Are SaaS dead because of AI agents?
In our agentic AI market deck, you will find everything you need to understand the market
SUMMARY
SaaS is not dead because of AI agents. What is dying is the old assumption that every employee needs a separate dashboard, a paid seat, and a collection of narrow tools to complete routine digital work.
Agents weaken the link between headcount and software revenue. One agent can serve many employees, complete thousands of actions, and reduce the number of people needed for some workflows, which makes seat-only pricing much harder to defend.
The interface is disappearing faster than the software underneath it. Employees may stop opening the CRM, help desk, contract platform, or analytics dashboard, while agents continue relying on those systems for trusted data, permissions, transactions, and audit trails.
Current revenue data does not support an industry collapse. The eight large cloud vendors examined here were all still growing, with a median growth rate of roughly 18%, while several were using agent demand to sell more data, workflow, identity, and observability products.
AI-assisted custom development creates a real threat, but mainly by setting a price ceiling on simple software. A company can now credibly replace an overpriced dashboard or approval tool; rebuilding a secure, compliant system of record with years of edge cases is a different job entirely.
The most exposed products combine generic output, shallow workflows, weak proprietary data, and a price tied to human seats. Writing tools, basic prospecting, level-one support, simple reporting, scheduling, data entry, and lightweight internal apps fit that profile.
The strongest beneficiaries sit on the other side of the same shift. Agents increase the value of authoritative data, identity controls, observability, regulated workflows, transaction systems, and software that proves what an agent was allowed to do and what actually happened.
Pricing is moving before the software category disappears. Per-seat contracts are giving way to platform fees, credits, usage charges, and outcome pricing because vendors need to capture machine activity without handing customers an unpredictable bill.
Agent economics will split the market. Generic features can carry expensive model and tool-use costs without earning much extra revenue, while verified outcomes tied to labor savings, risk, or revenue can support strong margins.
The likely result is not the end of recurring software. It is a harsher market with fewer standalone dashboards, more consolidation among point solutions, stronger systems of record, and a new generation of AI-native companies that still look commercially like SaaS even when customers call their products digital workers.
This market map, featured in our agentic AI market deck, highlights top companies and startups in the agentic AI market
Why are people saying AI agents will kill SaaS now?
AI agents make the death-of-SaaS argument feel serious today because they can perform parts of a job instead of merely helping an employee do it faster.
Traditional SaaS sold better tools to people. A support platform helped a representative answer tickets. A CRM helped a salesperson manage prospects. A design application helped a marketer create assets. The new pitch is much more aggressive: the software can answer the ticket, research the prospect, update the CRM, and produce the asset with limited human involvement.
Customer service gave us the clearest early example. Klarna said its AI assistant handled 2.3 million conversations in its first month, covering two-thirds of support chats and doing work comparable to 700 full-time agents. Average resolution time fell from 11 minutes to under two minutes. Intercom, HubSpot, Salesforce, Sierra, and Zendesk have since pushed the same idea: customers can buy completed resolutions or outcomes rather than another tool for a human support team.
The pressure now reaches far beyond call centers. Coding agents can finish multi-file development tasks, sales agents can research and qualify accounts, and workplace agents can gather information across several company systems. A recent Microsoft study of tens of thousands of its engineers linked command-line coding-agent adoption with roughly 24% more merged pull requests. A separate field study covering 802 developers and 196,000 pull requests found that output per engineer eventually reached just over twice its earlier level, although the researchers warned that the design could not assign every gain directly to AI.
The result threatens a basic SaaS assumption: more business output usually requires more employees, and more employees create more software seats. AI agents weaken that link.
What would it actually mean for AI agents to kill SaaS?
AI agents are nowhere close to killing SaaS if we mean cloud software sold through recurring contracts.
