Will AI agents kill per-seat SaaS pricing?
In our agentic AI market deck, you will find everything you need to understand the market
SUMMARY
AI agents will not kill per-seat SaaS pricing, but they are already killing the idea that human seats alone can determine what software should cost.
The pricing shift will happen before any large collapse in employment. An agent can multiply the amount of work produced by a fixed team, raising both the customer’s value and the vendor’s infrastructure costs without creating another paid user.
That is why the first visible change is not disappearing licenses. It is the addition of credits, actions, resolutions, tasks and other machine-activity meters beside existing human seats.
Seven of the eight major platforms reviewed here already charge separately for at least some autonomous work. The terminology varies, but the economic conclusion is remarkably consistent: machine labor cannot remain unlimited inside a fixed employee subscription.
The likely successor is a three-part contract combining a platform commitment, licenses for human access and a shared allowance for agent activity. In operational software, the agent component may eventually become the largest part of the bill.
Customer support, sales prospecting and repetitive back-office processing face the most pressure because their outputs can be counted. A resolved conversation, recommended lead or processed invoice is a more useful billing unit than the number of employees watching the workflow.
Systems of record are in a stronger position than their interfaces may suggest. Employees may stop opening a CRM or finance application directly, but agents still need its data, permissions, audit trail and transaction controls.
Raw usage pricing solves the vendor’s margin problem but creates a budgeting problem for the customer. Credits, committed allowances, spending caps and alerts are becoming essential because open-ended agent bills are difficult for procurement teams to approve.
Outcome-based pricing will spread, but only where success is frequent and easy to verify. Support resolutions and processed documents fit that model; creative work, strategic advice and final sales revenue usually do not.
Total software spending can still rise while seat growth slows. Agents create more demand for models, APIs, data, security and workflow infrastructure, but they also make thin applications and replaceable user interfaces easier to bypass.
The practical conclusion is that seats survive as the price of identity, access and accountability. They simply stop being the universal meter for software value once machines begin doing a meaningful share of the work.
This market map, featured in our agentic AI market deck, highlights top companies and startups in the agentic AI market
Will AI agents kill per-seat SaaS pricing?
Why is per-seat SaaS pricing under pressure now?
Per-seat SaaS pricing is under real pressure now because software can produce far more work without adding another human user.
For most of the SaaS era, the number of employees was a reasonable proxy for software value. A company hired 20 salespeople, then bought 20 more CRM licenses. A support team expanded, then added helpdesk seats. The customer’s payroll and the vendor’s revenue usually moved in the same direction.
AI agents weaken that link. A support agent can answer thousands of additional questions without receiving thousands of user accounts. A coding agent can work for hours while one developer supervises several tasks. A sales agent can research every account overnight without appearing anywhere in the customer’s headcount.
Gartner currently estimates that up to $234 billion of enterprise application spending could be exposed to this “agentic arbitrage” by 2030, equivalent to roughly 20% of enterprise application SaaS spending. Gartner is describing software work moving through agents while employees interact less with the original applications.
Recent operating data shows why vendors are taking the risk seriously. Salesforce reported more than $1 billion in Agentforce annual recurring revenue and 3.8 billion completed Agentic Work Units in its latest quarter. OpenAI also reported more than 5 million weekly Codex users, more than six times the number recorded after its desktop launch, with knowledge workers becoming its fastest-growing group.
These figures remain small beside the full SaaS market, but they expose the mismatch. Machine activity can rise quickly while the number of paid employees barely moves.
What would it actually mean for AI agents to kill per-seat pricing?
AI agents would have killed per-seat pricing only when human headcount stops anchoring most of a software contract’s value.
A complete disappearance of seats looks unlikely. Companies will continue paying for employee identities, permissions, collaboration histories and access to sensitive data. Many products genuinely need to know who is using them.
A more realistic change is already underway: the seat remains on the invoice, but it no longer captures most future growth. Customers pay a base fee for human access, then spend more when agents complete actions, run workflows or deliver outcomes.
