Author: qinbafrank

In the past few days, I have gone through the financial reports of Agent companies in the US and A-shares and had a strong impression: Agents are beginning to appear in financial statements.

In the past year, the market observed Agents mainly focusing on whether the model could invoke tools, complete multi-step tasks consecutively, and how many Agents a company had released.

Now, during this earnings season, the observational metrics have significantly changed. We are now looking at how much ARR Agents contributed, how much ACV they generated, how many new orders they brought in, how many customers entered production environments, and how much work was actually completed each quarter.

This change is important.

Being able to run the product only indicates that the technological direction is valid. Revenue, contracts, payments, and profits beginning to appear is what shows that companies are willing to continuously pay for this capability.

Thus, I tend to define 2026 as the "financial verification starting point and the first year" for enterprise-level Agents. This refers to the verification first year; there is still a long way to go before the industry matures, but at least the first batch of real data that can be included in financial models has already emerged.

1. The US stock market has already provided the first round of answers

At present, the clearest companies are all essentially managing internal data, permissions, and workflows within enterprises.

1. ServiceNow reported a Q2 subscription revenue of $3.877 billion, up 24.5% year-on-year; the AI business ACV has surpassed $1 billion, and the deployment of Agents increased 9 times in nine months.

Previously, many viewed ServiceNow as merely IT ticketing and workflow software; now it increasingly resembles a control center for enterprise Agents.

Internally, enterprises may simultaneously run Claude, Gemini, OpenAI, and various vertical models in the future. Which Agent can access which data, represent whom in task execution, which actions require approval, and how to roll back after errors—all these questions require a unified governance layer.

ServiceNow has already positioned itself in this role. Its offerings are gradually extending from workflow software to the identity, permissions, monitoring, and execution systems of Agents.

2. Salesforce's validation is also quite direct

Agentforce’s ARR has exceeded $1.5 billion, an increase of more than 240% year-on-year; combined, Agentforce and Data 360's ARR is nearly $3.9 billion. In Q2, Agentforce and Slack completed 3.2 billion Agentic Work Units, an increase of 97% quarter-over-quarter.

In the past, discussions around Salesforce expressed concerns that general large models would bypass CRM and directly take over sales and customer service. Looking at it now, large models still need customer data, historical orders, contract status, sales leads, and permission systems to truly accomplish enterprise tasks.

Claude may be responsible for understanding intent, while Salesforce manages the business context and execution paths.

Of course, a 240% year-on-year growth rate needs some qualification. Starting this quarter, Salesforce expanded the statistical scope of Agentforce ARR to include other AI products, Slackbot, and Headless 360. The direction is very clear, but the specific growth rate still needs to be assessed in conjunction with the changes in calculation criteria.

3. Workday's financial report is also very representative

AI has contributed to over 25% of the new ACV, with more than 5,500 customers using at least one self-developed Workday Agent.

This set of data shows that enterprises are not simply bypassing original software platforms in sensitive scenarios like HR, finance, and auditing. Employee data, compensation rules, financial regulations, and organizational permissions are all embedded within Workday, and Agents need to operate along these deterministic tracks.

For many traditional SaaS companies, AI may indeed compress some screen pages, buttons, and low-value seats. However, platforms with core data and business rules might find new upselling opportunities due to the introduction of Agents.

4. Palantir represents another form

Q2 revenue grew by 93% year-on-year, with U.S. commercial revenue rising by 149%. Palantir's strength has always been in connecting enterprise data, business entities, permissions, and actual production processes, and then using software to make decisions and execute.

Many Agent products remain at the stage of "helping you generate an answer," while Palantir is closer to "helping enterprises achieve a result."

Manufacturing companies care about how much downtime has been reduced, supply chain companies care about how much inventory turnover has improved, and sales teams care about how much conversion rates have risen. Ultimately, enterprises are willing to pay a high price for these quantifiable business results.

When looking at these companies together, a very clear trend emerges:

The first to benefit from Agent dividends are typically those companies already situated within the core business systems of enterprises.