Cloud delivery remains the default because agents need hosted models, databases, APIs, identity systems, monitoring, and constant updates. Recurring contracts are thriving too. Glean recently passed $300 million in annual recurring revenue, Sierra reported more than $150 million, and Cursor has said its annualized revenue moved beyond $1 billion. These companies look different from the SaaS startups of the 2010s, yet they sell cloud software through recurring commercial relationships.
The vulnerable parts are more specific. Employees may stop opening a separate dashboard for every task. Per-seat pricing becomes harder to defend when one agent can complete work for many people. Narrow features can disappear into larger products or become tools that a general agent calls in the background.
For this article, “SaaS is dead” means a broad and sustained decline in recurring software spending. Current evidence shows customers changing interfaces, pricing, and product mixes while continuing to spend.
If you want more recent data on this point, please see our latest agentic AI market report.
As this chart shows, and as featured in our agentic AI market deck, search interest in AI agents has been rising rapidly
Is SaaS revenue already falling because of AI agents?
SaaS revenue is growing too quickly today for us to call the industry dead, even though weaker products are already losing ground.
We checked the most recent reported revenue growth of eight large cloud software companies across CRM, workflow automation, data, observability, marketing, creativity, identity, and agreements. Every company remained in positive territory. The range ran from 9% at Docusign to 33% at Snowflake, with a median of 18%.
ServiceNow’s newest result is particularly awkward for the collapse thesis. ServiceNow reported 24.5% subscription revenue growth in its latest quarter, $29 billion in revenue already under contract for future recognition, and more than $1 billion in annual contract value from its AI products. It also said agentic deployments had increased ninefold in nine months. Salesforce grew quarterly revenue 13%, HubSpot grew 23%, and Datadog grew 32% while selling more products for monitoring AI workloads.
A growing market can still punish plenty of vendors. Adobe is expanding more slowly than Snowflake. Docusign and Okta remain healthy but mature. Small point solutions face much harsher pressure than these platforms. Even so, a sector with large vendors growing around 18% at the median has not entered an AI-driven revenue collapse.
| Company | Latest revenue growth | What the result tells us |
|---|---|---|
| Snowflake | 33% | Demand for governed cloud data remains exceptionally strong |
| Datadog | 32% | More AI workloads create more systems to monitor |
| ServiceNow | 24% total, 24.5% subscription | Enterprise workflows are absorbing agents rather than disappearing |
| HubSpot | 23% | A traditional SaaS vendor can grow while shifting toward agents |
| Salesforce | 13% | CRM demand remains large, although organic growth is slower |
| Adobe | 13% reported | Generative AI has not erased paid creative software |
| Okta | 11% | Agent identities are expanding the security problem |
| Docusign | 9% | Mature SaaS can keep growing while its interface becomes less central |
Are AI agents replacing SaaS products or changing how people use them?
AI agents are currently replacing many SaaS screens faster than they are replacing the systems behind those screens.
A salesperson may soon ask an agent which deals are slipping instead of opening a CRM, building filters, exporting records, and preparing a summary. The employee sees less of Salesforce or HubSpot, but the agent still needs account history, permissions, emails, meeting notes, prices, and a reliable place to record its actions.
The same pattern appears in contracts and customer support. Docusign now connects its agreement system to ChatGPT, Claude, Gemini, Slack, Salesforce, SAP, and legal AI tools. Intercom’s Fin can work across different help desks. General agents increasingly sit above the application and call its data or functions when needed.
Once an agent becomes the starting point, customer ownership shifts. Employees may begin in ChatGPT, Microsoft Copilot, Claude, Salesforce, ServiceNow, or an internal company agent. A smaller SaaS product can then become an invisible supplier behind that interface, weakening its brand and making substitution easier.
Products that own authoritative data or a critical transaction retain leverage. The agent may decide what the user sees, but it cannot safely invent a contract status, customer balance, access right, or approved price. SaaS vendors increasingly live or die by whether the agent depends on their platform or can swap it out.
This chart, included in our agentic AI market deck, illustrates yearly VC funding for agentic AI startups
Can one AI agent really replace dozens of SaaS apps?
