The 2026 AI Agent Market: B2B Pricing and B2C Payments as Key Variables

The AI agent market is flooded with news every day, but the pace of development differs significantly between enterprise and consumer sectors. As of October 2026, the B2B sector is seeing its software pricing models shaken up, while in B2C, although search and recommendation systems have evolved, final payments still remain firmly in human hands.

This article is based solely on publicly available research and forecasts. A disclaimer is necessary: many of the figures below are taken from articles and blogs citing primary sources like Gartner, Deloitte, Bain, and Morgan Stanley, rather than the original reports. Therefore, I have noted the source for each figure in every sentence. Research data is labeled with its sample and criteria, and forecasts are clearly identified as such.

Why Figures on the AI Agent Market Vary So Much

First, let’s define our terms. Here, an AI agent refers to an AI that, when given a goal by a human, breaks it down into steps, utilizes tools and data, and drives the task to completion. Unlike chatbots that only provide answers, they are distinguished by their ability to execute actions.

When looking for the size of this market, the numbers vary wildly. According to a roundup by the [Software Strategies Blog on February 26, 2026], Gartner’s 2026 forecast of $201.9 billion in agentic AI spending is a broad definition that includes embedded features. Looking only at the independent agentic AI market, the estimate is around $7–8.5 billion. Despite sharing the same market name, there is a 24- to 29-fold difference, calculated by dividing the two figures.

Other estimates exist as well. [Market research firm MarketsandMarkets] estimates that the agentic AI market will grow from $7.06 billion in 2025 to $93.20 billion in 2032, a CAGR of 44.6%. Their definition and criteria differ from Gartner’s.

Therefore, this article focuses more on research showing what has actually changed and where, rather than market size figures. While size depends on definitions, companies adjusting price tags or consumers changing their purchasing paths are observable regardless of definitions. This is why this article compares B2B and B2C within the same framework, following the same sequence: what is changing, who is affected, and what remains unchanged.

B2B AI Agents: How Deeply Are They Adopted in Enterprises?

The first question in B2B is whether they are actually being used. The [Liferay 2026 Agentic AI Adoption and Governance Report], conducted by Pollfish, surveyed 500 US professionals involved in AI decision-making, implementation, and operations. 54% stated that their company is running AI agents in production or as a pilot. This was published on CMSWire on August 12, 2026, though the exact survey date was not specified. It should be noted that this survey was released by Liferay, a digital experience platform company.

One must look at the composition of the 54% figure. It groups production operations and pilot projects together, so it does not mean company-wide adoption is complete. In the same survey, only 25% of companies measured the impact of AI with clear KPIs. The gap—where over half of companies use them, but only a quarter measure the effectiveness numerically—best illustrates the current position of the B2B market.

Who is affected splits in two directions.

One is the SaaS vendor. The logic of charging per seat based on the number of people is wavering as agents do the work instead. The other is the IT and finance department of the adopting company. When invoices arrive based on usage rather than headcount, the budgeting method must change. This pricing discussion continues in the next section.

What has not changed yet is measurement and accountability. In a situation where only 25% of companies measure effectiveness with KPIs, it is difficult to guarantee how quickly performance-based pricing—assigning value to the results produced by agents—will spread.

SaaS Pricing Model Shifts: From Per-Seat to Usage and Outcome

Per-seat pricing is built on the premise that humans perform the work.

It’s a set amount per account; as staff grows, revenue increases. However, if an agent handles customer inquiries, the number of agent accounts may decrease, and vendor revenue could drop accordingly. This is why price tags must be tied to the volume or results of work rather than the number of people.

Figures illustrating this trend are mostly cited from [SoftwareSeni’s article](published March 23, 2026, updated September 23). I could not verify the original reports, and the sample and timing of each survey are not provided in the cited article.