Microsoft can keep selling Microsoft 365 Copilot by the user while charging separately for autonomous Copilot Studio activity. Zendesk can keep charging for support-team access while billing automated resolutions. Notion can preserve its workspace subscription while making Custom Agents consume credits.
The important threshold is the loss of seat-only pricing. When a vendor can no longer include unlimited agent activity in a fixed user fee without undercharging heavy customers or damaging its margins, the old model has already lost its central role.
We will therefore treat per-seat pricing as “killed” only where the number of employees becomes a minor part of what determines the bill. Keeping a small platform or access fee would not change that economic reality.
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
Are AI agents already shrinking SaaS seat counts?
AI agents are currently changing software bills much faster than they are shrinking payrolls.
McKinsey’s latest global survey found that 23% of respondents were scaling an agentic system somewhere in their organization, while another 39% were experimenting. Yet no individual business function had more than 10% of respondents scaling agents. Most companies are still testing narrow workflows rather than rebuilding entire departments.
The latest labor data also gives us little evidence of a broad seat collapse. PwC’s 2026 analysis found that companies in the most AI-exposed sectors had grown headcount by about 52% since its baseline, compared with roughly 36% among the least exposed companies.
PwC notes that the analysis has survivorship and reporting biases, so those percentages should not be read as a clean causal estimate. The relative gap still challenges the idea that AI exposure is already producing widespread employment contraction.
Companies are often using agents to handle more work, enter new markets or improve service before they reduce staff. A support team may absorb twice the ticket volume with the same headcount. A software company may ship more products without hiring at its previous rate. Both cases weaken future seat growth, even though the existing seats remain.
Reliability is another brake. Gartner recently forecast that 40% of enterprises could demote or decommission autonomous agents by 2027 because their governance systems fail to match each agent’s risk and autonomy. Human review, approvals and exception handling will remain common while companies solve those problems.
A sudden collapse in paid SaaS users looks premature. Slower seat expansion is much more plausible, especially in departments where agents let a fixed team handle steadily rising workloads.
Does one AI agent really replace one SaaS seat?
One AI agent rarely maps to one lost SaaS seat because agents replace task volume before they replace an entire job.
A customer-service agent may answer 60% of routine questions while human representatives handle unusual cases, complaints and sensitive conversations. The support team may become smaller later, but every remaining representative still needs access to the helpdesk, customer history and escalation tools.
The relationship can also run in the opposite direction. One employee may supervise a research agent, a coding agent and a reporting agent at the same time. OpenAI’s recent Codex research found that more than 10% of active users managed at least three concurrent agents during a typical week. The number of humans stays fixed while machine activity multiplies.
Machine identities complicate the calculation further. An agent might access a CRM, a document system and a communication platform during one workflow. Each vendor must decide whether that agent needs a license, a service account, a pool of credits or permission through an integration.
A SaaS seat disappears only after the agent automates enough work to change staffing and employees can bypass the application without losing access to its data. Pricing usually changes earlier because the agent is already consuming resources and producing value.
This chart, included in our agentic AI market deck, illustrates yearly VC funding for agentic AI startups
Which SaaS categories face the most pressure from AI agents?
Customer support, sales development and routine back-office workflows face the sharpest pricing pressure because their outputs can be counted without counting employees.
Customer support has moved first. Intercom charges $0.99 when Fin delivers a defined outcome. Zendesk meters AI activity through automated resolutions completed without human intervention. HubSpot charges credits when its Customer Agent resolves a conversation or its Prospecting Agent recommends a lead.
These products have frequent and visible completion events. A conversation was resolved, a lead was qualified or an invoice was processed. The customer can compare the agent’s price with the cost of having a person perform similar work.
Coding tools sit in the middle. Agents can write features, fix bugs and prepare pull requests, but developers still review code, discuss architecture and remain accountable for production systems. GitHub therefore combines user licenses with AI credits rather than charging only for completed code.