One AI agent can hide dozens of SaaS apps from the user, but replacing all their underlying functions would require rebuilding a surprising amount of dull, difficult software.
Imagine asking one workplace agent to find a contract, compare it with company policy, contact the supplier, arrange a meeting, request approval, and update the budget. The conversation can happen in one place. Behind it, the agent touches document storage, contract management, email, calendars, identity, procurement, finance, and audit systems.
Each system contains years of edge cases. The finance tool knows accounting codes and approval limits. The identity platform knows who can see what. The contract system preserves versions and signatures. The CRM keeps the commercial history. Replacing the visible apps is much easier than reproducing every rule, integration, permission, and record they contain. That is the slightly boring bit, but it is also the moat.
The most likely consolidation therefore happens among lightweight applications. A broad platform may absorb five small tools that offer forms, summaries, reminders, or basic dashboards. Core systems that hold money, rights, commitments, and official records should survive much longer.
If you want more recent data on this point, please see our latest agentic AI market report.
Are companies cancelling SaaS because AI makes custom software cheaper?
Some companies are already cancelling SaaS and using AI to build simple replacements, but this remains far from a mass migration.
Coding agents have lowered the cost of producing an internal dashboard, approval flow, reporting tool, or basic CRM. A fresh example came from healthcare company Curative, whose CEO said the company replaced a Salesforce contract costing roughly $600,000 a year with an internal CRM built in about two months. Cases like Curative give buyers a credible alternative during renewal negotiations.
Broader market data shows no large-scale retreat from commercial software. Salesforce had $67.9 billion of contracted revenue waiting to be recognized after its latest reported quarter. HubSpot reached almost 300,000 customers and increased average subscription revenue per customer by 6%. Docusign said 40,000 customers were already investing in its newer agreement-management product, which had reached 12.6% of annual recurring revenue.
The first version is also the easy part. An internal tool needs security, support, migrations, testing, permissions, compliance, integrations, and years of maintenance. AI reduces the engineering bill, but a home-built system can become another piece of software the company must operate forever.
Custom building will put a hard ceiling on the price of basic functionality. Vendors with deep workflows, large integration networks, regulated data, or constant product development have a much stronger defense.
This chart, included in our agentic AI market deck, shows how Cognition is positioned in agentic AI
Which SaaS products are most exposed to AI agents?
AI agents pose the greatest threat to SaaS products that package one predictable digital task inside a separate interface.
Generic writing tools sit near the top because major models can already draft, summarize, translate, and reformat text. Basic sales prospecting products face similar pressure when an agent can find accounts, research contacts, qualify leads, draft outreach, and update the CRM. Simple dashboards also lose value when users can ask questions directly against the underlying data.
Customer support is the clearest commercial battleground. HubSpot now charges $0.50 for a resolved customer conversation, Intercom starts at $0.99 per outcome, and Sierra sells customer agents around completed jobs. That pricing lets customers compare one AI resolution with the full cost of a human support agent.
Lightweight internal tools are exposed for a different reason. AI coding products make it easier to build forms, approval flows, record editors, and small operational apps. A standalone vendor must now compete with larger platforms, general agents, and the customer’s own development team.
| SaaS category | Exposure to AI agents | Main reason |
|---|---|---|
| Generic writing and content tools | Very high | Frontier models can produce the core output directly |
| Basic sales prospecting | High | Research, qualification, drafting, and CRM updates can be combined |
| Level-one customer support | High | The work is repetitive, measurable, and easy to price per outcome |
| Simple reporting dashboards | High | Users can question the data without navigating the dashboard |
| Lightweight internal tools | High | Coding agents lower the cost of building a replacement |
| Scheduling and data entry | High | The tasks follow clear rules across connected systems |
| Regulated vertical workflows | Moderate | Agents help, but errors and edge cases remain expensive |
| Systems of record | Lower | Agents depend on authoritative data and transaction history |
If you want more recent data on this point, please see our latest agentic AI market report.