• The figure cited as a Gartner forecast is that by 2030, over 40% of enterprise SaaS spending will shift to usage, agent, and outcome-based pricing.
• The figure cited from Deloitte TMT Predictions 2026 is a forecast that the revenue share of per-seat vendors will drop from 21% to 15%. The citation does not specify the period compared.
• The figure cited from a Bain study is that about 65% of over 30 major SaaS vendors have added AI usage-based pricing on top of per-seat fees.
• The figure cited from a Maxio study is that 83% of AI-native SaaS companies have shifted to usage-based pricing.

The emphasis of the 65% figure is on addition, not replacement. A hybrid model, keeping seat fees and adding usage items on top, appears to be the most common form today. The 83% is a figure from a narrow group of AI-native companies, so the population differs from the 65%; these two numbers should not be combined as a single average.

Zendesk is a case where pricing is visible. According to the cited article, Zendesk introduced performance-based AI agent pricing in August 2024. It is $1.50 per resolution for contracted customers and $2.00 for pay-as-you-go. I could not verify the official price list. Pay-as-you-go is about 33% more expensive than the contract, and assuming 1,000 cases are resolved automatically in a month, it would be $1,500 for the contract and $2,000 for pay-as-you-go. This 1,000 cases is a hypothetical for calculation. It is a structure where the number of resolved cases becomes the billing unit regardless of how many agent accounts exist.

There is also movement in the opposite direction. According to the same article, Salesforce has introduced the Agentic Enterprise License Agreement, a 2–3 year contract allowing unlimited use of Agentforce, Data 360, and MuleSoft for a fixed fee. It is also said that Slack, Google Workspace, and Salesforce have moved customers to AI-inclusive packages through forced SKU conversions. It can be interpreted that as usage-based pricing increases, it becomes harder to predict bills, leading to a parallel trend where purchasing departments prefer fixed fees.

B2C AI Agents and Agentic Commerce: AI Starts, Humans Pay

Agentic commerce refers to transactions where an AI agent handles product search, comparison, and payment on behalf of the consumer.

Data on the consumer side was verified through an [Exploding Topics survey introduced by techwyse](April 27, 2026). It is a proprietary survey by a Semrush-affiliated outlet, with 1,009 US consumers reflecting on the past 6 months. It has limitations as it relies on memory and only includes US respondents.

Many people use it. 77.6% of respondents said they used AI for shopping or purchasing decisions in the past 6 months, and 43.21% said they use AI for shopping at least once a week.

The problem is the purchasing path.

Only 2.18% responded that they finished the entire purchasing process within an AI tool. 44.8% said they started on a retail site and supplemented with AI, while 44.03% started with AI and paid on an external site. Combined, these three paths account for about 91%, and the remaining paths are not mentioned in the article.

How to read this is simple. AI has taken over the starting point and comparison stages, but payment is still handled by existing sites and human hands. This is the part that AI agents have actually changed in B2C today.

The impact is visible in sales first. 68.64% responded that they bought items they had no intention of buying originally because of AI. The response criteria followed the original expression. This can be read as spending being swayed at the stage of suggesting new reasons for purchase, even before the agent buys on behalf of the consumer.

The wall of trust is also shown in figures.

31.21% of respondents completely disallowed autonomous AI agent spending, and the most selected limit was $0. A $20 limit was 17.45%, a $50 limit was 20.74%, and only 11.71% allowed over $100 per transaction. Even among those who shop with AI more than once a week, 51.84% set a limit of $50 or less. Even frequent users do not entrust large sums of money.

The forecast is much larger. The figure known as the Morgan Stanley Research forecast is that by 2030, agentic shopping e-commerce spending will be $190–385 billion, 10–20% of the total, with 126 million shopping agent users. I confirmed this in a table in a summary document, not the original report, and could not verify the publication date. The denominator for the total range was not written in the summary. While direct comparison is impossible due to different criteria, placing the 2030 forecast next to the current 2.18% gives an idea of how large the gap to be bridged in the payment stage is.