Collaboration, identity and systems of record face less immediate seat erosion. Employees still need personal email, messaging accounts, security permissions and access to authoritative customer or financial records. AI can raise the value of those products without removing the employee’s identity from the workflow.
| SaaS category | Why the category is exposed | Current pressure on seats | Likely pricing direction |
|---|---|---|---|
| Customer support | Resolutions and escalations are easy to count | High | Platform access plus automated outcomes |
| Sales prospecting | Leads, research and meetings can be measured | High | Seats plus qualified leads or agent credits |
| Back-office processing | Documents, invoices and requests follow repeatable steps | High | Charges per task, workflow or document |
| Software development | Agents complete tasks, but developers still review and own the result | Medium | Developer licenses plus AI consumption |
| Collaboration tools | Every employee still needs an identity and shared history | Low to medium | Higher-value seats with included AI |
| CRM, ERP and HR systems | Agents depend on their governed records and permissions | Low to medium | Platform access plus agent actions |
| Identity and security | Named users and accountability remain essential | Low | Seats, protected identities and consumption |
If you want more recent data on this point, please see our latest agentic AI market report.
Why can AI agents break seat economics before jobs disappear?
Agent economics can damage seat pricing even with stable headcount because every extra task creates cost and value that the seat count misses.
Imagine a support company with 100 representatives. Its helpdesk vendor receives roughly the same seat revenue whether the team handles 20,000 or 200,000 requests. That arrangement works when employees perform most of the work and additional software usage costs the vendor very little.
An AI agent changes both sides of the equation. The vendor pays for model inference, data retrieval, tool calls and monitoring each time the agent works. Meanwhile, the customer receives more output without purchasing another human license.
Heavy agent adoption can therefore increase the vendor’s costs tenfold while its seat revenue stays flat. A powerful agent can also produce several employees’ worth of output through one administrator account.
Pricing teams are already describing this as a financial problem rather than a distant theory. In interviews conducted by Metronome, SaaS leaders said they could tolerate unused human-seat capacity because the marginal cost was limited. Unpriced AI usage was different because the vendor paid real infrastructure costs every time the customer used it.
Cheaper models may reduce the cost of each task, but customers often consume more AI as prices fall. Notion recently cut the credit cost of many Custom Agent tasks by 35% to 50% while continuing to meter every run.
This chart, included in our agentic AI market deck, shows how Cognition is positioned in agentic AI
Are major SaaS vendors actually abandoning per-seat pricing?
Major SaaS vendors are keeping human seats and adding a separate meter for machine work.
We reviewed the current public pricing of eight prominent SaaS platforms offering agents or agent-like features. Seven now separate at least some autonomous work from the standard human license. Slack is the clearest exception in this sample because most of its built-in AI remains packaged inside user tiers.
This is a small, deliberately selected group rather than a representative survey of the entire SaaS market. The consistency is still striking. CRM, coding, support, productivity and collaboration vendors have reached a similar answer despite using different terminology.
Salesforce calls the unit an action. Microsoft uses Copilot Credits. GitHub converts model tokens into AI credits. Intercom charges for outcomes, Zendesk for automated resolutions, HubSpot for resolutions and recommended leads, and Notion for credits consumed by each task.
The vendors have not agreed on a common meter, but they have largely agreed that machine work needs one.
| Vendor | Human-access charge | Agent charge | What currently determines agent spend |
|---|---|---|---|
| Salesforce | User and product licenses remain available | Flex Credits or conversations | Actions completed by Agentforce |
| Microsoft | Microsoft 365 Copilot costs $30 per user monthly | Credit packs or pay-as-you-go | Responses, actions and other operations |
| GitHub | Copilot Business and Enterprise require licenses | Pooled AI credits and overages | Model choice and tokens consumed |
| Intercom | Optional helpdesk seats | $0.99 per standard Fin outcome | Resolutions, procedure handoffs and other outcomes |
| HubSpot | Professional or Enterprise subscriptions and seats | HubSpot Credits | Resolutions, recommended leads and agent actions |
| Zendesk | Support and Suite plans remain seat-based | Automated resolution allocation | Issues resolved without human intervention |
| Notion | Business or Enterprise workspace seats | $10 per 1,000 shared credits | Complexity and frequency of Custom Agent tasks |
| Slack | Per-user workspace plans | Most built-in AI currently bundled | Plan tier rather than a separate agent meter |
Is usage-based pricing the obvious replacement for per-seat SaaS?
