SaaS Is Not Dead. It Is Becoming the Operating Layer for Enterprise AI

Every technology cycle produces a dramatic obituary.

Cloud killed on-premise software.
Mobile killed the desktop.
Social media killed the website.
Now AI is supposedly killing SaaS.

I do not buy it.

AI will undoubtedly disrupt the software industry. It will change interfaces, pricing models, product development, competitive dynamics, and how much software companies can charge for basic functionality.

Some SaaS products will disappear. Many point solutions will struggle to justify their existence. Software that merely stores information, moves it between screens, or adds a thin interface over a common workflow will face real pressure.

But saying “SaaS is dead” misses how large organisations actually operate.

A CTO responsible for thousands of employees across multiple markets is not going to wholesale replace the systems running finance, sales, HR, customer service, marketing, security, collaboration, and procurement because a new AI application has a better demonstration.

The enterprise software stack is not being demolished. It is being re-architected around intelligence.

The SaaS Obituary Confuses the Interface with the Infrastructure

The argument that AI will kill SaaS usually follows a simple logic.

Traditional software requires users to navigate menus, enter data, configure workflows, and move between applications. AI agents can potentially understand intent, retrieve information, make decisions, and execute actions on behalf of the user.

Therefore, the theory goes, employees will no longer need dozens of applications. They will simply tell an intelligent agent what they want done.

There is some truth in this.

The traditional software interface is changing. Employees may spend less time clicking through applications and more time interacting through natural language, voice, or AI assistants.

But the disappearance of the interface does not mean the disappearance of the underlying software.

When an employee asks an agent to update an opportunity, approve an expense, resolve a support ticket, change a marketing campaign, or onboard a new employee, something still needs to:

• Store the official record

• Enforce business rules

• Verify identity and permissions

• Trigger approvals

• Connect with other systems

• Maintain an audit trail

• Meet regulatory requirements

• Report on what happened

The conversational interface may belong to an AI assistant.

The operational backbone will still depend on enterprise software.

AI may reduce the number of screens employees see. It does not remove the need for systems of record, systems of action, and governed business workflows.

The Enterprise Data Tells a More Nuanced Story

Okta’s 2026 Enterprise AI Index analysed application access across more than 20,000 organisations between June 2022 and June 2026. The research found rapid growth among AI-native vendors such as Anthropic, OpenAI, and Cursor. It also found strong enterprise growth among established platforms including Microsoft 365, GitHub, Google Workspace, Figma, Slack, Adobe, and Notion.

This is not a market where one generation of software is simply replacing another.

It is a dual-engine market.

One engine consists of AI-native companies creating new tools, interfaces, and categories.

The other consists of established software companies embedding AI into platforms that already sit inside enterprise workflows.

Business Insider’s analysis of the Okta data reached a similar conclusion. Corporations are adopting specialised AI applications while continuing to expand their use of established platforms enhanced with AI.

That distinction matters.

It suggests the enterprise AI stack will not consolidate neatly around one model, one chatbot, or one agent platform.

It is more likely to become:

• Multi-model

• Multi-vendor

• Embedded within existing applications

• Connected across systems

• Governed through common enterprise controls

The competitive advantage is shifting away from simply choosing the best AI model.

It is moving towards enterprise orchestration.

No Serious CTO Starts with a Blank Sheet

Technology commentary often imagines enterprise transformation as though every organisation is a startup.

It is not. A multinational company may have thousands of applications, decades of accumulated data, contractual commitments, country-specific configurations, regulatory obligations, and deeply embedded operating processes.

A CRM system, for example, is not merely a database of customer names.

It may be connected to:

• Marketing automation

• Sales forecasting

• Partner management

• Customer service

• Finance

• Incentive compensation

• Territory planning

• Data warehouses

• Identity systems

• Regulatory reporting

Replacing it is not simply a software decision. It is an operating-model decision.

The same applies to ERP, HR, productivity, service management, content, design, and collaboration platforms.

A CTO operating across Singapore, India, Australia, Japan, Indonesia, and other APAC markets must also account for different data-residency requirements, languages, business practices, infrastructure maturity, and regulatory expectations.

No responsible technology leader will replace all of that because an AI-native product can complete one workflow more elegantly.

The more credible path is selective modernisation.