Which SaaS products become more valuable because of AI agents?
AI agents are already increasing demand for SaaS products that store trusted data, control access, monitor systems, or record important transactions.
Identity is an obvious example. Companies must decide which data an agent can read, which actions it can take, how long its access lasts, and who is responsible when something goes wrong. Okta’s latest revenue grew 11% and its contracted backlog grew 16% as it expanded its products around securing both human and non-human identities.
Agents also create more work for data and observability platforms. Snowflake’s product revenue grew 34%, its remaining performance obligations grew 38%, and its net revenue retention remained 126%. Datadog passed $1 billion in quarterly revenue while adding products for GPU monitoring, AI security, and agent operations. ServiceNow crossed $1 billion in AI annual contract value while selling the workflows and controls around those agents.
Contract, payments, payroll, healthcare, legal, insurance, and industrial software also hold up better because the records have real consequences. An agent can suggest or initiate an action, but businesses need a controlled system to validate the decision and preserve what happened.
As agents do more work, companies need clear answers to basic questions: Which source is correct? What was the agent allowed to do? Did the transaction complete? Can we audit the decision? Software that answers those questions should become more important, even when employees rarely open it themselves.
This chart, included in our agentic AI market deck, illustrates yearly funding for agentic AI startups
Are AI agents already breaking per-seat SaaS pricing?
AI agents are breaking per-seat SaaS pricing now because software usage no longer scales neatly with the number of human employees.
The shift is visible in real price lists. HubSpot charges for resolved conversations and recommended leads. Intercom charges per outcome alongside a smaller seat fee. Salesforce offers credits, conversations, and employee licenses for Agentforce. Microsoft measures Copilot Studio activity through credits whose consumption depends on what the agent does.
Vendors are experimenting because both extremes have problems. Pure seat pricing undercharges for an agent that completes thousands of tasks. Pure usage pricing can frighten customers with an unpredictable bill. Outcome pricing sounds clean until the buyer and seller disagree about what counts as a successful outcome.
Fewer employees could cut seat counts too, although the impact remains smaller than the hype suggests. Stanford’s latest AI Index found that one-third of surveyed organizations expected AI-related workforce reductions, while almost half expected little or no change. Fresh UK government data found AI adoption rising from about 12% to 35% among businesses with at least ten employees, yet average use increased only from 1.4 to 1.6 AI technologies per adopting company and had not produced widespread changes in total headcount.
Hybrid contracts now look like the most likely winner: a platform fee for access and governance, plus usage or outcome charges for agent work. SaaS vendors that cling to a seat-only model will struggle when customers hire fewer junior workers or let agents serve many employees at once.
| Pricing model | Why vendors like it | Where it can fail |
|---|---|---|
| Per human seat | Predictable and familiar | Revenue falls when agents replace or share human work |
| Per action or credit | Captures machine usage | Bills become hard to predict and compare |
| Per outcome | Links price to visible value | Defining a successful outcome can become contentious |
| Platform plus usage | Balances predictability and growth | Contracts become more complex |
Will AI agents ruin SaaS profit margins?
AI agents will squeeze some SaaS margins, but vendors can protect profits when customers pay for a valuable completed job.
Traditional software can serve another dashboard view at almost no cost. An agent may call several models, retrieve documents, use external tools, retry failed steps, and ask a human to review the result. Those costs rise every time the product performs more work.
Margins get ugly when a vendor bundles expensive agent usage into an old subscription without raising prices, or when the agent needs many hidden attempts to produce one billable result. A company can report spectacular revenue while spending heavily on models and infrastructure to earn it.
Agents can also lower the vendor’s own support, onboarding, sales, and engineering costs. Klarna estimated a $40 million annual profit improvement from its customer-service assistant. Datadog produced a 29% free-cash-flow margin in its latest quarter while expanding its AI product range. Adobe’s non-GAAP operating income grew while AI-first annual recurring revenue tripled to more than $500 million.
