From Telco “Dumb Pipes” to AI-Orchestrated Software: Observations on Where SaaS Might Be Heading

I’ve been thinking about an analogy that may or may not hold, but it keeps coming back when reading about the current AI vs SaaS debate.

It comes from telecom, where I started working more than two decades ago.

In the early 2000s, telecom operators assumed they controlled the digital ecosystem. They owned the infrastructure, the customer relationship, and the services themselves. Voice and SMS were the primary revenue sources, and operators experimented with early “value-added services” such as ringtones, news feeds, and mobile portals.

Then smartphones and mobile internet changed the game.

• Messaging migrated from SMS to applications like WhatsApp.

• Voice moved to VoIP.

• Content shifted to streaming services and platforms.

Operators still owned the networks, but much of the value moved above them into software platforms.

The telecom industry eventually described this outcome as the “dumb pipe” problem, where operators provide connectivity while application platforms capture most of the economic value.

I am not claiming enterprise software will repeat telecom history. But the pattern - value migrating upward in the technology stack - feels familiar when looking at what is happening with SaaS and AI today.

The SaaS model that dominated the last decade

For roughly two decades, enterprise software has been organised around SaaS applications.

Companies adopted cloud systems for specific functions:

• ERP for finance and operations

• CRM for sales and customer management

• HRMS for workforce management

• E-procurement and supply chain platforms

These systems essentially serve as systems of record - databases containing the organisation’s authoritative operational and master data.

If simplified, the architecture looked something like this:

User → SaaS application → database.

The application interface was the primary point of interaction.

The rise of orchestration layers

Over time, organisations accumulated dozens or even hundreds of SaaS tools. Once that happened, a new layer became necessary: integration and workflow orchestration.

Platforms emerged to connect these systems and coordinate processes across them. These include integration platforms and automation environments.

Examples include Microsoft’s Power Platform, Salesforce MuleSoft, or SAP’s Business Technology Platform.

These platforms allow workflows such as:

• synchronising data between systems

• triggering approvals

• coordinating multi-system processes.

Microsoft describes this capability as “automation across apps and services”.

In practice, this orchestration layer already acts as a process backbone for many organisations.

AI introduces another layer

Artificial intelligence appears to add yet another level above this architecture.

AI systems do not replace enterprise systems directly. Instead, they require two things:

1. access to enterprise data

2. access to enterprise workflows

For example, if AI is asked to prepare a sourcing strategy or evaluate supplier risk, it needs:

• supplier data from procurement systems

• financial data from ERP

• workflow rules for approvals

• access to contracts and documents.

This means AI operates on top of existing systems, not instead of them.

Increasingly, the architecture resembles something like this:

Systems of record → orchestration layer → AI intelligence layer.

Users interact with the AI, which then triggers workflows across enterprise systems.

The SaaS vs AI discussion may be misleading

Many articles currently frame the situation as AI replacing SaaS.

That seems unlikely. AI models do not maintain structured operational databases.

At the same time, SaaS systems alone cannot deliver advanced AI functionality without orchestration and analytics layers.

Instead, the layers appear to stack rather than replace each other.

• Enterprise systems store data.

• Automation platforms coordinate processes.

• AI systems analyse and trigger actions.

Major vendors are merging SaaS and platforms

One thing that stands out is how aggressively the large enterprise vendors are merging applications and platforms (SaaS and PaaS).

Microsoft

Microsoft’s ecosystem combines:

• Dynamics and Microsoft 365 applications

• The Power Platform for automation and integration

• Azure AI and Copilot capabilities

Microsoft positions the Power Platform as a way to connect applications and automate business processes across them.

The strategy can be interpreted as placing orchestration in the centre of the stack.

Salesforce

Salesforce is a global leader in CRM applications, but has gradually expanded into a broader platform ecosystem.

Today, their stack includes:

• CRM systems

• the MuleSoft integration platform

• Data Cloud for enterprise data unification

• Einstein AI and agent-based automation

Salesforce describes MuleSoft as a platform for “connecting applications, data, and devices across organisations”.

Again, the emphasis seems to be on orchestration rather than standalone applications.

SAP

SAP’s strategy looks similar.

The traditional ERP system (S/4HANA) remains the system of record, but SAP has developed a platform layer called Business Technology Platform (BTP) that provides integration, extension, and data services.

SAP describes BTP as a way to “integrate, extend, and build applications across SAP and third-party systems”.

AI capabilities are then added to this platform.

Why orchestration might become strategic

If the telecom analogy holds even partially, the key question becomes:

Which layer controls the ecosystem?

In telecom, the infrastructure layer lost strategic control when applications moved above the network.

In enterprise software, something similar could happen if AI agents start orchestrating enterprise processes independently of SaaS platforms.

For instance, an AI agent might:

• Read CRM data

• analyse ERP information

• trigger procurement workflows

• notify users in collaboration tools

In that scenario, the user interacts mainly with the AI layer rather than the application interface.

This is why vendors appear determined to keep orchestration inside their own ecosystems.

A tentative conclusion

Personally, I do not think SaaS is disappearing.

Enterprise systems remain essential as systems of record.

But the centre of gravity may be shifting toward the layers that coordinate and interpret those systems.

Those layers appear to be:

• workflow orchestration platforms

• enterprise data platforms

• AI intelligence layers.

Let's see whether SaaS vendors can successfully integrate these layers themselves, or whether independent AI systems will end up orchestrating enterprise software from outside the traditional application stack.

If the telecom history teaches anything, it is that value often migrates toward the layer that coordinates interactions across the ecosystem.

Whether the large SaaS vendors can keep control of that layer remains one of the most interesting questions in the enterprise software industry.