Keep the systems that provide scale, control, and continuity. Introduce new AI capabilities where they produce a meaningful advantage. Connect the two through secure, governed workflows.

This is less dramatic than declaring SaaS dead. It is also much closer to how enterprise change happens.

Incumbents Have Something AI Startups Want

AI-native vendors have important advantages.

They can move quickly. They are not constrained by legacy product architectures. Their entire user experience can be designed around AI rather than having AI added later.

But established enterprise vendors have advantages that are just as difficult to replicate.

They already have:

• Enterprise distribution

• Customer trust

• Procurement approval

• Security certifications

• Identity integrations

• Installed data

• Partner ecosystems

• Workflow context

• Long-term contracts

• Regulatory credibility

Most importantly, they sit where the work already happens.

Microsoft does not need to persuade an employee to move their documents, email, meetings, spreadsheets, presentations, and organisational directory into a new environment before Copilot can become useful.

Salesforce does not need to recreate the customer record before an agent can act on it.

ServiceNow does not need to rebuild years of IT workflows, approval chains, service catalogues, and business rules before introducing AI into service operations.

Adobe does not need to persuade creative teams to abandon their content supply chain before adding generative capabilities.

The incumbents’ moat is not necessarily the intelligence of their model.

It is their proximity to enterprise data and workflow.

The Evidence Is Already Appearing at Scale

In June 2026, Microsoft announced that Infosys, TCS, and Wipro had each expanded Microsoft 365 Copilot deployments to more than 100,000 employees. Collectively, this represented over 300,000 licences across the three organisations. Microsoft said the deployments were being integrated into engineering, service delivery, and corporate workflows rather than treated only as standalone productivity experiments.

Whatever view one takes of vendor-reported adoption figures, the scale is instructive.

These organisations are not replacing their productivity platforms with an entirely new AI operating environment.

They are putting AI inside the environment their employees already use.

ServiceNow is pursuing a similar strategy from a different position. Its 2026 Action Fabric announcement opened ServiceNow workflows to agents built using systems such as Claude, Copilot, or customers’ own internal tools. Those agents can interact with existing workflows, approvals, catalogues, and business rules while actions remain governed, permission-scoped, and auditable.

That is a powerful clue about where the market is heading.

ServiceNow is not betting that every enterprise will choose one model.

It is betting that whichever agent the enterprise chooses will still need a governed place to execute work.

Salesforce is making a comparable argument. Its Agentforce IT Service product operates across Salesforce, Slack, Microsoft Teams, email, web, and voice. The proposition is not that one interface will replace every other system. It is that agents can coordinate work across the platforms employees already use.

Even OpenAI, one of the most important AI-native challengers, describes its enterprise direction in terms of connecting models and agents to company context, internal systems, external data, permissions, and controls. It has also acknowledged that enterprises are frustrated by disconnected AI point solutions that do not communicate with one another.

The market is converging on the same enterprise requirement:

Intelligence must connect to workflow.

What Is Actually Dying?

SaaS is not dead. But parts of the traditional SaaS model are under pressure.

Thin Point Solutions

A product that performs one narrow function and has little proprietary data, workflow depth, or integration value may be vulnerable.

If the capability can be reproduced by a general-purpose model, bundled into a larger platform, or performed by an agent at low cost, customers will question why they need another vendor.

Seat-Based Pricing Without Clear Usage

The traditional model of charging every employee the same monthly fee becomes harder to defend when AI usage varies significantly.

Some employees may use an agent hundreds of times per month. Others may barely use it. Some AI tasks may consume substantial compute, while others cost almost nothing.

Vendors are already experimenting with usage credits, consumption pricing, hybrid models, and outcome-based fees because the economics of AI do not map cleanly onto traditional per-seat pricing.

Passive Systems of Record

Software that only stores data will increasingly be expected to interpret it, recommend actions, and initiate workflows.

Customers will no longer be satisfied with dashboards that tell them what happened.

They will expect systems to help determine what should happen next.

Poor User Experiences

AI will place pressure on complicated interfaces and excessive administrative work.

Employees will increasingly ask why they need to navigate ten screens to complete an action that an agent could execute through a single instruction.

That pressure is healthy. It does not eliminate enterprise software. It forces enterprise software to become more intelligent and easier to use.