Margins will depend less on whether a product uses AI and more on whether customers pay more than the full cost of delivering the work. Generic features will be hard to monetize. A verified outcome tied to revenue, risk, or labor savings gives the vendor much more room.
This chart, included in our agentic AI market deck, compares the main business model options for autonomous AI agent platforms
Can established SaaS companies adapt before AI startups replace them?
Established SaaS companies currently have the upper hand over most AI startups because they already hold the data, integrations, and customer relationships that AI agents need.
Salesforce closed more than 29,000 Agentforce deals by the end of its last full year, with Agentforce annual recurring revenue reaching $800 million. ServiceNow said agentic deployments had increased ninefold in nine months. HubSpot moved its customer and prospecting agents toward outcome-based pricing while keeping them inside its existing platform.
They also control permissions, sales channels, and systems that customers already trust with important work. A startup with a smarter demo must persuade the buyer to connect sensitive systems, pass security reviews, migrate workflows, and add another supplier.
Incumbents can lose when they treat agents as a feature pack attached to yesterday’s interface. Customers will notice when a product requires the same manual work with an AI button placed on top. They will also push back if vendors protect seat revenue rather than pricing the value agents create.
Right now, the strongest incumbents are redesigning workflows without asking customers to rebuild their whole technology stack. Startups have room to win focused categories, but the blanket idea that established SaaS companies cannot adapt has already aged badly.
If you want more recent data on this point, please see our latest agentic AI market report.
Are AI-native startups replacing SaaS or becoming the next SaaS companies?
AI-native startups are becoming the next generation of SaaS companies.
Glean moved from $100 million to $300 million in annual recurring revenue in roughly fifteen months. Sierra reached $100 million in seven quarters and later reported more than $150 million. Cursor said it passed $1 billion in annualized revenue after previously crossing $500 million. Their speed is unusual even by software standards.
Their products also look different. Glean sells a context layer and agents connected to company knowledge. Sierra sells customer outcomes. Cursor sells an environment where coding agents perform a growing share of development work. Customers pay for more work and fewer clicks.
Underneath the new language, the commercial structure remains familiar. These companies host the product in the cloud, sign recurring contracts, expand inside enterprise accounts, and report annual recurring revenue. They have changed the unit of value while keeping the recurring software model.
What has changed most is the source of advantage. Building a feature is easier, while reliable context, distribution, security, evaluations, and workflow knowledge remain difficult. AI-native startups that master those layers can replace older SaaS vendors. The winners will look like software companies, even when customers describe them as digital workers.
This chart, featured in our agentic AI market deck, shows the share of revenue generated by each customer segment in the agentic AI market
Are AI agents reliable enough to run critical SaaS workflows?
AI agents are currently reliable enough for narrow jobs with clear checks, but giving them broad control over critical business systems would be reckless.
Adoption has moved quickly, yet deep deployment remains rare. Stanford’s latest AI Index found that 88% of surveyed organizations used AI somewhere and 70% used generative AI in at least one business function. Agent deployment, however, remained in the single digits across nearly every function. The newest UK business survey tells the same story from another angle: many more firms use AI, but most have added only one or two technologies rather than rebuilding their operations around agents.
Agent capability is improving fast too. METR’s latest measurements show that frontier agents can now complete much longer software tasks than earlier models, and some recent results run beyond the range its benchmark can measure reliably. Recent field studies also found lasting productivity gains from coding agents rather than a brief novelty effect.
Reliability falls apart when the work is vague, long, poorly documented, or expensive to verify. Agents can choose the wrong tool, misread a record, repeat an action, or confidently continue after an early mistake. In a ten-step workflow, one early mistake can poison everything that follows.
For critical work, companies need permissions, tests, spending limits, logs, approval points, and clear handoffs to people. SaaS platforms already contain much of that machinery. Better agents may reduce the number of manual steps, but they also raise the value of the controls around them.
