscription. Another might consume credits, tokens, resolutions or some other usage metric.

Embedded AI can create new software cost-control problems when those additional consumption models are spread across suites the enterprise already uses.

Again, knowing the application is not enough.

The enterprise has to know which capabilities are active, what they can do and what they cost.

AI capabilities can reach beyond the application

Managing individual capabilities becomes even more important when a feature reaches outside the product in which it appears.

AI capabilities increasingly depend on data, identity, permissions, integrations and business context that comes from other systems. That means the application where a feature appears is not always the only place that matters.

An agent can appear to be a CRM capability while using information from ERP. An employee assistant can surface within a productivity suite by relying on HR data. A collaboration AI feature can turn meetings and chats into information that later becomes available to search, analytics or other AI tools.

This is one reason AI agents do not always map cleanly to individual applications. One SaaS product can contain multiple agents with their own access patterns, while application inventories or governance tools might see only the parent application.

That makes application ownership necessary but insufficient.

The application team might know where the feature lives and how it is configured. But software teams also need to understand which data, systems and workflows the capability depends on or affects beyond that application.

That broader view can help teams distinguish routine AI features from capabilities that require closer review.

Enterprises do not need the same governance process for every embedded feature.

They need enough visibility to know which ones deserve attention.

A limited AI feature that summarizes information or assists a small group of users might require little more than ordinary application management.

A feature that expands access to sensitive data, adds autonomous action, affects employees or customers, changes workflows, creates significant new costs or interacts with multiple enterprise systems deserves more scrutiny.

The exact threshold depends on the organization. But the first requirement is visibility.

A company cannot evaluate an AI capability it does not know is active. It cannot manage cost if it does not know how the capability is metered. It cannot assign responsibility if nobody knows which team is responsible for the feature. And it cannot decide whether to restrict or disable something if the AI layer has become effectively invisible inside the application stack.

Software management needs a more detailed view

Enterprise software management is not abandoning applications. It increasingly needs to look inside them as well.

The application remains an important management unit for procurement, administration, architecture and support. But embedded AI keeps adding capabilities whose data access, autonomy, cost, governance and business effects can vary even within the same application.

That changes the basic management question. It is no longer sufficient to ask what applications the enterprise has. Organizations increasingly need to ask what capabilities are active inside those applications and what those capabilities are allowed to do.

James Alan Miller is a veteran technology editor and writer who leads Informa TechTarget's Enterprise Software group. He oversees coverage of ERP & Supply Chain, HR Software, Customer Experience, Communications & Collaboration and End-User Computing topics.