Undifferentiated AI Wrappers

Ironically, some of the products most at risk may be the first generation of AI applications.

A thin wrapper around a foundation model, with no distinctive data, workflow integration, governance capability, or distribution advantage, will be difficult to defend.

The model provider may add the feature. An incumbent may bundle it. A competitor may replicate it.

The new software moat will not come from adding a chat box.

It will come from owning a valuable part of the workflow.

The New Enterprise Software Stack

The next-generation enterprise stack will probably have several layers.

1. Foundation Models

Organisations will use multiple models based on capability, cost, latency, language performance, risk, and data requirements.

The model may change without the business workflow changing.

2. Employee and Customer Interfaces

These may include ChatGPT, Copilot, Gemini, Slack, Teams, voice interfaces, specialised applications, and embedded assistants.

Employees may interact with AI through several interfaces depending on the task.

3. Systems of Record

CRM, ERP, HR, finance, customer, product, and operational platforms will continue to maintain authoritative enterprise data.

These systems will remain essential because agents require reliable context.

4. Systems of Action

Workflow platforms will translate intent into execution.

They will trigger approvals, update records, assign work, connect departments, and enforce business rules.

5. Integration and Orchestration

This layer will connect models, agents, data, applications, APIs, and human decision-makers.

It may become one of the most strategically important parts of the stack.

6. Identity and Governance

Every human, application, and AI agent will require defined permissions.

Organisations will need to know what an agent can access, what it can change, where human approval is required, and how its actions can be audited.

The future enterprise stack will therefore not be smaller in every sense.

It may be more abstracted at the interface, but more sophisticated underneath.

The Real Competitive Battle Is Workflow Ownership

For SaaS vendors, the central question is no longer:

How do we add AI to our product?

It is:

Which part of the customer’s workflow do we have the right to orchestrate?

That right is earned through a combination of data, trust, integration, business rules, and demonstrated outcomes.

A CRM provider may have the right to orchestrate parts of the revenue workflow.

A service-management platform may have the right to orchestrate IT, employee, and customer-service processes.

A collaboration platform may become the interface through which employees access multiple agents.

A design platform may orchestrate the journey from creative brief to approved asset.

A specialist AI-native vendor may own a high-value workflow that incumbents cannot perform well enough.

This is why I do not believe one platform will absorb everything.

Different vendors will own different layers and moments of work.

The enterprise advantage will come from making those components work together.

A Marketing Example

Consider a campaign running across eight APAC markets.

The work may involve:

• Customer insights from a research platform

• Audience data from a customer platform

• Planning in Microsoft 365 or Google Workspace

• Creative production in Adobe, Canva, or Figma

• Collaboration through Teams or Slack

• Activation through advertising and marketing automation platforms

• Legal, brand, and regulatory approval

• Localisation across languages and cultures

• Performance reporting through analytics tools

A general-purpose AI assistant can help draft the campaign brief or produce initial copy.

That is useful.

But the larger opportunity is to orchestrate the entire process.

An intelligent workflow could:

1. Analyse customer, category, and campaign data.

2. Prepare a brief using the company’s approved framework.

3. Generate market-specific content variants.

4. Check outputs against brand, legal, and responsible-AI rules.

5. Route high-risk claims for human review.

6. Send local-language variants to market teams.

7. Move approved assets into activation platforms.

8. Monitor performance and feed results into the next planning cycle.

No single model delivers this operating capability.

The advantage comes from combining models, software, data, controls, and human judgement.

That is execution fluency.

Why APAC Makes the “SaaS Is Dead” Argument Even Less Credible

APAC is fragmented by design.

A platform that works for an English-speaking workforce in Singapore or Australia may need significant adaptation for Japan, Korea, Indonesia, Thailand, or Vietnam.

Large organisations must manage differences in:

• Language

• Culture

• Regulation

• Data residency

• Cloud infrastructure

• Labour practices

• Customer behaviour

• Digital maturity

• Local technology ecosystems

A single global AI layer may provide consistency, but it cannot eliminate these market realities.

Regional enterprises need common standards combined with local flexibility.

This is another reason the future will not be a clean replacement of SaaS with one universal agent.

The more pragmatic model is:

• Centralise identity, security, governance, and architectural standards.

• Standardise selected enterprise platforms where scale matters.