While model capabilities are certainly important, deploying Agents in enterprises also requires data, permissions, workflows, auditing, and customer entry points. Simply connecting a model and creating a chat interface will quickly lower the threshold. This ties back to a previous lengthy discussion about the engineering era of AI adoption: True SaaS that possesses workflow and exclusive core data holds the entry points to work, business entities, business semantics, user permissions, historical operation data, system records, industry rules, final action execution, and customer result feedback.

2. The most important change in this earnings season is actually the change in measurement methods

In the past, when judging AI products, the market often looked at registered users, trial customers, and token invocation levels. These metrics can prove interest but are hard to validate commercial viability.

Now increasingly more companies are starting to disclose: how much new ACV AI contributed, how much ARR Agents generated, how many customers transitioned from trial to production environments, and how many tasks were actually completed.

Salesforce's use of Agentic Work Units to measure workload is a significant signal.

In the past, SaaS mainly charged based on seats: if a company has 10,000 employees, it purchases 10,000 accounts. In the future, as Agents complete numerous tasks for enterprises, the charging method might gradually change to:

Base seat fee + AI usage + task completion volume;

Further down the line, in some high-value scenarios, even a results-based fee structure might emerge.

For example, how many tickets the customer service Agent resolved, how many leads the sales Agent screened, how many invoices the tax Agent processed, and how many risk events the security Agent closed.

Once the billing unit shifts from "how many people are using the software" to "how much work the software has completed," the income ceiling for SaaS will also change.

It will start cutting into the enterprise's labor costs, outsourcing expenses, and operational budgets, rather than merely reallocating within the IT budget.

3. Why the improvement in model intelligence will make Agents a long-term industry

There is a very clear capability threshold for the Agent line: assuming an Agent has a 95% success rate for executing a single-step task, after executing ten steps consecutively, the probability of successfully completing the entire task drops to about 60%.

Thus, early Agents often faced a problem: each step appeared quite intelligent, but when placed within longer processes, errors began to occur. Enterprises still needed to arrange for employees to check, correct, and rework, resulting in little time saved.

As the model's reasoning capability, context length, tool invocation, and memory abilities improve, this situation will gradually get better: the more steps an Agent can complete consecutively, the fewer points will need human intervention, and tasks that were previously impossible to automate will surpass commercialization thresholds.

This is not a uniformly progressive process.

An improvement in model capabilities might only lead to smoother answers; however, once reliability surpasses a certain threshold, an entire workflow has the opportunity to be completed automatically.

Thus, Agent income may display strong non-linearity in the future.

Today, it can only organize information, the next step might include filling out systems, and afterward making submissions for approvals, tracking progress and handling exceptions. Each additional completed step will significantly enhance commercial value.

This is why Agents do not need to first achieve so-called general artificial intelligence.

Financial, customer service, IT operations, sales lead screening, and document processing all involve a high volume of tasks with clear boundaries, high repetition rates, and verifiable results. This relates back to the second scenario discussed after the topic of 'bad coding':

As long as Agents can reliably complete a substantial portion of these tasks, the client ROI is already established.

4. The A-shares have also begun validation, but the path differs from the US market

The US market is currently the first to realize the potential of enterprise platforms and workflow companies; while the A-shares are more focused on vertical scenarios.

1. Kingsoft Office reported revenue of 3.313 billion yuan in the first half of the year, a year-on-year increase of 24.69%, with WPS 365 maintaining over 60% growth for six consecutive quarters.

The advantages of Kingsoft Office are quite intuitive. It possesses document access, formatting capabilities, collaborative relationships, and corporate office data. Future office Agents could directly generate, modify, review, and deliver documents, which is far more valuable than merely placing a chat window next to a document.

However, Kingsoft Office's net profit attributable to the parent grew by over 200% in the first half, much of which comes from investment income. When examining the core business, these one-time projects still need to be stripped out.

2. Taxu Friends is a relatively typical case of a vertical Agent in the A-shares

In the first half, revenue was 1.012 billion yuan, an increase of 9.72% year-on-year; net profit after deductibles grew by 44.04%. AI products and business repayments now account for 33% of the intelligent finance and tax business repayments, higher than the 28% for the entire year of 2025.

The finance and taxation scenarios are very suitable for Agents. The rules are relatively clear, the data is highly structured, the results can be verified, and clients are willing to pay to reduce labor costs and tax risks.

