Wall Street Is Suddenly Asking Which Software Companies GPT-6 Astra Could Make Obsolete

GPT-6 Astra Could Change the Economics of Software — and Wall Street Is Taking Notice

When AI Stops Using Software and Starts Doing the Work

Wall Street has reopened one of the most uncomfortable questions hanging over the artificial intelligence boom: what happens when increasingly capable AI systems stop merely helping people use software and begin doing the work that software companies charge people to perform? The launch of OpenAI's GPT-6 Astra has pushed that question back to the centre of the market, sending several major enterprise software stocks sharply lower.

Salesforce and Intuit fell roughly 4% on Tuesday, September 8, while ServiceNow lost about 5%. The wider S&P 500 software and services index declined 1.4%, extending its fall for a second consecutive session as investors confronted a possibility that has repeatedly unsettled the sector throughout 2026: advanced AI agents may not simply become another feature inside existing software products. They could change how many of those products are bought, priced and used.

Astra Has Turned an Old Fear Into a New Market Problem

GPT-6 Astra was introduced by OpenAI on September 3 as its most capable model yet, with major improvements in computer use, browsing, software engineering, cybersecurity, science and professional work. OpenAI says the system can perform multistep workflows across computers and professional software rather than merely answering questions inside a chatbot.

That distinction matters enormously for the software industry. Astra can fill in online forms, update records inside customer relationship management systems, organise calendars, conduct research, work with email, analyse data, create websites and perform other tasks that once required a person to move manually between specialised applications. OpenAI also says Astra completes computer-use tasks substantially faster than GPT-5.6 Sol in its testing.

The threat therefore sits deeper than better text generation.

Traditional software largely assumes that a human employee will open an application, navigate its interface and perform work through it. Enterprise software companies then charge organisations for access to those applications, often according to users, seats, modules, features or consumption.

AI agents threaten to insert themselves between the employee and the software.

Instead of asking a worker to log into several systems, locate information, update records and complete a workflow, a company could increasingly give an AI agent the objective and allow the agent to perform those actions itself.

That potentially changes where the economic value sits.

Wall Street Is Questioning the Value of the Software Seat

The uncomfortable scenario for software investors is not necessarily that Salesforce, ServiceNow, Intuit or Workday disappear.

It is that businesses eventually need fewer conventional software seats to accomplish the same amount of work.

Imagine a company employing 1,000 people across sales, finance, administration and customer support. Today, hundreds of those workers may require their own licences for numerous applications because they personally interact with those systems throughout the working day.

An increasingly autonomous AI environment could look very different.

A smaller workforce supported by powerful agents might complete more transactions, process more customer requests, prepare more reports, update more databases and perform more administrative work while requiring fewer human-facing software licences.

That possibility attacks one of the great financial characteristics of software-as-a-service: recurring revenue tied to large populations of users.

The rise of autonomous AI agents has already shown how the model could extend from software development into customer service, research, personal administration and wider business operations. Astra gives investors another reason to ask how quickly that theoretical shift could become commercially significant.

The User Interface Could Become Less Important

There is an even more fundamental danger beneath the licence question.

Much of the modern software industry has been built around graphical interfaces.

Companies spend enormous sums designing dashboards, menus, workflows, forms, buttons and screens that make complicated systems usable by people.

But an AI agent does not necessarily need to interact with software in the same way a human does.

If an agent can communicate directly with databases, APIs and business systems, the interface that once helped differentiate one software application from another may become less important.

The employee may simply say what needs to happen.

Find the overdue invoices. Contact the customers. Update the CRM. Reschedule tomorrow's meetings. Analyse last month's sales. Prepare the presentation. Identify unusual expenditure. Build the report.

The agent then decides which systems need to be used.

That represents a potential change from an application-centred computing world to an outcome-centred one.

Software does not disappear in that world. Databases, identity systems, security controls, accounting records, transaction engines and enterprise infrastructure remain essential.

But some of the value moves away from the screen where the employee works and towards the intelligence coordinating everything underneath it.

Salesforce and ServiceNow May Also Have Powerful Defences

The sell-off should not be mistaken for proof that today's large software companies are doomed.

There is a strong counterargument.

Businesses such as Salesforce and ServiceNow hold enormous quantities of customer context, permissions, workflows, integrations and institutional history. An AI agent attempting to perform serious business work still needs somewhere reliable to find customer records, authorised processes, financial information and organisational data.

That can make existing enterprise platforms more valuable rather than less.

ServiceNow, Salesforce and other major software companies are already building their own agentic AI products. The strategic battle may therefore become less about AI versus software and more about which company controls the layer through which AI performs enterprise work.

This is an important difference.

A company whose product is little more than a convenient interface around a repeatable task could face severe pressure.

A company that owns essential data, complex workflows, compliance infrastructure, security permissions or deeply embedded systems may be far more difficult to displace.

