The number is impressive. The context is more complicated.
Salesforce just told the market its AI agent business, Agentforce, grew by over 200%. On the surface, this reads as yet another victory lap in the enterprise AI arms race. But if you look beneath the press release, you will find something far more interesting: a quiet transformation of how software companies will charge for value, a data moat that is more durable than any model capability, and a looming pricing conflict that could reshape the entire SaaS ecosystem.
The bubble burst, the lessons remain — but this time, the lessons are about business models, not tokens.
As a macro watcher who has spent nearly three decades tracking how technology capital flows through global markets, I see this as a watershed moment. The "200% growth" headline is not the story. The story is what it tells us about the maturation of enterprise AI, the shifting economics of software, and the uncomfortable truth that your customer service representative may soon be an algorithm with a Salesforce badge.
Let me break this down with the rigor it deserves.
The Hook: When the Growth Number Conceals More Than It Reveals
The announcement came through the usual channels: a quarterly earnings call, a carefully crafted press release, and a CEO with a rehearsed talking point. "Agentforce bookings grew over 200% quarter-over-quarter," the narrative went. The market nodded approvingly, analysts adjusted their models, and the stock ticked upward.
But I have been in this game long enough to know that growth percentages without absolute bases are like yields without risk adjustments. If last year's base was $50 million, a 200% increase means $150 million. That is a rounding error against Salesforce's total revenue of approximately $37 billion. The message is not wrong, but it is incomplete.
The question is not whether Agentforce is growing. The question is whether this growth trajectory is sustainable and whether the underlying economics will hold as the volume of conversations scales.
This is exactly the kind of situation where a "quantitative skepticism engine" must be activated. We can't just accept the headline; we need to look at the underlying mechanics of the business model. In a market as noisy as enterprise AI, it is precisely the numbers that aren't in the press release that matter.
The Context: What is Agentforce, Actually?
For the uninitiated, Agentforce is Salesforce's AI agent platform, a direct bet on the "agentic AI" trend. It is not a foundation model. It is a layer of orchestration and business logic that sits on top of models from OpenAI, Anthropic, Google, and others, routing their capabilities through a "reasoning engine" called Atlas and translating outputs into actions inside the Salesforce ecosystem via "Atomic Actions."
The technical architecture is not groundbreaking in the way a new model release is. It is groundbreaking in a different way: it brings enterprise structure to AI chaos.
This is where the story gets interesting. Agentforce is not about the model's inherent intelligence. It is about applying that intelligence to the structured business data that Salesforce has been accumulating for over two decades. This includes customer records, order histories, service tickets, and marketing engagement. The platform then executes the action in a business context.
A conversational AI agent that can say "I understand you want a refund" is one thing. An agent that can access the customer's order history, validate the refund policy, check inventory levels, and then initiate a refund through the company's ERP system is something else entirely.
The engineering challenge is not the model; it is the integration, the permissioning, the audit trail, and the reliability. Salesforce's technical moat is the data access layer and the workflow orchestration, not the algorithm.
The Core: Decoding the Commerce of Agents
This brings us to the real core of the story: the commercial model. Salesforce has made a daring move that challenges the entire pricing philosophy of the software industry. Instead of charging per user or per seat, Agentforce charges per conversation at $2 each.
The Pricing Paradigm: From Seats to Outcomes
This is a fundamental shift. The old SaaS model—charging a flat fee per employee—is a proxy for value. You assume that if you have 500 customer service agents using your software, they will generate 500 seats' worth of value. But with Agentforce, you are essentially betting on a "result-oriented" pricing model.
The model is a major break from the traditional "per seat" model. It aligns the cost of the software with the actual task, not the potential productivity of a human. This is the future of software: you don't pay for the tool; you pay for the job it does.
This model has an elegant logic. It aligns the cost of the tool with the value it creates. However, it also shifts the risk from the customer to the vendor. If the agent fails to resolve a customer's query, Salesforce doesn't get paid for that conversation. If the agent is inefficient and requires 20 conversations to solve what a human could solve in one, the customer is effectively paying 20 times for a single outcome, which creates a perverse incentive.
The Unit Economics and the Hidden Risks
The biggest concern I have is the unit economics. The $2-per-dialog price is a public number, but what is the underlying cost? Salesforce doesn't train its own foundation models. It calls APIs from OpenAI, Anthropic, and Google. Every single conversation runs inference on a model, and that has a real cost.
If the cost of a single conversation is close to $2, the margin is extremely thin. If model providers raise prices, or if more complex conversations require more expensive models, the margin could quickly turn negative. The "200% growth" could turn into a "200% margin squeeze" if the underlying costs are not carefully managed.
My analysis indicates that Salesforce is relying on its enormous scale as a customer to negotiate volume discounts with model providers. But this is not a sustainable long-term moat, as the bargaining power of a few customers cannot last indefinitely against a handful of dominant model suppliers.
The "Base Effect" and the Absolute Numbers
We must also consider the "base effect." A 200% growth rate is an amazing number, but it doesn't tell us the absolute revenue. If the base is small, the absolute numbers might not be significant yet. For Salesforce, a $37 billion revenue company, an AI agent business contributing a few hundred million dollars is still a marginal business, not the main driver.
This is a common narrative trap in the crypto and tech worlds. We celebrate percentages without looking at the absolute numbers. The "200% growth" is a good headline, but it's the absolute revenue contribution and the path to $1 billion+ that will matter for the stock price.