Usage-based pricing will capture a large share of agent revenue, but raw consumption is too awkward to replace seats on its own.
Metronome’s survey of 100 SaaS companies found that 85% had adopted some form of usage-based pricing. It also found usage pricing at 77% of the largest software companies it examined. Nearly half of the companies using it had introduced the model during the previous two years.
AI strengthens the case because vendor costs genuinely rise with activity. Charging for agent calls, completed steps or model usage prevents a small number of intensive customers from consuming unlimited resources inside a flat subscription.
The difficult part is choosing a unit customers understand. Tokens reflect the vendor’s model bill but say little about business value. Agent steps can punish customers when a poorly designed system takes too many actions. Compute time rewards speed rather than usefulness. A “task” might mean answering one question or spending an hour analyzing several databases.
GitHub’s current model reveals the tension. Customers still buy licenses for each user, while model usage is converted into AI credits according to the tokens consumed. The arrangement protects GitHub from expensive activity, but the customer must understand allowances, model rates and possible overages.
Pure usage pricing can also suppress adoption when employees feel that every interaction adds another charge. In practice, usage is becoming the expansion meter while a subscription or committed allowance keeps the contract predictable.
This chart, included in our agentic AI market deck, illustrates yearly funding for agentic AI startups
Can customers live with unpredictable AI-agent bills?
Enterprise customers will accept variable AI-agent bills only when vendors put firm limits around the variability.
Traditional SaaS contracts are easy to forecast. Procurement counts the expected employees, negotiates a price and knows roughly what the year will cost. Agent usage can jump because employees adopt the tool, customer demand rises or a workflow starts running more frequently.
Microsoft now offers three ways to control that uncertainty in Copilot Studio. Customers can buy packs containing 25,000 Copilot Credits for $200 per month, pre-purchase larger commitments or use a pay-as-you-go meter.
Notion uses a shared workspace balance for Custom Agents and sends alerts when consumption reaches 80% and 100%. HubSpot lets administrators purchase credit packs, enable pay-as-you-go spending and set limits. GitHub pools the AI credits included with employee licenses so that light users leave capacity for heavier users.
Customers will tolerate a moving bill when they can set caps, receive warnings and see which agent consumed the budget. Completely open-ended metering will remain difficult to sell outside technical infrastructure.
Will outcome-based pricing win for AI agents?
Outcome-based pricing will win in customer support and a few other measurable workflows, while most agent products will stop at usage or credits.
Customer support offers unusually clean events. Intercom charges $0.99 when Fin resolves a conversation, completes a configured handoff or delivers another defined outcome. Zendesk charges for automated resolutions completed without human intervention and uses a model to verify whether the conversation qualifies.
Outcome pricing is now moving into sales, although the price and definition change. Intercom charges $9.99 when Fin for Sales qualifies a prospect according to criteria chosen by the customer. HubSpot uses 100 credits when its Prospecting Agent recommends a lead.
These examples show how quickly the idea becomes complicated. A support resolution can usually be observed within one conversation. A sale may occur months later after advertising, product demonstrations, negotiations and human follow-up. Charging for a qualified lead is easier than proving that one agent caused the final contract.
Orb’s analysis of AI-agent products found outcome pricing in only 4.5% of the sample. Two-thirds of those companies also used another pricing method. That low adoption rate reflects the work required to define success, verify it, handle disputes and decide who deserves credit when several systems contribute.
Outcome pricing will spread fastest where results are frequent, visible and mostly controlled by the agent. Support resolutions, processed documents and completed compliance checks fit that description. Strategic advice, creative work and revenue generation rarely do.
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, compares the main business model options for autonomous AI agent platforms
Why are hybrid AI pricing models becoming the default?
