Can SaaS Overcome Its Paradox? Conditions for SaaS Survival in the AI Era

It has been a while since “the SaaS apocalypse” — the idea that AI agents will make existing SaaS unnecessary — became a popular topic.

I have written before that for SaaS companies to survive, simply adding AI is not enough. What matters is how they change the existing seat-based pricing model, and how they use AI to add new value (Beyond the “End of SaaS” Narrative: A Future Where Rule-Based Accuracy and AI Agents Converge).

Looking at this quarter’s SaaS earnings, I think an answer is starting to emerge.

The companies that take AI as an ally — adding new value on top without putting too much pressure on their seat-based pricing — and the companies that fail to capture the value of AI-driven efficiency in their own revenue are starting to part ways.

The key concept here is the SaaS paradox.

For SaaS companies, AI is a powerful tool that improves customer productivity. At the same time, the more efficient customers become, the more pressure builds on growth models that assume more users and more seats.

In other words, AI raises the value of SaaS, while also forcing the existing revenue model to be redesigned.

So the point is not simply whether a company has added AI features.

Has the company built new monetization points using its own data and workflows? Has it captured the new operational layer that the AI era specifically requires?

In my view, this is the key fork in the road for SaaS companies going forward.

Datadog uses AI as a guard rather than being eaten by it

The clearest winner this quarter was Datadog.

Datadog’s stock rose sharply after earnings. According to Datadog’s Q1 FY2026 results, Q1 revenue was $1.006 billion (up 32% year over year), free cash flow was $289 million, and the full-year outlook was raised.

What is interesting about Datadog is that it does not use AI to make existing tasks more efficient. It turns the new anxieties of running AI in production into a revenue source.

Once a company starts using AI apps and AI agents in production, it is not enough to simply run a model. They have to constantly monitor things like:

• Which model is running
• Which prompts are failing
• How many tokens are being consumed
• Whether answer quality is degrading
• Whether GPUs are being used efficiently
• What is causing incidents
• Whether hallucinations or unsafe responses are happening

The volume of logs, prompts, responses, latency, errors, cost, and quality drift that AI generates quickly exceeds what a human can check one by one.

This is where Datadog’s strength comes in.

Datadog is extending its existing cloud monitoring business into the monitoring of AI apps and AI agents. LLM Observability monitors prompts, responses, latency, errors, quality, and cost for AI apps. Bits AI works as an AI agent that helps engineers investigate incidents and reduce operational load.

In other words, Datadog is not on the side that runs AI. It is on the side that watches AI.

For most AI app companies, more token consumption and more GPU usage means higher cost.

For Datadog, however, the more AI is used, the more there is to monitor — more logs, metrics, traces, and token-related data.

This is a very clear winning structure in the AI era.

It is not “a SaaS that adopted AI.” It is “a SaaS that is needed to operate AI.”

Atlassian turns its internal data into AI value

Atlassian is taking a different path from Datadog to add AI value.

Atlassian’s stock also rose sharply in after-hours trading following earnings.

The driver is the market’s appreciation of Rovo, Atlassian’s AI search and agent feature. Atlassian’s Q3 FY26 shareholder letter explains that customers using Rovo see ARR growth roughly twice as high as customers who do not. Monthly AI credit usage is also growing more than 20% month over month.

Rovo is not just a chatbot.

It is an AI engine that searches across internal information scattered in Jira, Confluence, Slack, Google Drive, GitHub, and so on, understands the work context, and connects to the next action.

Jira holds the history of tasks, bugs, spec changes, owners, progress, and decisions.
Confluence holds specs, meeting notes, internal knowledge, and the reasoning behind past decisions.

These are not the kind of open data that a general-purpose LLM can easily learn from.

But they are directly tied to a company’s actual work.

In other words, Atlassian is trying to turn the company-specific work data accumulated over years into new AI-powered value.

In my view, this is one of the strong directions for SaaS to survive in the AI era.

