Perché l'intelligenza artificiale non metterà fine al software aziendale

The most dramatic prediction about generative AI and enterprise software is also one of the easiest to understand. Employees will stop opening dozens of applications, ask an AI assistant to complete the work instead, and leave much of the software-as-a-service industry behind.

There is already some evidence for the first part of that argument. AI agents can search documents, draft responses, summarise meetings, update records and move information between systems. Apple, Microsoft, Google and a growing group of enterprise-software companies are designing assistants that sit above conventional applications and allow users to complete work through natural-language instructions.

What follows is less certain.

The disappearance of familiar interfaces does not mean that the systems behind them become unnecessary. A sales agent still needs an authoritative customer record. A procurement assistant needs approved suppliers, spending limits and an audit trail. An AI tool preparing an employee contract still depends on specialist HR, identity and document-management systems.

Generative AI may weaken some SaaS products, particularly those that offer little more than a convenient interface around a generic task. It is unlikely to remove the enterprise software stack that allows companies to control data, permissions, workflows and legal responsibility.

For IT teams, the result looks less like an apocalypse than a more complicated layer of administration.

The Interface Is Becoming Less Important

For much of the SaaS era, enterprise software was sold through its interface. Vendors competed over dashboards, menus, collaboration features and the number of functions available inside an application.

AI agents challenge that model because users may no longer need to navigate every product directly. An employee could ask an assistant to identify overdue invoices, prepare a report or create a project update without opening the finance, analytics and project-management tools where the relevant information is stored.

This changes where the user experiences value. The AI assistant becomes the visible working environment, while specialist SaaS products operate further in the background.

That shift is already shaping the largest workplace platforms. Microsoft reported in its 2026 Work Trend Index that the number of active agents in the Microsoft 365 ecosystem increased fifteenfold year on year and eighteenfold among large enterprises. The report describes a workplace in which employees increasingly delegate execution to AI while retaining responsibility for direction, judgement and review.

It is tempting to conclude that one sufficiently capable agent will replace a collection of individual subscriptions. In practice, the agent needs somewhere reliable to retrieve information, execute approved actions and record what it has done.

The interface may become optional. The underlying system does not.

Specialist Software Still Holds The Record

Enterprise applications exist partly because businesses cannot run important processes through unstructured conversations and generated text.

A customer relationship management platform contains defined accounts, ownership rules, sales stages and contractual histories. An enterprise resource planning system connects orders, inventory, suppliers and financial records. Device-management software determines which employee can access a company laptop, which applications are installed and what happens when the device is lost.

These platforms are systems of record. Their value lies in the structure, permissions and operating logic built around the data, not simply in what appears on screen.

The original Apple @ Work argument makes this distinction clearly. IT departments are unlikely to replace established security, identity, device-management and business-software vendors with a general-purpose model from OpenAI, Anthropic or another AI provider. Enterprise environments rely on products designed for narrow and consequential functions that a conversational assistant cannot safely improvise.

An AI agent may make those products easier to use. It may also reduce the number of employees who require full access to every interface. Neither change removes the need for authoritative records or controlled execution.

This is where the strongest SaaS companies retain their position. A platform deeply embedded in finance, compliance, logistics or security owns more than a user workflow. It holds operational history and provides evidence that a process occurred in an approved way.

That becomes more valuable when machines begin initiating the work.

Apple’s Role Is At The Device And Privacy Layer

Apple’s significance in this discussion comes from its position inside the enterprise endpoint rather than from being a conventional SaaS provider.

The company controls the hardware, operating system and increasingly the intelligence available across the Mac, iPhone and iPad. Apple Intelligence combines on-device processing with Private Cloud Compute for requests that require larger models. Apple says the architecture is designed so that more complex processing can be handled in the cloud without exposing users’ personal data to Apple in a conventional server environment.

At WWDC in June 2026, Apple introduced a new generation of Apple Intelligence and a more capable Siri, extending AI functions across its software platforms and opening additional intelligence features to developers.

For enterprise IT, this creates both an opportunity and a governance problem.

Employees may be able to summarise, search and act on information across applications with fewer manual steps. Developers can add system-level intelligence to business tools without building every model capability independently. On-device processing may also reduce some of the privacy concerns associated with sending sensitive prompts and documents to a remote AI service.

Yet these features still have to operate within company policy. IT teams need to determine which devices support the functions, what corporate information may be processed, which external models or services can be used and how AI-enabled applications interact with managed data.

Apple’s deployment framework already allows organisations to manage hardware, applications, services and security settings at scale through Apple Business and third-party device-management services. In March 2026, the company also announced Apple Business, an integrated platform with built-in device-management capabilities for configuring employee groups, security policies and applications.

Generative AI does not diminish that administrative layer. It gives IT teams more behaviour to control.

SaaS Economics Will Still Change

The absence of an apocalypse does not mean enterprise software remains commercially untouched.

Seat-based pricing becomes harder to defend when an AI agent can complete work on behalf of several employees. A company may question why every user needs a premium licence for an application that is rarely opened directly. Some functions that once justified a separate SaaS product can now be generated or automated inside a larger platform.

This puts pressure on software vendors whose differentiation rests mainly on presentation, basic content generation or a thin workflow that can be replicated by an agent.

The products in the strongest position will own a trusted dataset, a regulated process, a difficult integration or a transaction that cannot be completed through language alone. They may charge for usage, automated outcomes, data volume or agent activity rather than relying exclusively on the number of human users.

