Signals Inbox·July 28, 2026·Frontier AI

Is software dead now?

Software is not dead; the old SaaS playbook is. Spending is still rising, but agents, cheap code and weaker seat growth are moving value away from visible apps and toward data, infrastructure and outcomes.

Q1Why are people suddenly saying software is dead?

Software looks fragile today because AI is attacking its old pricing and product model faster than revenue can show the damage.

The mood changed because three pressures arrived together. AI can now produce decent code in minutes. Agents are starting to complete work across several applications. At the same time, slower hiring has weakened the old SaaS habit of selling another paid seat whenever a customer adds an employee.

Investors reacted before those pressures showed up clearly in company revenue. Bain found that broad software indices had fallen about 25% from their 12-month highs during the first major “SaaSpocalypse” sell-off. A recent Investopedia review of S&P Dow Jones data found that the S&P 500 software index remained over 25% below its 2025 high while the wider market was up.

The sell-off made the claim feel more proven than it really was. Software shares can fall because investors expect weaker future growth, lower prices or higher AI costs. None of those outcomes requires customers to stop using software.

What is really up for grabs is the value captured by visible apps, human seats and traditional SaaS vendors. Agents, data platforms and outcome-based products all want a share of it.

Q2What do people actually mean when they say software is dead?

“Software is dead” currently mixes together several claims, and only some of them hold up.

Some people mean that customers will stop buying digital tools; the latest spending data contradicts that. Others mean traditional SaaS pricing will fail, which is much more plausible when an agent can perform work that once required several paid users.

The phrase can also refer to interfaces. Employees may stop opening many apps because an agent can use those apps for them. It can refer to software jobs, where junior coding work is already under pressure. Or it can refer to valuations, which can fall long before revenue does.

Those claims need to stay separate. Otherwise, a falling stock price gets mistaken for falling software demand, or a disappearing screen gets mistaken for a disappearing system.

What “software is dead” could mean

Q3Are companies spending less on software now?

Companies are spending more on software now, at a pace that rules out any broad collapse in demand.

In its latest forecast, Gartner expects software spending to rise from about $1.27 trillion in 2025 to $1.47 trillion in 2026. The increase reaches 15.5%, adding nearly $200 billion in one year.

Spread across the year, the extra $197 billion equals roughly $540 million of additional software spending per day. A market adding that much demand cannot reasonably be described as dead.

The mix is changing fast, though. More money is flowing into AI-ready software, data platforms, security, cloud services and tools that coordinate agents. Some older applications will lose budget as customers redirect spending toward those areas.

Companies are buying more software. The fight is over which vendors receive the money.

Q4Are companies cancelling their SaaS tools?

Companies are pruning duplicate SaaS tools these days while keeping stacks that still average more than 100 applications.

BetterCloud found that the average organization used 106 SaaS applications, down from 112 in the previous comparison period. The roughly 5% drop still leaves the typical company with more than 100 cloud tools.

The pattern looks like cleanup after years of easy purchasing. Teams are removing unused licenses, combining overlapping products and asking whether one large platform can cover work previously split among several smaller vendors. BetterCloud also found that 70% of organizations preferred a unified platform where possible.

Point solutions feel the pressure first. A separate meeting-summary app, simple reporting tool or narrow workflow product can disappear when a suite includes a good-enough version in an existing contract.

Customers still need the work done. They are simply less willing to pay five vendors for it.

Q5Have the biggest software companies stopped growing?

The biggest software companies are still growing strongly today, and their latest numbers are too large to dismiss as old contracts slowly running out.

ServiceNow increased quarterly subscription revenue by 24.5% to $3.88 billion and reported about $29 billion of future contracted revenue. SAP grew cloud revenue by 22%, while its cloud backlog due within 12 months rose 27% to €22.9 billion. Salesforce increased subscription and support revenue by 14% to $10.6 billion, with $67.9 billion of future contracted revenue. Adobe said annual recurring revenue from its AI-first products had passed $500 million and tripled year over year.

Those four companies sell different products, serve different departments and report on different schedules. Their results still point in the same direction: large organizations continue signing, renewing and expanding software contracts.

But the growth is going to fewer winners. Vendors with trusted data, deep integrations and large customer bases have an easier route into AI. A small product selling one reproducible feature faces a much harder market.

What the largest software vendors are showing

Q6Why are software stocks falling if customers still pay?

Software stocks are falling because investors expect slower growth, weaker pricing and higher AI costs, even while customers keep paying.

Bain found that established software vendors were still keeping roughly 90% or more of their customers’ recurring revenue during the sell-off. The problem came after renewal: customers added fewer seats, questioned every upgrade and bought fewer extra products.

Lower expansion can crush a software valuation. Traditional SaaS companies were prized because revenue rose predictably as customers hired more people, added departments and accepted annual price increases. AI can loosen each link in that chain. One agent may serve work previously done by several users. New competitors can copy familiar features faster. Running AI also creates a real computing bill each time the product acts.

