Is SaaS dead because of AI agents?

No. The data, permissions, and audit trails inside CRM, ERP, and HR systems are too valuable to throw away. SaaS is being demoted instead: the interface and business logic move to an AI coordination layer, and the apps become back-end capabilities that layer calls. Vendors survive, but with less control over the customer.

What is the intelligence layer in business software?

It's a coordination layer above departmental apps. It understands how sales, delivery, staffing, and finance affect one another, turns goals into actions across tools, carries them out within agreed limits, and measures whether results showed up. Think of it as an operating system for business decisions rather than another application.

Why will AI change how SaaS is priced?

Seat pricing assumes a person does the work. When software does the work, fewer seats is the goal, so vendors move to charging per outcome or per action. Intercom bills its Fin agent at $0.99 per outcome, and Salesforce bills Agentforce at $0.10 per action. The open problem is defining outcomes both sides trust.

Who will win the AI layer in enterprise software?

Probably not traditional SaaS vendors bolting on AI, which are limited to their own domain, or standalone agent startups, which have reach but little context and no responsibility for results. The winners will combine business context, cross-functional decisions, execution through existing tools, and accountability for outcomes.

What are the risks of agentic AI in business operations?

Unreliable execution, messy data, a larger security attack surface, and weak governance. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027. The companies that succeed treat autonomy as something the system earns through audit trails, permissions, and clear escalation rules.

SaaS Isn't Dying, It's Being Demoted: The AI Layer War

Your CRM isn't going to die. Something more awkward is going to happen to it: it will be demoted.

It will keep your data, enforce your permissions, and log your history, much as it does now. But it will stop being where work gets decided. That moves up a level, to an intelligence layer that understands the whole business, works toward goals, and answers for results. The CRM becomes something that layer calls. So do your ERP, your HRIS, and your marketing stack.

That's the real AI shift in enterprise software. It isn't SaaS with a chatbot. It's a fight over who runs the business, and most of the companies talking loudest about it are badly placed to win.

The department was always the wrong unit

Business software was split up by function because that's how budgets were split up, not because that's how businesses work. Sales bought CRM. Finance bought ERP. HR bought an HRIS. Each came with its own interface, data model, and view of the world.

Then reality shows up. A services firm closes a large deal late in the quarter. Delivery needs three engineers in six weeks. Staffing has to pull them off another project. Marketing is still spending to win more of a segment you can no longer serve. Finance finds that the payment terms make month two's payroll uncomfortable.

Four systems each see one slice. Nobody's software sees the deal.

So people fill the gap: an ops lead, a Monday leadership meeting, a spreadsheet that one person understands. The most important logic in your company runs in people's heads and in meetings, because your software was never built to hold it.

That's the opening. For the first time, software can read across systems, weigh trade-offs, and act through APIs. The gap departmental SaaS left behind can now be filled, and whoever fills it captures the value.

Be careful with incumbent "collapse" talk

Microsoft's Satya Nadella said it most bluntly. On the BG2 podcast in late 2024 he described business applications as CRUD databases with a bunch of business logic and said the logic is moving to agents that work across back ends. ServiceNow's Bill McDermott made the same case: traditional apps turn into databases that feed a platform where the real work happens.

Listen to what's actually being said. "Applications will collapse" means your applications will collapse, into my platform.

That's a land grab dressed up as a forecast. It doesn't make the forecast wrong. It tells you what the prize is. Nobody is fighting over the app. Everyone is fighting over the layer above it.

Five things that stop working

1. The department stops being the unit of software

The useful question moves from "what's in my pipeline?" to "if this pipeline closes, what breaks?" Software that only answers the first question becomes a feed for software that answers the second.

Gartner expects 33% of enterprise apps to include agentic AI by 2028, up from under 1% in 2024. Most of those will be agents trapped inside a single app. That's the transition phase, not where this ends.

2. The workflow stops being the unit of work

For twenty years we drew boxes and arrows and called it automation. In practice, workflow software made people do the thinking ahead of time and froze it into a process.

Goal-driven systems flip that around. You state the outcome and the limits: "hit this hiring target, keep gross margin above 35%, don't touch the Q4 marketing budget." The system works out the steps. Your job moves from designing processes to stating constraints.

Most companies have never written their real constraints down. That's about to become an urgent problem.

3. The interface stops being the moat

SaaS companies spent billions making screens that were pleasant to live in, because whoever owned the screen owned the user. Gartner's line is blunt: agentic AI will remove the need to interact with websites and applications.

The connections are becoming common infrastructure too. In December 2025, Anthropic donated the Model Context Protocol to the Linux Foundation's Agentic AI Foundation, with OpenAI, Block, Google, Microsoft, and AWS behind it. When any orchestrator can reach any app through a neutral standard, the screen isn't a moat. It's just one way in.