Hybrid AI pricing is becoming the default because vendors want predictable commitments without leaving agent activity unpriced.
Stripe recently found that 86% of SaaS platforms offering AI features were already charging for them. Yet 44% expected to change their AI pricing more than once during the following year. Vendors know AI must generate revenue, but many remain unsure whether customers should pay for tokens, credits, tasks, outcomes or a premium tier.
A separate Stripe survey of AI-company leaders found that 56% used hybrid pricing and 38% relied on pure usage pricing. AI-native businesses can begin with consumption because they have no old contract structure to protect.
Established SaaS vendors are moving more cautiously. Bain reviewed more than 30 incumbents and found that around 65% had layered an AI meter or access charge onto seat pricing. The remaining 35% mainly raised seat prices or bundled AI into more expensive tiers. None had converted its entire business to a standalone AI charge.
A base subscription keeps budgets and revenue predictable. Seats continue to control human access, while credits, tasks or outcomes capture autonomous work. The structure also lets vendors change the balance gradually as customers learn how much agent activity they need.
| Research sample | Companies studied | Current finding | What we can reasonably conclude |
|---|---|---|---|
| Bain review of established SaaS vendors | More than 30 | About 65% added an AI meter to seat pricing | Incumbents are extending seats rather than rapidly replacing them |
| Orb analysis of AI-agent products | 66 | Around 92% used more than one pricing component | AI-agent vendors strongly prefer hybrid structures |
| Stripe survey of AI-company leaders | Not publicly specified in the summary | 56% used hybrid pricing and 38% used pure usage pricing | Consumption dominates, but subscriptions remain common |
| Metronome SaaS survey | 100 | 85% had some form of usage-based pricing | Usage has already entered mainstream software pricing |
| Stripe vertical SaaS research | SaaS platforms in its survey | 44% expected several AI pricing changes within a year | The winning hybrid formula remains unsettled |
Will software budgets shift from SaaS seats to digital labor?
Digital labor is starting to become a software budget line, although most companies still buy agents through existing SaaS contracts.
Salesforce has made the shift unusually explicit. Its Flex Agreement allows customers to move some spending between traditional user licenses and Agentforce consumption. A company can therefore rebalance its contract as employees and agents take different shares of the work.
The comparison becomes easiest in workflows with a known labor cost. A support leader already knows the approximate cost of handling a request with an employee or outsourced provider. Paying $0.99 for an automated outcome can be evaluated against that baseline.
The calculation becomes less precise for research, management and creative work. An agent may save an analyst four hours without reducing payroll or creating directly measurable revenue. The company receives value, but the spending still looks like productivity software rather than a replacement for labor.
Software, cloud infrastructure, employees and outsourced services are also approved by different teams. Digital-labor spending will usually enter through an existing software budget first, then expand when the vendor can prove savings against a human or outsourced workflow.
This chart, featured in our agentic AI market deck, shows the share of revenue generated by each customer segment in the agentic AI market
Can AI agents expand software demand while weakening seat growth?
AI agents can increase total software demand even as they weaken seat growth because machines can call many systems far more often than employees do.
One employee has a limited working day. Several agents can research, write, retrieve records and update systems in parallel. That activity creates demand for models, databases, APIs, security controls, monitoring and the applications that hold the original business data.
A recent RBC Capital Markets survey of more than 100 technology leaders found that every respondent had allocated money to AI, while 91% were creating new AI budgets rather than only moving money from other software. No respondent reported cutting broader software spending because of AI. One survey cannot settle the market’s direction, but it argues against an immediate “SaaSpocalypse.”
Growth will be uneven. Agent orchestration, data infrastructure and security products can benefit from higher machine activity. A CRM that stores authoritative customer records may process more queries even when fewer employees visit its dashboards.
Thin applications face the opposite result. An agent can reproduce the convenience of opening a small tool, copying information and moving it into another system. Software spending can therefore rise overall while shifting away from products that mainly provide a replaceable interface.