A general-purpose AI model is smart, but it does not have the company-specific work context.

SaaS, on the other hand, has unique data accumulated through daily use.

If this unique data can be activated through AI, the result is an AI that handles that company’s specific work knowledge.

ServiceNow is trying to control the entry point of work data for AI agents

ServiceNow is taking a direction close to Atlassian.

ServiceNow announced Action Fabric as a layer that lets external AI agents access data inside ServiceNow and run workflows. Anthropic, as the first design partner, is connecting Claude Cowork to ServiceNow’s work execution layer.

What ServiceNow is trying to do is build an interface that lets AI agents access enterprise work data and execute tasks.

The structure is: data, workflows, approvals, permissions, and execution history inside ServiceNow are made safely usable by AI agents, and ServiceNow charges for those actions.

This is a strategy that treats SaaS-held data and work context as a moat.

No matter how smart an AI agent gets, it cannot finish real work unless it can connect to a company’s actual work data and approval flows.

ServiceNow is trying to control that connection point.

In this sense, ServiceNow is closer to Atlassian than to Datadog.

Atlassian uses AI to activate the internal information accumulated in Jira and Confluence.
ServiceNow turns its own work data and workflows into an interface that AI agents can use.

Both share the direction of turning SaaS-held unique data and work context into value for the AI era.

The SaaS paradox: HubSpot faces concerns that AI will hit existing seat-based pricing

In contrast, HubSpot’s stock fell sharply after earnings.

In Q1 2026, HubSpot reported revenue of $881 million, up 23% year over year, and customer count grew to roughly 299,000. The numbers alone do not look bad (HubSpot Q1 2026 results).

Even so, the stock fell sharply, which I think reflects the concern that AI will pressure the traditional seat-based pricing model (Barron’s coverage).

HubSpot is a typical SaaS company offering CRM, Marketing Hub, Sales Hub, Service Hub, and so on.

Its value has come from being software that the humans in sales, marketing, and customer support use day to day.

That means revenue grows when more people use it, when teams expand, and when seat counts and usage scope increase.

But what happens when AI writes the sales email,
AI classifies the leads,
AI automates customer support,
AI generates marketing copy,
AI handles inbound inquiries?

Customers can run the same operations with fewer people.

For customers, that is efficiency.
For HubSpot, it is the risk of fewer seats.

For example, if work that used to require three people in sales, marketing, and support can now be done by one person with AI, that is a productivity gain for the customer.

But it may mean fewer IDs and fewer seats for HubSpot.

This is the SaaS paradox.

Adopting AI raises customer value.
But if that AI replaces too much human work, the seat-based pricing model is broken from within.

This is not just HubSpot’s problem.

It is a structural challenge that every SaaS that has charged based on human workload or number of users now faces.

Salesforce’s challenge: it is unclear whether AI is additive to CRM or a replacement for it

Salesforce is also an important company for thinking through this issue.

Through Agentforce, Salesforce is trying to make AI-agent-driven work execution a new revenue source. The company explained in its FY2026 Q4 earnings that Agentforce ARR was $800 million, up 169% year over year.

Looking at this number alone, Salesforce is not failing at AI.

But Salesforce faces the same SaaS paradox concern as HubSpot.

The core value of Salesforce has been in the place where humans enter, refer to, update, and manage customer information for sales, marketing, customer support, HR, and operations.

In other words, Salesforce has fundamentally grown as a SaaS used by human workers in business roles.

When AI agents enter this space, two possibilities open up.

One is that AI uses data on Salesforce to improve outcomes in sales and customer support.

In this case, AI is additive — it raises Salesforce’s value.

The other is that AI replaces human work, reducing the workload of sales reps and support agents.

In this case, AI is efficiency for the customer, but it can pressure Salesforce’s seat counts and the growth of existing licenses.

The difficulty for Salesforce is that it is not yet clearly visible whether Agentforce will truly create new revenue on top of existing CRM, or whether it will partially replace the existing human-facing CRM usage.