Markets price those future pressures today. A company can keep growing while its shares fall if investors once expected 25% growth and now expect 12%.

Q7Are AI agents already replacing business software?

AI agents currently replace pieces of work, but they rarely replace a full business system.

Stanford’s latest AI Index found that 88% of surveyed organizations used AI and 70% used generative AI in at least one business function. Yet agent deployment remained in the single digits across nearly every function. Companies have moved quickly on copilots and chat tools; autonomous execution is much less common.

A recent paper based on S&P 500 SEC filings found that 11% of companies had deeply integrated AI into business processes, while another 10% used it directly to produce goods or deliver services. Those figures show real adoption, although most large companies still have not put AI deep into their day-to-day operations.

Reliability explains much of the gap. Salesforce researchers tested agents on 300 realistic customer relationship management tasks. The best closed models completed at most 39% without examples. Demonstrations lifted the top result to around 50%.

At that reliability level, an agent can save time under supervision. Giving it sole control of a major enterprise workflow would be reckless.

Q8Will people stop opening software apps?

Many employees will open fewer software apps as agents take over the clicking, searching and form-filling.

Gartner estimates that up to $234 billion of enterprise application spending could be exposed to “agentic arbitrage” by 2030, equal to roughly 20% of projected enterprise application SaaS spending. The idea is simple: an agent can complete work across several systems, so fewer people need direct access to every interface.

Consider an expense approval. An employee could ask an agent to find the receipt, check the company policy, enter the claim and send it for approval. Most employees just want the expense approved correctly; they do not care whether that requires visiting three screens.

The software underneath still has work to do. It must store the transaction, enforce permissions, apply company rules and keep an audit trail. In many cases, the interface loses value while the system behind it becomes more important.

Gartner currently considers large-scale replacement of enterprise applications unlikely through 2030. A more believable outcome is that agents become the front door to many systems that remain in place.

Q9Is per-seat SaaS pricing breaking?

Per-seat SaaS pricing is already breaking in workflows where agents perform the work instead of named employees.

Salesforce now sells Agentforce through Flex Credits, with a standard action priced at about $0.10. Intercom charges $0.99 for an outcome completed by its Fin support agent. Microsoft measures Copilot Studio activity through credits consumed by agent actions.

All three tie price to work performed rather than simply counting logins. A bot may complete thousands of tasks without needing thousands of paid users.

Outcome or usage pricing makes sense for some products and looks forced for others. Figma can still charge by seat because each designer actively uses the product. A customer-support platform has a harder time defending the same model when an agent resolves a large share of incoming requests.

Usage pricing also brings a new problem. AI products incur model and infrastructure costs every time they act. Microsoft’s cloud gross margin fell from 69% to 66% year over year in its latest reported quarter, with the company pointing partly to AI infrastructure investment.

Expect more hybrid contracts: a platform fee, some human seats, usage credits and perhaps a charge for completed outcomes. Pure per-seat SaaS will survive, but it will no longer fit every kind of software.

Q10Can companies build their own software instead of buying it?

Companies can build far more internal software now, especially small tools and custom workflows, but production systems remain expensive to own.

As coding became cheaper, software activity accelerated. GitHub recorded 986 million commits in 2025, up 25%, while more than 36 million developers joined the platform and users created over 230 repositories per minute.

The explosion in activity shows what cheaper coding does in practice: people create more software. A finance team can now create a simple dashboard or approval app without waiting months for a traditional development project.

The difficult work begins after the first version. Production software still needs secure login, clean data, testing, monitoring, backups, integrations and someone who fixes it when a dependency changes.

Developers themselves remain cautious. Stack Overflow found that 46% distrusted the accuracy of AI tools, compared with 33% who trusted them. Two-thirds complained that generated solutions were “almost right,” and 45% said debugging AI-written code could take longer.

Companies will build more tools that are small, temporary or unique to their operations. They will keep buying systems where reliability and long-term maintenance matter more than the first week of coding.

Q11Is code becoming so cheap that software is a commodity?

Code is becoming cheap; dependable software still requires judgment, trusted data and someone willing to take responsibility.

AI can already generate familiar interfaces, database queries, tests and integrations. That makes a codebase less defensible when its main achievement is reproducing features that customers have seen elsewhere.

The hard part has moved away from typing code. Someone still has to decide what the product should do, account for unusual cases, connect it to reliable data, test the output and take responsibility when it fails.

Stack Overflow’s survey shows why demos can be misleading. “Almost right” code can look impressive while creating hours of hidden checking and repair. In a payroll, healthcare or security system, a small error is a product failure, not a minor inconvenience.

Cheaper construction will probably increase the number of software products while reducing the value of shallow ones. Products with proprietary data, trust, distribution or control over a critical workflow can still defend themselves.

Q12Are AI startups already replacing older software companies?

AI startups are already taking meaningful revenue from older software categories, while opening markets that barely existed a few years ago.