Is SaaS dead because of AI agents?
No. SaaS is growing, although AI agents are forcing its biggest redesign since cloud computing.
Cloud delivery, recurring contracts, and enterprise software spending remain firmly alive. Major SaaS vendors continue to grow, and some of the newest AI-agent companies are reaching recurring-revenue milestones faster than earlier SaaS leaders ever did.
The real danger sits with a particular type of product: a narrow application with generic features, no unique data, no critical transaction, and a price tied to every human employee. Agents can absorb its interface, reproduce its output, or make it cheap for the customer to build an alternative.
We expect the next few years to produce fewer standalone dashboards, weaker seat-based pricing, and brutal consolidation among lightweight point solutions. At the same time, spending should keep flowing toward systems of record, data platforms, identity, observability, regulated workflows, and products that can charge for completed work.
The winners will be platforms that agents rely on and AI agents strong enough to become major SaaS businesses themselves. The easy formula of packaging a small workflow behind a login and charging every employee forever has reached the end of the road.
If you want more recent data on this point, please see our latest agentic AI market report.
This chart, included in our agentic AI market deck, shows how autonomous AI agent platform technology has evolved over time
OUR METHODOLOGY
Questions such as “Are SaaS dead because of AI agents?” attract strong opinions before the market evidence points clearly in one direction. We broke the debate into the dimensions that would actually show whether a structural decline is happening: revenue growth, product architecture, customer behavior, pricing, software economics, competitive pressure, and operational reliability.
For each dimension, we reviewed recent company disclosures, earnings results, product launches, pricing pages, customer adoption figures, technical studies, and large-scale surveys. We prioritized observable behavior over forecasts: revenue, contracted backlog, deployments, pricing changes, usage, measured productivity, and operational outcomes carried more weight than executive predictions or broad industry narratives.
We separated illustrative cases from market evidence. Klarna and Curative show what has become technically or commercially possible, but neither case alone proves an industry-wide shift. Broader conclusions were based on patterns repeated across large vendors, AI-native startups, independent research, and public datasets.
To test the collapse thesis directly, we compared the latest reported revenue growth of eight large cloud software companies spanning CRM, workflow automation, data, observability, marketing, creativity, identity, and agreements. The group was selected to cover different parts of the SaaS market rather than to represent every public software company. We used the median to reduce the influence of the fastest- and slowest-growing names.
We treated systems of record, identity, data infrastructure, observability, and regulated workflows differently from lightweight point solutions because agents depend on authoritative records, permissions, controls, and transaction history. This distinction helped separate software whose interface may disappear from software whose underlying function remains difficult to replace.
Pricing analysis relied on current public offers from HubSpot, Intercom, Salesforce, and Microsoft to track the shift from seats toward credits, usage, conversations, and completed outcomes. We used those published structures as evidence of experimentation, not as proof that one pricing model has already won.
Reliability and workforce conclusions drew on the Stanford AI Index, UK government adoption statistics, METR’s agent measurements, and recent field research on coding-agent adoption. We used these sources to separate capability gains from actual organizational deployment, which remains much shallower than the most aggressive agent forecasts suggest.
Key sources used for this analysis include: OpenAI’s Klarna case study, ServiceNow investor disclosures, Salesforce investor disclosures, HubSpot investor disclosures, Snowflake investor disclosures, Datadog investor disclosures, Adobe investor disclosures, Okta investor disclosures, Docusign investor disclosures, Glean’s newsroom, Sierra’s newsroom, Cursor’s blog, the Stanford AI Index, METR, the Microsoft command-line coding-agent study, the UK Department for Science, Innovation and Technology, Intercom Fin, Salesforce Agentforce, Microsoft Copilot Studio pricing, and HubSpot pricing.
In our agentic AI market deck, we identify pain points entrepreneurs should prioritize
Related blog posts