B2B vs B2C AI Agent Comparison: A Side-by-Side Framework

Summarizing the figures seen so far with the same questions, we get the following table. All figures in the cells were discussed with their sources in the previous sections.

Table. B2B vs B2C AI Agent Market Comparison (As of October 2026, based on research/forecasts cited in this article)

Category	B2B AI Agent	B2C AI Agent
What is changing	Software pricing: From per-seat to usage/outcome	Starting point of purchase: Search/comparison moves to AI
Who is affected	SaaS vendors, enterprise IT/finance	Retail sites, shopping consumers
Representative figures	54% operating/piloting, ~65% of vendors added AI usage items	77.6% shop with AI, 2.18% complete purchase within AI
Unit of money movement	Contracts/agreements (e.g., 2-3 year fixed contracts), resolution counts	Individual purchases, consumer-set spending limits
What hasn’t changed	Effectiveness measurement (25% of companies measure with KPIs)	Payment (31.21% completely disallow autonomous spending)
2030 forecast	Over 40% of SaaS spending shifts to new pricing (cited from Gartner)	$190–385 billion in agentic shopping spending (Morgan Stanley summary)
Sample basis	Survey of 500 US professionals + secondary citations	Survey of 1,009 US consumers

The most notable part of the table is that the units in which money moves are different.

In B2B, contracts and price tags move. When a vendor modifies billing items, the billing method for thousands of locations changes at once. In B2C, spending limits set by each individual move. Since the paths through which changes spread are so different for the same AI agent, the speed is bound to be different as well.

Common bottlenecks are also visible in both markets.

In B2B, 25% of companies measured effectiveness, and in B2C, 31.21% of consumers completely disallowed autonomous spending.

Even if the appearance is different, the structure is the same: it stalls when the party entrusting the task lacks the means to verify results or a basis for trust. This part is an interpretation of reading the two surveys side by side; the surveys themselves did not directly demonstrate causality.

The limitations of the samples are also clear. The figures in the table are all centered on US respondents, and the entities conducting the surveys are often vendors or vendor-affiliated media. They are useful for reading trends, but it is difficult to apply the individual numbers directly to the Korean market.

MCP·A2A·AP2: Connection and Payment Standards Connecting the AI Agent Market

For agents to work, they must be connected to tools, other agents, and payment methods.

According to the explanation on the official AP2 site, roles are divided.

MCP connects agents with tools, APIs, and data.

A2A allows agents to exchange messages and tasks with each other.

AP2, an extension of A2A, handles payment-specific communication such as payment mandates.

The first two are the base layers, and AP2 is the payment layer on top of them.

The MCP server for AP2 is in development, and the schedule has not been released.

There is a figure with a reference date for A2A.

According to a [Linux Foundation press release on April 9, 2026](the 1st anniversary of A2A’s launch), the number of organizations supporting A2A increased from about 50 in April 2025 to over 150 in April 2026. Supporters include AWS, Cisco, Google, IBM, Microsoft, Salesforce, SAP, and ServiceNow. The core repository has over 22,000 GitHub stars, and SDK support has expanded from just Python to production support for 5 languages. In payments, over 60 entities in the payment and finance sectors have declared support for AP2.

There is a point to be careful about when reading these numbers. The number of supporting organizations is not the number of adoptions. The press release announced integration with major cloud platforms and production use cases, but it did not disclose how many organizations are using it in production. It is safer to read the 150 as the scale of declared support.

These standards touch the two markets in different ways.

In B2B, it becomes a channel for agents in business software like SAP, Salesforce, and ServiceNow to call each other.

In B2C, AP2 directly addresses the problem seen earlier: that humans still do the payment. As the name ‘mandate’ suggests, it is a device to set how much payment can be entrusted to an agent, so it can be read as a technical attempt to solve the same problem as consumers who chose a $0 limit. However, no adoption figures for AP2 have been released.