Cursor, an AI coding product, reportedly passed $2 billion in annualized revenue after doubling in roughly three months. Lovable, which turns natural-language instructions into applications, said it had reached about $500 million in annualized revenue and was generating around one million new projects each week.

These are private-company figures, so they deserve more caution than audited public results. Even with that caveat, the scale and speed are exceptional. Customers are clearly willing to pay for new software that removes enough work.

Replacement will happen product by product. An AI coding tool can take time and budget from a traditional development environment. An AI support agent can reduce demand for older help-desk add-ons. A general research agent can weaken several narrow information products.

Many AI startups also buy cloud capacity, databases, security tools and model access. Their growth creates software and infrastructure demand elsewhere.

Q13Which software products could AI kill first?

AI will hit narrow, easy-to-copy products first, especially when they sit between a general model and data the customer already owns.

Thin model wrappers are the clearest case. When the main product is a prompt, a basic interface and access to another company’s model, competitors can reproduce it quickly and customers can switch with little pain.

Simple internal apps are also exposed. Forms, approvals, dashboards and basic database tools are becoming much cheaper to recreate. Small point solutions face a different threat: a larger suite can add a similar feature and include it in the customer’s existing contract.

The most vulnerable products share several traits. They own little unique data, require limited integration, create low switching costs and charge for a workflow that an agent can complete elsewhere.

Software products most exposed to AI

Q14Which software products are hardest for AI to replace?

AI will struggle most with software that owns the official record, moves money or carries the blame when something goes wrong.

An enterprise resource planning system records financial transactions, inventory and supply-chain activity. A customer relationship management platform holds years of customer history and permissions. Identity software decides who can enter a system. Core banking and healthcare products must follow strict rules and preserve detailed audit trails.

Recreating the visible features is the easy part. Replacing the system means migrating years of data, rebuilding integrations, validating controls, retraining employees and accepting the risk of disruption.

Bain’s review of the software sell-off reached a similar conclusion: systems of record are harder to dislodge because customers depend on their data and workflows. Gartner also expects enterprise application software to keep growing at double-digit rates through 2030, even as agents reshape how people use it.

Agents will attack the edges of these systems first. They can simplify data entry, answer questions and complete routine steps. The core record usually stays where it is until a replacement proves equally reliable.

Q15Are software engineers losing their jobs to AI?

Junior software engineers are already losing ground to AI, though the wider profession is still growing.

Stanford found that US employment among software developers aged 22 to 25 had fallen nearly 20% from its 2024 level, while employment for older developers continued growing. Few numbers show more clearly how AI is changing the bottom of the career ladder.

Routine implementation, basic debugging and documentation are easier to automate. Companies may therefore hire fewer beginners for work that once helped them become experienced engineers.

Beyond entry-level roles, the picture is far less clear. The US Bureau of Labor Statistics still projects employment for software developers, quality assurance analysts and testers to grow 15% between 2024 and 2034, with about 129,200 openings per year. Forecasts can be wrong, especially during a fast technology shift, but those numbers do not describe a dying profession.

The role is moving toward system design, product judgment, security, evaluation and responsibility for production outcomes. People who only translate detailed instructions into routine code face a rougher market. Engineers who can decide what should be built and verify that it works remain valuable.

Q16Can software spending rise while software companies fail?

Software spending can rise quickly while a large number of software companies lose customers, shrink or disappear.

The contradiction disappears once we follow the money. A company may cancel five small applications, buy an AI platform, increase cloud usage and spend more overall. The total market grows even though several vendors lose their place in the budget.

The money is moving in three directions. Large suites can bundle features that once supported separate companies. AI-native products can charge for results rather than seats. Infrastructure, data and security vendors collect more revenue as agents perform more work.

A narrow SaaS company can therefore fail without proving that customers no longer value software. Its feature may simply have become cheaper, moved into another product or stopped deserving a separate contract.

Category growth offers no protection to a vendor whose product loses its place in the stack. We are likely to see record software demand and unusually high software-company mortality at the same time.

Q17Is software dead now?

No, software is not dead now.

The old SaaS formula is weakening, while the broader software market is still expanding fast.

Worldwide software spending is adding nearly $200 billion in a year, major enterprise vendors are still posting double-digit growth, and new AI software companies are reaching substantial revenue at unusual speed. Those facts rule out a collapse in software demand.

The real damage is concentrated elsewhere. Interfaces are losing importance because agents can operate applications for us. Per-seat pricing fits fewer workflows. Cheap code has made features much easier to copy. Junior developers face a tougher entry-level market. Small products that own little data or distribution can be bundled or bypassed.

Software that stores authoritative data, controls important transactions, provides security or becomes the operating layer for agents remains difficult to remove. In many cases, users will see less of it while relying on it more.

The headline is exaggerated. People will click through fewer screens, vendors will charge more often for actions and outcomes, and shallow features will be copied faster. Plenty of software companies will die during that transition, even as the world produces more software, more code and more total spending.