4. The seat stops being the unit of value

Seat pricing assumes a person does the work. When software does it, cutting seats is the point. You can't charge per seat for a product designed to remove seats.

The meters are already changing. Intercom bills its Fin agent at $0.99 per outcome. Salesforce moved Agentforce to $0.10 per action in May 2025.

It's messy: Salesforce has since added per-user licences back, and Fin customers complain about unpredictable bills. Don't read that as a retreat. It's the industry discovering that you can only charge for outcomes you can define. Support resolutions are easy to define. "Grew revenue" isn't, yet. Whoever builds trustworthy outcome measurement gets to set the price.

5. The human stops being the default operator

The end state isn't humans out of the loop. It's humans moved from doing the work to supervising it: systems run, monitor, adjust, and escalate the calls that need judgment.

Gartner expects at least 15% of day-to-day work decisions to be autonomous by 2028. That sounds small until you notice it was 0% in 2024, and that the 15% will be the repetitive, high-volume decisions that take up most of people's working days.

Four bets

These are my predictions, not settled facts. I'm putting them on record so they can be checked later.

Bet 1: A new category forms above systems of record. Call it the system of accountability. Systems of record answer "what happened?" The next category answers "what should happen, who decided, and did it work?" I expect it to be a recognised software category within five years, with its own budget line.

Bet 2: The buyer changes. Department heads buy departmental software. A layer that trades sales against staffing against cash flow has no natural departmental owner, so the COO or CFO buys it. Vendors still selling to VPs of Sales will find the real decision being made one level higher.

Bet 3: Departmental SaaS vendors become suppliers. Think of component makers selling into a car manufacturer. A well-run, agent-friendly app with clean APIs and strong permissions stays profitable. But the brand, the relationship, and most of the margin go to whoever assembles the whole.

Bet 4: The org chart gets refactored. Conway's law says organisations build systems that mirror how they communicate. The reverse is also true: departmental software has helped keep companies departmental. Once one layer handles cross-functional coordination, many roles that exist mainly to pass information between silos will shrink. That's the uncomfortable part almost nobody says out loud.

Why most attempts will fail

None of this happens because a demo looked good. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, and estimates that only about 130 of the thousands of "agentic" vendors are offering the real thing.

The failures will come from four places:

Context. Half-filled CRMs and finance data stuck in spreadsheets. An intelligence layer on bad data makes bad decisions faster and with more confidence.
Reliability. A system that's right 95% of the time across a thousand actions a day is wrong fifty times a day.
Security. An agent that can act across sales, finance, and HR is one breach away from all three.
Governance. The hardest one. Executing is getting easy. Answering "why did it do that, and who approved it?" is still hard.

Having shipped multi-agent systems in production, I can say the model is rarely what breaks. Permissions, audit trails, rollback, and the escalation threshold are what break.

The winner owns accountability, not intelligence

So who wins? I don't think it's the SaaS incumbent adding agents, because it's limited to its own domain. A CRM's agent will always see the world through the CRM. And I don't think it's the agent startup with broad reach and nothing at stake, because an agent that can't answer for results is just a feature waiting to be copied.

The winner puts four things together across the systems companies already run:

Layer	The question it answers
Context	How does this business actually fit together?
Decisions	What's the right trade-off across functions?
Execution	Can it act through the tools already in place?
Accountability	Did it work, can we see why, and who's responsible?

Intelligence is becoming a commodity. Models get cheaper and better every quarter. Accountability isn't a commodity. It's built slowly, through results that can be measured, actions that can be audited, and a track record that lets a CFO grant a bit more autonomy each quarter. That's the moat.

What to do on Monday

If you build SaaS: stop treating your interface as the product. Your data and your actions are the product now. Make them excellent for agents to call: clean APIs, fine-grained permissions, audit logs that explain themselves. Fighting the orchestrators is a fight you'll lose.

If you run a business: write down the constraints people currently apply by instinct. Margin floors, approval limits, staffing rules, the things "everyone knows." That's the instruction set for whatever coordinates your business next, and you'll run better for having it even if you never deploy an agent.

If you're building the layer: don't lead with autonomy. Lead with visibility. A system that says "this deal creates a staffing gap in six weeks" earns trust. Then recommend. Then act. Autonomy is granted, not shipped.

Ownership is the product

The current era of SaaS split business software into departments. The next one will be built around the connections between them.

Applications will survive, demoted to capabilities that something smarter directs. The prize isn't the best app or the cleverest agent. It's the right to coordinate the whole business.

The next stage of SaaS isn't software with AI. It's software that takes ownership of outcomes. Whoever earns that ownership first will matter more than any CRM ever has.

SaaS Isn't Dying. It's Being Demoted.