Korean AI Agent Market: 58% Corporate Adoption, 8% Full Agentic Adoption

Domestic data was taken from the Strand Partners report, “Realizing Korea’s AI Potential in 2026.”

It is a survey commissioned by AWS, with 1,000 domestic business leaders and 1,000 members of the general public responding. It is based on an MTN News report from September 14, 2026 and an article on skax.co.kr from October 1, 2026.

Korea’s corporate AI adoption rate is 58%, up from 48% the previous year.

Looking at it by stage, the change is clearer.

The basic stage focused on efficiency decreased from 70% to 54%, and the advanced stage using customized AI systems or agentic AI more than doubled from 11% to 26%.

Looking at agentic AI alone, 37% of companies have heard of it, 29% are in the experimental/pilot stage, and 8% have fully adopted it. Details of the respondent population are not written in the article. Since it is unknown whether the three numbers overlap or were counted separately, they were not summed. It is impossible to rank it against the 54% from the US Liferay survey because the samples and questions are different.

As for the effects reported by adopting companies, an 81% productivity improvement, 22% average revenue increase, and 14 hours of average weekly savings per company were introduced. This is a report commissioned by AWS, and the population details require verification of the original. It is more appropriate to view this figure as data close to the upper limit of the adoption effect.

By sector, the ratio of the advanced stage was reported as 41% for ICT, 35% for manufacturing, 35% for startups, 25% for SMEs, and 9% for large enterprises. The 9% for large enterprises is as written in the report, but it is a figure that requires contrast with other items. Since the result that large enterprises are the lowest contradicts intuition, I will not use it as a basis until verified by the original text. 24% responded that they are fully or very prepared for next-generation AI, and 43% of startups said the same.

The materials verified this time lack domestic agent service cases and policy details. Therefore, that part was not covered.

Risks and 1-2 Year Outlook for the AI Agent Market

The most frequently cited warning is Gartner’s prediction from June 25, 2025.

It states that over 40% of agentic AI projects will be canceled by the end of 2027. This was verified not from the original Gartner text but from a secondary page citing it; an article from February 18, 2026, also cited the same figure. It is a prediction, not an aggregated result.

While there is no material directly explaining the reasons for cancellation, there are domestic figures showing the situation.

In the aforementioned Strand Partners/AWS report, 47% of domestic companies answered they have no reliable way to measure AI investment returns. 31% of companies decide to expand or stop projects by looking at AI metrics, and 26% track business performance metrics. As obstacles to adoption, 46% cited a lack of AI/digital capabilities, 38% cited data-related difficulties, and 33% cited a lack of internal talent and business transformation capabilities. If performance cannot be measured, the basis for expanding or stopping is weak. This is a figure in the same direction as the 25% KPI in the Liferay survey. However, there is no basis to say these figures have a causal relationship with the cancellation rate.

Governance figures are also low.

67% of domestic companies view AI as a top priority or important strategic task, but only 27% have an official and comprehensive AI strategy. 22% responded that they are prepared for the AI Framework Act system. Risks on the consumer side appear as the 31.21% who disallow autonomous spending, as seen earlier.

When moving to the outlook, one must check the reference date.

The figure known as a Gartner press release dated January 15, 2026, is that worldwide AI spending in 2026 will total approximately $2.52 trillion (also noted as $2.53T depending on the citation page), a 44% increase from the previous year, and will exceed $3 trillion in 2027 with a growth rate of about 30%.

Narrowing it down to agents, the Software Strategies Blog summary on February 16, 2026 states that Gartner forecasts agentic AI spending will overtake chatbot spending in 2027. According to a summary on December 22, 2025, the proportion of application software spending that includes agent features will go from 2% in 2024 to over 50% in 2028, and spending on software without agents will begin to decrease from 2027.

The direction points to one place, but the definitions of the numbers differ. Like the gap between the $201.9 billion and $7–8.5 billion seen earlier, one must first determine whether embedded features are included when reading forecast figures.