• Allow model and application flexibility where business needs differ.

• Build integration so information and actions can move safely.

• Retain human accountability for consequential decisions.

This is ecosystem-first adoption.

It accepts complexity rather than pretending it can be wished away.

The Governance Question Gets Harder, Not Easier

Embedding AI into enterprise software creates a significant governance challenge.

An employee may begin a task in Microsoft 365, retrieve information from a CRM, involve a specialist model, trigger a ServiceNow workflow, and communicate the result through Slack.

Which system produced the answer?

Which model processed the data?

Who authorised the action?

Which record is authoritative?

Where is the audit trail?

Who is responsible if the outcome is wrong?

Governance can no longer be organised only around a list of approved AI tools.

It must follow the workflow.

A governance-ready enterprise should be able to identify:

• Which models and agents are active

• What data they can access

• Which actions they are authorised to perform

• What risk tier applies to each use case

• Where human approval is mandatory

• How decisions and actions are logged

• How agent identities are created and removed

• Who owns incidents and remediation

• How third-party model changes are assessed

This reflects the practical, responsible adoption philosophy at the heart of my broader work. Enterprise transformation should convert complexity into clarity, preserve accountability, and keep human judgement at the centre.

The objective is not to slow adoption.

It is to make scale survivable.

A Playbook for CTOs and Business Leaders

1. Stop Debating Whether SaaS Will Survive

That is the wrong altitude.

Instead, identify which parts of your stack provide genuine operational value, which are becoming commodities, and which should be replaced or consolidated.

2. Map Workflows, Not Just Applications

An application inventory tells you what you own.

A workflow map tells you how the business operates.

Identify where data moves, where decisions happen, where approvals stall, and where employees repeatedly perform low-value work.

3. Protect Systems of Record

AI outputs are probabilistic.

Core enterprise records cannot be.

Define which systems remain authoritative and ensure agents cannot create conflicting versions of business truth.

4. Build for Multiple Models

Avoid designing an enterprise architecture that depends entirely on one foundation model.

Models will improve, pricing will change, regulations will evolve, and different tasks will require different capabilities.

The architecture should allow substitution without requiring the entire workflow to be rebuilt.

5. Standardise the Control Plane

Organisations may permit multiple models and applications, but identity, access, logging, security, risk classification, and incident response should not be fragmented.

Flexibility at the application layer requires discipline at the governance layer.

6. Measure Business Outcomes

Do not measure AI success through licences deployed, prompts submitted, or content generated.

Measure:

• Revenue impact

• Cycle-time reduction

• Cost-to-serve

• Error rates

• Customer satisfaction

• Employee capacity

• Compliance incidents

• Speed and quality of decision-making

The unit of value is not the AI interaction.

It is the improved workflow.

7. Rationalise Aggressively, but Not Recklessly

AI creates a legitimate opportunity to remove redundant software.

Use it.

But do not confuse application reduction with strategic transformation.

Eliminating ten tools and creating one uncontrolled AI dependency is not simplification. It is risk migration.

The Takeaway for Builders

For software companies, the message is uncomfortable but clear.

Adding AI features will not be enough.

The strongest vendors will need to demonstrate that they can:

• Own or orchestrate an important workflow

• Connect safely with other enterprise systems

• Work across multiple models

• Protect customer data

• Support agent identity and permissions

• Produce measurable business outcomes

• Adapt their commercial model as usage changes

• Remain relevant even when the visible interface belongs to someone else

Some SaaS businesses will not survive this transition.

But that is different from saying SaaS itself is dead.

The category is not disappearing.

It is evolving from software employees operate into infrastructure through which humans and agents get work done.

The Takeaway for Decision Makers

The enterprise AI battle is not really OpenAI versus Microsoft, AI-native versus incumbent, or agents versus SaaS.

It is a battle over who owns the workflow, who holds the context, who governs the action, and who can prove the outcome.

The organisations that win will not be those that replace the most software or select the most powerful model.

They will be those that orchestrate intelligence across the software ecosystem they already have, while steadily removing what no longer earns its place.

Before approving the next AI tool or announcing another platform consolidation programme, ask:

Does this strengthen the way our organisation works, or does it simply add another layer of technology?

SaaS is not dead.

But passive, disconnected, and undifferentiated software should be worried.