That is why the software sell-off is unlikely to end with every company receiving the same verdict.

The Most Vulnerable Software Could Be the Simplest

The part of the software market facing the greatest structural danger may be the collection of specialised applications that automate relatively narrow knowledge-work tasks without controlling essential underlying infrastructure.

Simple content generation, basic reporting, scheduling, repetitive research, lightweight customer communication, routine data transformation and other easily described tasks can increasingly be performed by general-purpose AI systems.

Companies selling those capabilities as standalone products may discover that their feature is becoming a prompt.

This is the same deeper threat behind efforts to build AI systems capable of performing the functions of entire software businesses. The economics become uncomfortable when intelligence that once required several specialised tools becomes available through one increasingly general agent.

Large enterprise platforms possess another defence: switching is difficult.

Major businesses cannot casually replace years of integrations, security configuration, databases, compliance procedures and employee processes because a new AI model performs well in a demonstration.

Replacing mission-critical software can take years.

That gives incumbents time to adapt.

Astra Could Create Winners Inside Software Too

Artificial intelligence is not producing a simple division where semiconductor companies win and software companies lose.

Some software businesses may become considerably more valuable because of AI.

Data platforms can provide the information agents require. Cybersecurity businesses can protect increasingly autonomous systems. Cloud providers can host them. Governance products can determine what agents are permitted to do. Workflow platforms can coordinate them.

The companies capable of turning themselves into the operating infrastructure for AI agents could therefore capture an enormous new market.

This helps explain why recent software performance has been so uneven. Investors have repeatedly moved between fears that AI will destroy established software economics and optimism that the strongest platforms will become essential components of the AI economy.

The financial pressure surrounding the wider AI boom has already forced Wall Street to distinguish between companies merely exposed to artificial intelligence and companies capable of converting that exposure into sustainable cash flow.

Astra adds another test.

Software companies must now show not only that they can add AI features, but that AI strengthens rather than destroys the economics of their core products.

Wall Street Is Beginning to Price the Transition

Tuesday's losses came during a difficult session for the broader US market. The S&P 500 fell 0.58%, the Dow declined about 1.2% and the Nasdaq slipped 0.3%, with rising oil prices, Middle East tensions and uncertainty over US interest rates also weighing on sentiment.

That wider pressure matters because Astra was not solely responsible for the market's decline.

The more revealing signal was what happened inside technology.

Software companies weakened while parts of the semiconductor industry rose. Intel climbed about 9% and Qualcomm gained approximately 3.2% following an AI chip agreement involving Amazon, illustrating the division investors increasingly see between companies supplying the infrastructure behind artificial intelligence and companies whose existing products could potentially be disrupted by it.

The market is effectively trying to answer two questions simultaneously.

Who supplies the machines?

And whose business gets consumed by them?

GPT-6 Astra Does Not Need to Replace a Company to Damage Its Valuation

This is where the discussion about AI replacing companies can become misleading.

Astra does not need to eliminate Salesforce, ServiceNow or Intuit for investors to become nervous.

It only needs to alter expectations about future growth.

If AI allows one employee to do the work previously performed by several people, software vendors could face pressure on seat counts. If agents make interfaces less important, some products could lose differentiation. If general-purpose models absorb features previously sold separately, specialised vendors may lose pricing power.

Even modest changes can matter when investors are valuing software companies on years of expected recurring growth.

Markets price the future long before that future is fully visible.

That is why a technological capability can hit a stock even while the affected company continues producing billions of dollars in revenue.

The Next Battle Is Over Who Owns the Agent

The deeper question created by GPT-6 Astra is therefore not whether software disappears.

It is which layer of the software economy becomes dominant.

One possibility is that today's enterprise giants successfully absorb artificial intelligence into their platforms. Salesforce agents operate Salesforce data. ServiceNow agents manage ServiceNow workflows. Microsoft agents move across corporate productivity systems. Established software becomes the trusted infrastructure underneath increasingly autonomous work.

The other possibility is more disruptive.

General-purpose agents become the primary interface through which people interact with digital systems. The user stops thinking about individual applications and instead asks an agent to achieve an outcome. Software companies then become invisible services underneath a more powerful intelligence layer, competing for access rather than controlling the customer relationship themselves.

That would represent one of the largest changes in software economics since cloud computing replaced much of the traditional installed-software model.

GPT-6 Astra has not proved that this transformation will happen.

But its ability to combine reasoning with computer use, browsing, coding and professional workflows has made the possibility harder for investors to dismiss.

Wall Street is no longer asking only how much money artificial intelligence will create.

It is increasingly asking where that money will come from.

For some software companies, the answer could eventually be uncomfortable: part of the extraordinary value created by AI may come from compressing the very margins, licences and workflows on which the previous generation of technology giants built their empires.