On top of this, when AI agents actually take over sales and support work, what customers want from CRM also changes.

Until now, the value of CRM was that humans could manage customer information easily.

In the AI era, what matters becomes whether AI agents can understand customer information and autonomously execute the next action.

This shift is a big opportunity for Salesforce, but it puts pressure on the existing seat-based pricing model.

The more AI adoption progresses, the more the difficulty of moving between the old human-facing CRM model and the new AI-agent pricing model is becoming visible.

This is where Salesforce differs from Datadog or Atlassian.

For Datadog, more AI usage means more things to monitor.
For Atlassian, AI activates the value of internal data and deepens dependence on existing workflows.
For ServiceNow, AI agents access work data and execute through it — and ServiceNow controls that entry point.

Salesforce, on the other hand, holds both possibilities at once: AI raising the value of existing CRM, and AI putting pressure on existing human-facing seat pricing.

Microsoft Copilot also struggles to add value if it stays as an extension of human-facing UI

Microsoft 365 Copilot has the same problem.

Putting AI into Word, Excel, PowerPoint, and Teams is a natural move.

But these tools were originally built on the assumption that humans operate them through a screen.

Writing a document in Word.
Building a spreadsheet in Excel.
Making a deck in PowerPoint.
Running a meeting in Teams.

Putting AI into this human-facing UI does make work faster.

But faster work and Microsoft earning meaningful additional revenue are two different things.

If AI just makes writing and slide-making more efficient, that is a productivity gain for the customer.

Whether that creates enough new revenue on top of existing Office products is a much harder question.

Microsoft is pushing Copilot adoption, but it has also pointed out that AI adoption in the workplace faces organizational hurdles. Microsoft’s Work Trend Index shows that even people who recognize the need for AI often consider it safer to prioritize their existing goals.

Putting AI into Office alone does not become AI-native value.

Real value emerges only when AI is connected to enterprise data, permissions, workflows, decisions, and execution.

So the question is: are we just putting AI on top of human-facing UI? Or can AI evolve into a layer that understands the work, executes it, and continuously improves it?

This is the fork in the road.

Two winning paths for SaaS to survive

Looking at this quarter’s earnings and market reactions, I see two main winning paths for SaaS in the AI era.

The first is to turn the SaaS’s unique data and work context into AI value.

Atlassian and ServiceNow are on this path.

SaaS holds company-specific data accumulated over years of daily use:

Project history.
Customer support history.
Sales activity history.
Incident response history.
Internal knowledge.
Approval flows.
Decision context.
Execution history.

This is information that general-purpose models do not have.

And it is directly tied to actual work.

A SaaS that can activate this private, work-adjacent, continuously updated data through AI may actually raise its own value.

In the AI era, SaaS data is not just stored information. It becomes a moat.

The key is whether that moat can be connected to search, reasoning, execution, and pricing.

The second is to capture the new operational layer that the AI era specifically requires.

Datadog is on this path.

As AI agents and AI apps grow, simply running a model is not enough.

Companies need to monitor AI behavior, check quality, track cost, detect incidents, optimize GPU usage, and contain risk.

This is not something humans can keep up with manually.

Logs, prompts, responses, token consumption, errors, latency, and quality changes generated by AI are new monitoring targets specific to the AI era.

Datadog is going after this AI-native operational territory.

So Atlassian and ServiceNow are companies turning SaaS-held work data into a moat for the AI era.

Datadog is a new infrastructure SaaS that monitors AI operation itself.

Both are “SaaS that takes AI as an ally,” but the nature of each is different.

These need to be thought of separately.

The SaaS paradox — adopting AI raises customer productivity while pressuring the traditional seat-based pricing model — is the central dilemma. In my view, whether a company can overcome this paradox and turn its unique data and the AI-native operational layer into revenue will be the deciding fork in the road for SaaS survival.