Will AI agents let companies bypass entire SaaS applications?
AI agents will hide many SaaS interfaces from employees, but the underlying systems of record will remain hard to remove.
An employee may ask an agent for the latest sales pipeline rather than opening Salesforce, filtering opportunities and building a dashboard. The employee interacts with the agent, while Salesforce still stores the accounts, permissions, activity history and forecast data.
The same pattern can appear in finance and HR. An agent can submit an expense, answer a benefits question or prepare a purchasing request. It still needs an approved ledger, employee record or procurement system behind the conversation.
Gartner’s $234 billion estimate focuses heavily on this interface problem. As agents work across several applications, customers may question why every employee needs full access to each one. Vendors that relied on frequent screen time as proof of value will face difficult renewal conversations.
Removing the underlying platform is much harder because agents need reliable data, stable APIs, access rules, audit logs and a clear place where the final transaction is recorded. SaaS companies that own those layers can charge agents for access and actions even when employees stop visiting the interface. Companies that own only a convenient screen have fewer defenses.
This chart, included in our agentic AI market deck, shows how autonomous AI agent platform technology has evolved over time
Which SaaS companies will win and lose the AI pricing reset?
The AI pricing reset favors vendors that own trusted data, measurable workflows or a required control point in the process.
Systems of record start from a strong position. Agents need accurate customer, employee, financial and product information. A vendor that controls those records can charge for platform access, API activity, agent actions and governance even when employees visit the interface less often.
Workflow platforms also have an advantage when they can verify completion. Support vendors can meter resolutions. Document platforms can count processed files. Security companies can charge for protected assets, investigations or events.
Human collaboration products can preserve seats by making AI part of each employee’s workspace. Slack still charges per user while including several AI features in its plans. Microsoft includes internal agent-building capabilities with its Copilot user license. These products continue to benefit from identity, communication and shared organizational context.
The weakest position belongs to a narrow application that mainly helps a person move information between other systems. An agent can often perform that work directly through APIs while giving the user one conversational interface.
| SaaS position | Effect of AI agents | Pricing outlook |
|---|---|---|
| System of record with trusted proprietary data | Agents depend on its records, permissions and audit trail | Strong |
| Platform controlling a measurable workflow | The vendor can charge for completed work | Strong |
| Security, identity or governance layer | More agents create more identities, permissions and risks to manage | Strong |
| Personal productivity or collaboration workspace | Human seats remain useful while AI raises value per user | Medium to strong |
| Specialist tool with unique data or expertise | The interface may fade, but the underlying capability remains valuable | Medium |
| Generic point solution connecting other applications | Agents can reproduce or bypass much of the workflow | Weak |
| Seat-only product offering unlimited expensive AI | Usage can grow while revenue stays fixed | Weak |
If you want more recent data on this point, please see our latest agentic AI market report.
Will SaaS vendors eventually earn more from agents than human seats?
A minority of SaaS vendors will eventually earn more from agent work than human seats, especially in high-volume support, security and transaction workflows.
Consider a support platform charging a hypothetical $100 per month for each of 1,000 human representatives. That produces $100,000 in monthly seat revenue.
If the same platform handles 500,000 automated outcomes at $0.99 each, the agent component generates $495,000. Human seats remain important, but machine work produces almost five times as much revenue.
That comparison is most realistic where agents perform millions of similar tasks, such as support conversations, fraud checks, security events, invoices and insurance claims.
Collaboration, design and executive productivity software will look different because employees still need personal workspaces and the agent’s value is spread across tasks that are difficult to measure.
Competition will also push down the price of common agent actions. Proprietary data, reliable execution, regulatory trust and deep workflow integration will decide which vendors can charge far above the underlying compute cost.
In our agentic AI market deck, we identify pain points entrepreneurs should prioritize
Will AI agents kill per-seat SaaS pricing?
Our final judgment is that per-seat pricing survives, while seat-only pricing loses its default status wherever agents perform work autonomously.
