The most important sector-level stock story on September 8 was the selloff in enterprise software after investors reassessed how quickly frontier AI models could replace parts of traditional software workflows.
Salesforce, Intuit and ServiceNow fell roughly 4% to 5%, while the S&P 500 software and services group declined 1.4% for a second consecutive session.
The trigger was not an earnings miss.
It was a change in perceived competitive risk after the rollout of GPT-6 Astra, OpenAI’s newest frontier model.
Astra is designed for computer use, browsing, software engineering and complex professional work. OpenAI says the model can operate software, fill forms, update customer records, conduct research and execute multi-step workplace tasks.
Those capabilities create an obvious question for software investors:
If AI agents can perform workflows directly, how much value remains in the software layer that previously organized those workflows?
What Actually Changed With GPT-6 Astra?
The biggest change is not simply better text generation.
Astra is designed to interact with computers and business applications.
That matters because the historical software model was built around a simple structure:
the user opens an application;
the application presents a workflow;
the user performs tasks inside that workflow.
AI agents can invert that relationship.
The user can describe an objective.
The AI can navigate multiple applications and complete the work.
That creates pressure on software products whose differentiation comes mainly from workflow interfaces rather than unique data, distribution or regulated infrastructure.
Why Did Salesforce Fall?
Salesforce is one of the most obvious companies investors use to test the agentic-AI thesis.
CRM systems contain valuable customer data, but they also require users to update fields, prepare reports, create tasks and move information between systems.
If an AI agent can perform more of that work automatically, Salesforce could benefit because its data becomes more valuable.
But the risk is that users need fewer paid seats or less application-layer complexity.
That is the tension.
AI can make Salesforce more useful.
AI can also reduce the amount of human interaction with Salesforce.
Why Did Intuit Fall?
Intuit has a similar two-sided exposure.
Tax preparation, bookkeeping and small-business finance involve structured workflows.
A capable AI agent can potentially gather documents, categorize transactions, reconcile accounts and prepare outputs.
Intuit can integrate AI and improve its own products.
But investors have to ask whether the company can maintain its pricing power if general-purpose AI performs more of the same work.
The durable moat is not the interface.
It is the data, compliance knowledge, embedded financial relationships and trust.
Why Did ServiceNow Fall?
ServiceNow sits at the center of enterprise workflow automation.
That makes it both a potential winner and a potential target.
If companies deploy many AI agents, they will need governance, permissions, orchestration and audit trails.
ServiceNow can become the control layer for those agents.
But if general-purpose AI can complete simple workflows without requiring a large specialized interface layer, some parts of traditional workflow software may become commoditized.
The stock market is now trying to price which effect is larger.
Is the Selloff Rational?
Yes, but the size of the long-term risk is still uncertain.
The market is rational to lower valuation multiples when technology changes the competitive structure of an industry.
Software has historically earned very high multiples because of recurring revenue, high gross margins and strong switching costs.
If AI reduces switching costs or lowers the number of seats required, those multiples should be lower.
But that does not mean every software company becomes obsolete.
The companies with proprietary data, deep integrations, regulatory complexity and strong distribution can still be highly valuable.
The likely result is not “AI kills software.”
It is that software becomes more differentiated between durable systems of record and replaceable interface layers.
Why Is This Bigger Than One Model Release?
Because Astra is evidence of a broader capability trend.
The model can interact with software directly instead of only generating suggestions.
That shifts AI from copilot toward operator.
Once that happens, the economic value of software interfaces changes.
This is a structural theme that can persist for years.
What Does This Mean for SaaS Valuations?
Investors should focus on four questions.
First, does AI reduce paid-seat growth?
Second, does AI increase usage enough to offset fewer seats?
Third, does the software company own unique data?
Fourth, does the company become the orchestration layer for AI agents?
Companies that fail those tests are more vulnerable.
Companies that pass them may eventually become stronger.
What Is the Bull Case?
The bull case is that AI increases the amount of software activity.
If AI agents create more workflows, more customer interactions and more automated tasks, the underlying systems of record become more valuable.
Salesforce still owns customer data.
ServiceNow still owns workflow context.
Intuit still owns financial relationships and tax workflows.
AI may increase the frequency with which these systems are used.
The value may shift from user seats to usage, transactions or platform access.
What Is the Bear Case?
The bear case is that general-purpose agents reduce the need for expensive application-layer software.
A business may no longer need every employee to have a separate interface-heavy seat.
AI can interact with underlying systems on behalf of users.
That threatens per-seat pricing.
It can also make switching easier if the AI layer abstracts away application differences.
What to Watch Next
Watch how Salesforce, ServiceNow and Intuit discuss agent pricing.
Watch net retention.
Watch seat growth.
Watch usage-based revenue.
Watch whether customers reduce software licenses while increasing AI spending.
Watch whether Astra and competing models gain deeper enterprise access.
The most important question is not:
“Will AI replace software?”
It is:
Which parts of software remain valuable when users no longer need to interact directly with every application interface?