AI agents are changing what organisations buy as software. Mistral AI CEO Arthur Mensch has argued that more than half of enterprise software could shift towards AI-built applications and workflow automation, putting pressure on traditional SaaS models. For CIOs, the opportunity is real — but so are the governance and integration challenges.

The idea of a “SaaS stack” has defined enterprise IT for more than a decade: buy best‑of‑breed tools, integrate them, and accept the subscription sprawl.

But AI is starting to change the shape of that stack. In recent interview comments picked up across the tech press, Arthur Mensch, CEO of Mistral AI, suggested that more than half of enterprise software could shift to AI — with procurement moving away from traditional SaaS licences and towards AI-driven workflows and AI-built applications.

If that sounds dramatic, it’s worth pausing. Not because SaaS is disappearing tomorrow — it isn’t — but because the mechanism behind the claim is plausible: AI compresses the time and cost of creating “just enough software” for a specific workflow.

What people mean by “AI replacing SaaS”

This isn’t about swapping every SaaS product for a chatbot.

It’s about shifting spend from:

• standalone tools that require humans to click through screens

…to:

• AI-assisted workflows that operate across tools (create, update, route, summarise, schedule)

• custom, thin apps built quickly on top of your existing systems of record

In practice, that means some SaaS seats get reduced because the workflow becomes “agent-led” rather than “UI-led”.

Why the shift is happening now

Three conditions are aligning:

1) Agents can now take actions, not just answer questions

We’re moving from “AI that explains” to “AI that executes” — updating tickets, creating tasks, drafting docs, scheduling follow-ups and more.

2) Building small, workflow-specific apps is faster than ever

If your team can ship an internal tool in days (not months), the build‑vs‑buy equation changes.

3) Data is becoming the real moat

Workflows are only as good as the data they can access. That pushes organisations to focus on:

• clean, permissioned knowledge

• systems of record

• governance and auditability

What’s likely to change for CIOs and IT leaders

Software procurement becomes more outcome-led

Instead of buying a tool and hoping adoption happens, leaders will prioritise:

• cycle time improvements

• reduction in manual handovers

• fewer missed follow-ups and errors

The “integration layer” becomes more strategic

If agents operate across multiple systems, organisations need:

• clear permission boundaries

• reliable connectors

• logging and audit trails

The long tail of SaaS gets squeezed first

The most vulnerable subscriptions are tools bought to solve narrow problems where:

• switching costs are low

• data is easily exported

• the workflow can be recreated with AI + your existing platforms

What probably won’t change overnight

Regulated workflows will still need strong controls

AI doesn’t remove compliance requirements — it increases the need for:

• access control

• retention policies

• human approvals

• clear accountability

Switching costs are still real

Enterprise systems are sticky because of:

• integrations

• data models

• training and change management

AI shifts the pressure point, but it doesn’t magically remove the hard work of operational change.

Practical steps: how to respond without chasing hype

If you’re in IT, Ops, Product, or Digital, this is how to test the thesis safely.

Step 1: Choose one workflow with repeatable pain

Good candidates:

• customer request intake → triage → resolution

• project updates and reporting

• contract review support

• knowledge search → action creation

Step 2: Define what the agent is allowed to do

Start with low-risk actions:

• drafting and summarising

• creating tasks

• adding comments

Require approval for:

• changes to customer records

• financial commitments

• access/permissions

Step 3: Measure impact in weeks, not quarters

Track:

• time saved per case

• fewer handovers

• reduction in rework

• stakeholder satisfaction

Step 4: Build a shared operating rhythm

The fastest way AI fails is when every team uses it differently.

Create:

• prompt and workflow templates

• review checklists

• a simple governance policy

Where Generation Digital fits

If you’re exploring how AI changes your collaboration stack, we can help you identify which workflows are worth automating — and how to roll them out safely.

Summary

The claim that “over half of SaaS spending could shift to AI” is less about wiping out existing platforms and more about changing how work gets done. As agents become capable of taking actions across systems, the value moves from interfaces to outcomes.

Next steps

• Pick one workflow to pilot with an agent.

• Put guardrails and approvals in place.

• Measure results, then scale with templates.

6. FAQs

Q1: Did Mistral AI’s CEO really say over half of SaaS could shift to AI?
Multiple outlets reporting on the CNBC interview and related posts attribute that claim to Arthur Mensch. Treat it as a directional prediction, not a guaranteed forecast.

Q2: Does this mean SaaS is dead?
No. Core systems of record remain sticky. The likely change is seat reduction and consolidation where AI-driven workflows replace repetitive UI work.

Q3: Which SaaS tools are most at risk?
The long tail of narrow, low-switching-cost tools where workflows can be recreated with AI on top of existing platforms.

Q4: What should CIOs do first?
Run a pilot on one repeatable workflow, restrict what the agent can change, require approvals for high-risk actions, and measure outcomes.

Q5: How do you keep agent adoption safe?
Use least privilege, log actions, separate read vs write permissions, and train teams in repeatable workflows and review standards.