Copilot gutted our business, now AI agents are eyeing SaaS seats

A tweet crossed my feed this morning saying Apple is dead.

The argument was simple:

AI agents will generate just-in-time interfaces on the fly, hit backends directly through MCP, and make the App Store moat irrelevant.

I sat with that for a minute.

Not because Apple is what worried me.

Because if that argument is even half right, phones are not the first thing to break.

Per-seat SaaS is.

Take a CRM.

• A 200-person sales team pays for 200 seats.
• Everyone logs in.
• Everyone clicks around.
• Everyone updates pipelines, chases tasks, writes notes, checks dashboards and moves deals along.

That model makes perfect sense when humans are the workers.

Now change one thing.

Agents arrive.

• They qualify leads.
• Update records.
• Summarise calls.
• Trigger follow-ups.
• Prepare quotes.
• Check entitlements.
• Write the next email.
• Push the manager the exceptions.

Suddenly that 200-person team becomes 40 people using a layer of agents.

The CRM is still doing work.

Possibly more work than before.

• More API calls.
• More automations.
• More records updated.
• More workflows triggered.
• More data moving through the system.

But the vendor just lost 80% of the seats.

The product is being used more.

The revenue collapsed.

That is the maths sitting under a lot of very calm SaaS board decks right now.

And honestly?

I feel sorry for the companies that did everything right.

• They built beautiful products.
• They earned trust.
• They created huge audiences.
• They trained customers to live inside their workflows.
• They built onboarding, reporting, partner channels, certification paths and enterprise sales motions around humans clicking through screens.

And now they are staring at the ceiling at 3am realising the machine they built so well may be optimised for the wrong user, the wrong buyer and the wrong pricing model.

Because building an “agentic layer” is not just another feature release.

It is an admission that the front door of your product is becoming optional.

The workflow underneath survives.

The interface stops being the main event.

• Conversational AI books the holiday.
• Checks the entitlement.
• Raises the approval.
• Updates the CRM.
• Creates the quote.
• Answers the customer.
• Renders a chart in the chat if someone really needs to see one.

The software still matters.

The screens matter less.

And while incumbents maintain every page, button, dashboard, permission model and onboarding flow for the humans still clicking around, an AI-first competitor can start with the outcome and skip a lot of the legacy weight.

That is the real threat.

Not “AI will kill SaaS”.

It won’t.

But AI will expose which SaaS companies were selling outcomes, and which were renting access to screens.

Salesforce has already seen the direction of travel. Agentforce pricing now mixes user licences, conversations and action-based credits, which tells you even Salesforce knows the old per-seat model won’t cover every agentic use case.

Because per-seat pricing was a brilliant proxy for value when humans were the ones doing the work.

Once agents become part of the workforce, the proxy breaks.

And you can see the scramble starting:

• Zendesk charging per resolved ticket.
• Salesforce experimenting with actions and credits.
• SaaS companies talking about usage, consumption, resolution, automation and outcomes.

Outcome-based pricing is not a trend.

It is an evacuation route.

The brutal part is this:

The better you executed the old model, the heavier the renovation.

If you have thousands of customers trained around seats, dashboards, licences, admins, user permissions, onboarding videos, sales decks, procurement forms and renewal conversations, you cannot just wake up one morning and become AI-native.

You have to reprice the plane while flying it.

Helen and I know a smaller version of this feeling.

Collab365 spent years building training for Microsoft 365 professionals.

Deep libraries.
Real audience.
Steady revenue.
Courses people genuinely valued.

Then Copilot landed.

Search traffic crumbled.
AI started gatekeeping inboxes.
Training demand shifted.
The half-life of content went from years to weeks.

We lost directors.

We shrank.

We rebuilt.

Now it is just me and Helen in Telford rebuilding Collab365 as an AI-native platform called Spaces.

So when I read that tweet, I was not reading theory.

I was reading the same shape of problem we have already lived through once, now coming for companies a hundred times our size.

The customer does not care how beautifully your old model worked.

They care whether the new way gets them to the outcome faster.

That is the bit I think a lot of SaaS companies are underestimating.

The first-order question is:

“How do we add AI to our product?”

The second-order question is much more dangerous:

“What happens to our pricing, our interface, our onboarding, our sales motion and our renewal base when AI does the work our users used to do?”

That is where the real disruption sits.

Not in the demo.

In the renewal.