Such businesses do not need models to dominate entirely. As long as they perform reliably in a narrow scenario, they can generate real income.

3. Hand Information

In the first half, AI smart business revenue was approximately 220 million yuan, more than doubling year-on-year, and has already accounted for a double-digit percentage of the company's revenue.

Hand has long been involved in ERP, supply chain, manufacturing, and financial system implementation. The biggest challenges in implementing enterprise Agents often arise from connecting old systems, non-standard data, and complex approval processes. Hand is familiar with these stages and has the opportunity to become the deployment and integration layer for domestic enterprise AI.

However, A-shares should particularly pay attention to fiscal quality, so while there are great opportunities for Agents here, the analysis may be more complex than for US stocks.

Many companies are still primarily project-based revenue, whether AI businesses can be standardized, whether gross margins can improve, and whether it can transition from one-time implementations to subscriptions and ongoing usage fees still require ongoing observation.

5. The Agent industry chain may form a four-layer structure

The first layer consists of basic models and computing power, which determines the intelligence ceiling and reasoning costs of Agents.

The second layer includes data, identities, permissions, and governance. ServiceNow, Salesforce, and Workday are currently competing for this layer.

The third layer is workflow orchestration and execution. Palantir, UiPath, and some enterprise digital service providers are competing in this realm.

The fourth layer is vertical industry Agents, including finance and taxation, office, legal, medical, manufacturing, customer service, and sales.

Currently, the clearest financial validation exists within the second layer.

Businesses are hesitant to allow an Agent that lacks permission boundaries and cannot be audited to directly operate core systems, so data and governance platforms have a strong positional advantage.

The long-term space that has more elasticity is likely the fourth layer. Vertical Agents can correspond directly to specific workloads, ROI is easy to calculate, and it is also easier to tap into labor and outsourcing budgets.

The third layer determines whether Agents can truly enter production environments. It's not enough for models to be able to think; they also need to connect to old systems, invoke tools, handle exceptions, and complete processes fully.

UiPath will be a very important observation sample. The company plans to release its Q2 earnings report on September 3, 2026, Eastern Time; at that time, further judgments can be made on the speed of traditional RPA upgrading to the Agent execution layer.

6. How to observe Agent companies going forward

In the future, when observing this type of company, I believe several indicators will become increasingly important:

1. First look at AI-related ARR, ACV, and real income, and then observe the number of customers in production environments; the reference value of trial customers will decrease;

2. Next, look at the contribution of AI to new contracts. Workday’s disclosure that “AI accounted for over 25% of new ACV” is more meaningful than reporting how many Agents were released;

3. Also, keep an eye on task invocation volume, full task success rates, renewal rates, and customer expansion rates;

4. Finally, return to financial reports to observe gross margins, implementation cycles, accounts receivable, and operating cash flows after deducting reasoning costs.

If each Agent deployment requires a large number of engineers for half a year of customization, it is more akin to traditional IT services.

If Agents can replicate quickly, the more clients use them, the lower the unit cost will become, and the operational leverage of software companies will truly be unleashed.

Finally, understanding this Agent line:

1) Model intelligence enhancement determines how much Agents can accomplish;

2) Enterprise data, permissions, and workflows determine whether this intelligence can safely enter production systems;

3) Real task results determine how much clients are willing to pay.

Every upgrade to the model pushes the boundaries of tasks that can be automated outward.

Today it is programming, customer service, and documents; later it will enter finance, sales, supply chain, medical, and more specialized scenarios. As reliability improves, manual checks will decrease, and unit economics will steadily enhance.

Therefore, Agents might be a new efficiency-oriented industry that can sustain for many years.

However, this round of opportunities will not be evenly distributed.

The first group of winners will likely be those platforms that have already controlled enterprise data and business processes. The second group will comprise companies capable of deploying AI into production environments and genuinely completing task loops. The elasticity in A-shares comes more from vertical scenarios, but fiscal quality and income standardization need to be monitored closely.

2026 is more likely to be the financial verification first year for Agents. In the past, market discussions focused on how intelligent models were; going forward, there will be increasing attention on: how much work they actually completed, how much they helped clients save, and how much value ultimately enters the revenue, profits, and cash flows of listed companies.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。