Enterprise AI is therefore likely to reorganise the SaaS market rather than erase it. Applications become less visible, integrations become more valuable and pricing moves closer to the volume or value of work performed.

AI agents also create activity rather than merely replacing it. A system capable of analysing every support request, reviewing every contract or checking every device configuration can initiate more actions than an overstretched human team previously had time to complete. That increases the need for software capable of processing, validating and documenting those actions.

The SaaS layer becomes an execution and control environment for AI.

IT Will Have To Manage Non-Human Workers

The most consequential change may be the arrival of software identities that act inside corporate systems.

Traditional IT administration is built around people and devices. An employee receives an account, defined permissions, approved applications and a company-managed laptop. When the employee leaves, access is revoked.

An AI agent also needs an identity, permissions and limits. It may need access to email, cloud storage, customer records or financial systems. It may act continuously and complete transactions without a person approving each individual step.

That creates several questions familiar to security teams but harder to answer in an autonomous environment. Who owns the agent? Which information can it retrieve? Can it send messages externally? What spending threshold can it approve? How is its activity recorded? What happens when its instructions conflict with company policy?

Microsoft’s Agent 365 initiative illustrates the direction of travel. The company introduced tools that allow administrators to authorise, monitor, quarantine and secure agents in a way comparable with the management of employees and devices. The product emerged in response to concerns over control, security and the ability to assess whether autonomous workplace tools produce sufficient value.

Independent research is pointing to the same requirement. A 2026 study of enterprise AI roles found that greater automation is creating a need for revised permission structures, governance functions and oversight models rather than eliminating human responsibility.

IT teams will increasingly manage a mixed workforce of people, devices, applications and autonomous agents. The administrative burden does not disappear simply because the software becomes easier to speak to.

Integration Will Decide Which Tools Survive

Many organisations already operate with overlapping SaaS products, fragmented data and multiple versions of the same information. Generative AI can hide some of that complexity from employees, but it cannot automatically resolve it.

An agent asked to prepare a revenue forecast still needs to know whether the authoritative numbers are held in the finance system, a CRM report or a spreadsheet maintained by one business unit. It must recognise duplicate customer records, apply the correct access policy and understand whether a document is current or archived.

Without reliable integration and data governance, generative AI can produce a more fluent version of the organisation’s existing confusion.

This gives IT teams a more strategic role. They need to decide which systems remain authoritative, retire redundant applications, improve data quality and define the interfaces through which agents can act. Procurement also becomes more demanding because every AI-enabled SaaS vendor may bring an additional model provider, data-processing arrangement and security dependency.

The eventual enterprise architecture may contain fewer visible applications and fewer duplicative tools. The remaining platforms will probably be more interconnected and more deeply governed.

That is consolidation, not collapse.

Security Becomes Harder To See

Conventional SaaS risks are reasonably familiar. IT departments assess how a vendor stores data, manages identities, handles incidents and supports regulatory requirements.

Generative AI adds less visible failure modes. Sensitive information may be included in prompts, retrieved from the wrong source or exposed through an overly permissive agent. Generated instructions may be inaccurate. Malicious content can attempt to manipulate an agent into revealing information or completing an unauthorised action.

The convenience of natural language makes these risks easier to overlook. An employee asking an agent to “send the latest figures to the team” may not know which file the system selected, whether every recipient was authorised or what information was removed before the message was sent.

This is another reason specialist software persists. Identity platforms, security tools, data-loss prevention systems and audit services provide controls that a general-purpose AI interface cannot be trusted to recreate independently.

The market may eventually produce an orchestration layer that chooses between models and applications according to cost, security and performance. Even then, the orchestration platform itself becomes another critical enterprise system requiring procurement, monitoring and governance.

The Weakest SaaS Products Are Exposed

The sensible rejection of a SaaS apocalypse should not become complacency.

Generative AI makes it easier to create software and lowers the cost of reproducing basic features. Companies will have less patience for isolated products that require employees to enter the same information repeatedly or maintain another dashboard without contributing distinctive data or operational control.

Some vendors will be absorbed into broader platforms. Others will become features rather than independent companies. Businesses may reduce the number of licences they buy as AI performs work on behalf of users.

The distinction will be based on how close a product sits to the actual operation of the company.

A generic writing assistant can be replaced relatively easily. A validated system controlling payroll, manufacturing, clinical records or access to corporate devices cannot. The closer software is to money, regulated information, physical infrastructure or legal accountability, the harder it is to remove.

Generative AI will reveal which applications were genuinely running the business and which were merely adding another place for employees to click.

IT Remains Accountable For The Outcome

The most persuasive promise of enterprise AI is that employees will no longer need to understand the architecture behind every task. They will describe an outcome and allow intelligent systems to find the information, select the application and complete much of the work.

Someone must still design and govern that architecture.

IT teams will determine which agents may operate, what data they can access and which systems remain authoritative. They will investigate failures that cross several applications and negotiate contracts whose costs may depend on unpredictable model usage. They will also have to explain why an automated decision was made when a regulator, employee or client challenges it.

The work becomes less concerned with supporting individual menus and more concerned with controlling the movement of data and decisions across the organisation.

Generative AI may reduce the importance of the SaaS interface. It will expose weak vendors, challenge conventional licensing and accelerate consolidation across overcrowded software categories.

It will not remove the need for specialist systems or the teams responsible for making them work safely together.

The enterprise-software stack is not disappearing. It is becoming less visible to employees and considerably more consequential for IT.