The strongest version of the claim is unsupported today. AI agents have not eliminated enough jobs, application users or software contracts to cause a broad collapse in seats. Most companies are still experimenting, and scaled deployments remain concentrated in a small number of functions.
The economic change is already real. Agents generate costs and value without increasing human headcount. Seven of the eight major platforms we reviewed now meter at least some machine work separately from employee access. Surveys from Bain, Stripe, Orb and Metronome point toward the same hybrid structure.
Customer support, sales development and repetitive processing will move furthest away from seats. Their outputs are frequent and measurable, making tasks or outcomes better billing units than employees.
Collaboration, productivity, security and systems of record will keep a strong seat component. Those products still need to identify people, preserve responsibility and control access to company data. Agents may raise their revenue per employee rather than reduce it.
The dominant SaaS contract will increasingly contain three pieces: a platform commitment, human licenses and a pool of agent activity. The balance will vary by product. In operational software, the agent meter may eventually become the largest component.
So AI agents are unlikely to erase per-seat SaaS pricing. They are ending the period when counting employees was enough to price almost every kind of business software.
If you want more recent data on this point, please see our latest agentic AI market report.
OUR METHODOLOGY
This analysis tests whether AI agents are genuinely displacing per-seat SaaS pricing. We examine the issue through the factors that determine whether a pricing model is actually losing relevance: enterprise adoption, software usage, employment and seat growth, agent operating costs, current vendor billing models, differences between software categories, budget predictability and the ability to measure completed outcomes.
We separated three developments that are often mixed together: agents performing more work, companies reducing employee headcount and vendors changing how they charge. Rising agent activity does not prove that jobs or software seats have already collapsed. It can still undermine seat-only economics when machine usage creates additional value and infrastructure costs without creating another paid user.
We prioritized observable behavior over broad predictions. Current pricing documentation shows what vendors meter, include, limit and charge for today. Company disclosures and adoption surveys show whether agent activity is reaching meaningful scale. Labor evidence helps test whether that activity has already translated into widespread employment or seat contraction.
The review of eight major SaaS vendors is deliberately selected rather than statistically representative of the whole market. It covers prominent platforms across CRM, productivity, coding, customer support and collaboration so that we can compare how different categories are responding to the same pricing problem.
Human licenses, agent credits, actions, automated resolutions, outcomes and other billing units were treated as separate pricing components even when they appeared inside the same contract. A vendor was considered to be moving beyond seat-only pricing when autonomous activity could create an additional charge, consume a limited allowance or require a separate commitment.
We also distinguished raw usage pricing from outcome pricing. Tokens, credits and actions mainly measure activity or vendor cost. Resolutions and qualified leads attempt to measure customer value. We treated outcome pricing more cautiously because it requires a clear definition of success, reliable verification and an agreed method for assigning credit.
Evidence from systems of record, collaboration products and operational workflows was assessed separately. A product that controls authoritative data, permissions or audit trails can preserve strong access pricing even when employees stop using its interface directly. Products built around countable operational tasks face greater pressure to connect their prices with machine output.
Key sources include Gartner’s analysis of enterprise application spending exposed to agentic arbitrage, Salesforce’s Q1 FY2027 results, Salesforce’s definition of Agentic Work Units, McKinsey’s global AI adoption survey, PwC’s 2026 Global AI Jobs Barometer, Microsoft’s Copilot Credits documentation, GitHub’s usage-based Copilot billing documentation, Intercom’s Fin outcome definitions, Zendesk’s automated-resolution methodology, Metronome’s survey of usage-based SaaS pricing, and Stripe’s research on AI pricing models.
The final judgment comes from convergence across these sources rather than from one company or statistic. Vendor pricing pages show the move away from unlimited machine work, company results show rising agent activity, surveys show that deployment remains early and labor data shows that a broad seat collapse has not yet occurred. Together, those findings support a hybrid future in which human access remains important but no longer captures the full economic value of the software.
This chart, included in our agentic AI market deck, shows the share of revenue by region across Europe, Asia, North America, Africa, and South America in the agentic AI market