The Contrarian Angle: This is a defensive play, not a new horizon
The market is reading this as a bold new frontier for Salesforce. I see it differently. This is a defensive move.
Salesforce's core business—its CRM dominance—is being threatened. On one side, there are AI-native startups like Sierra, which are building customer service experiences from the ground up, without the baggage of legacy CRM systems. On the other side, there are platform giants like Microsoft, which are integrating AI agents into their Office and Azure ecosystems.
If Salesforce doesn't provide the AI layer for its existing customers, someone else will. Agentforce is a way to protect its core business by using the AI wave to strengthen its position.
The Moat: Data and Workflow
The real moat is not the AI itself but the data and the workflow. Once a company starts running its customer service through Agentforce, it is essentially training the system on its most sensitive business logic and customer data. This is not a relationship you can easily walk away from.
The "composability" of these agents is a double-edged sword. On the one hand, the modular design of "Atomic Actions" allows for flexibility and customization. On the other hand, the complexity of these workflows makes it very difficult to leave the ecosystem.
Switching costs are high, and the data flywheel starts to spin. The more conversations the agent handles, the more data it generates, which makes the agent smarter and more effective, which attracts more users. This is a powerful cycle.
The Real Threat: The "Tax" of Complexity
However, the "data moat" is also a weakness. The same complexity that makes it sticky also makes it expensive. The "Atomic Actions" library, which maps model outputs to business processes, has a limited scope. For non-standard, long-tail business processes, the customization cost could be significant.
This is a hidden risk. The "composability" of the system is a "double-edged sword." It allows for rapid deployment, but it also creates a new layer of complexity that requires specialized skills. The "AI agent" may be easy to deploy, but the "AI agent" that is truly integrated into a business is complex and expensive to build.
The Macro View: The Labor Market and the "Digital Labor" Revolution
We need to look beyond the immediate business and consider the macro implications. What does it mean for the labor market when a company like Salesforce is selling "digital labor" at $2 per conversation?
The Employment Question
It's not a "if" but a "when" and "how fast". Customer service representatives, telemarketers, and junior marketing roles are the most exposed to this shift. We are talking about a global, multi-million-person labor force.
This is the biggest risk for Salesforce, not from a financial perspective, but from a social and political one. If large-scale deployment of AI agents leads to massive job losses in the customer service sector, it will create a wave of public backlash and calls for regulation. This is exactly what happened in the crypto market with the question of "decentralization" versus "centralization."
The Regulatory Crosswinds
The regulatory environment will be a major factor. The EU AI Act, for example, could classify Agentforce's use in customer service as a "high-risk" AI system, requiring transparency, human oversight, and data governance. This could slow down its adoption in Europe.
The Chinese market is also a factor. As a cross-border payments researcher, I see the regulatory challenge of the Chinese market. Salesforce's global AI services cannot be directly deployed in China, and they need to partner with local providers like Alibaba Cloud to comply with the local AI model filing requirements.
The Cultural Shift in Software Pricing
The most profound macro impact is on the software industry itself. If Salesforce's "pay-per-conversation" model proves successful, it could trigger a new era of "outcome-based" pricing across the entire SaaS industry. This is a move away from the "per-seat" model to a "per-value" model, which is a huge shift.
This is a "shift" in the software industry, where the value of software is no longer measured by the number of employees who use it, but by the actual business results it delivers. This is a more mature and more honest way to price software, but it's also a lot harder to pull off.
The Trust Layer: The Invisible Architecture
One of the most overlooked components of the Agentforce architecture is the Einstein Trust Layer. This is the component that ensures that sensitive data is not used for model training, prevents prompt injection attacks, and provides an audit trail for all agent actions.
In my experience, the enterprise adoption of AI is not a matter of "what is the best model?" but "is it safe to use?" The Trust Layer is the "trust" that allows enterprises to let AI agents loose on their customer base.
The risks of data leakage, hallucination, and bad behavior are real. The challenge is not the model's intelligence but the ability to control it. This is where Salesforce is investing heavily, and it is the right place to invest.
The Responsibility Question
However, there's a question that remains unanswered: who is responsible for the AI agent's mistakes? If an agent gives wrong advice, or makes a bad promise, who is accountable? Is it Salesforce, the customer, or the model provider?
The law is not clear on this. This is the "new frontier" of enterprise AI, and it's a legal and ethical minefield.
The Investment Lens: The "P/E" of the Future
From an investment perspective, Salesforce's stock price reflects its "AI premium." The market is already giving it a higher P/E ratio than traditional software companies, based on the expected growth of Agentforce.
The "P/E" of the Future
The investment logic is simple. If Salesforce is the "operating system" for the digital workforce, it will be a huge business. If it is just a "feature" of the CRM, it will be a nice add-on but not a major revenue driver.
The market is betting on the former. The current valuation is already pricing in the "agentic AI" narrative.
The Sustainability of the 200%
The question is whether the "200% growth" can be sustained. The key is not just the number of conversations but the value of each conversation. A $2-per-conversation model will create a very high revenue per customer, but it also means that a high failure rate will lead to a high churn rate.
The "AI agent" is not a one-time purchase; it's a continuous service. If it doesn't deliver value, the customers will stop using it. The "200% growth" is only the beginning; the "sustainability" is the real test.
The "digital labor" and the Global Macro
As someone who has been watching cross-border payments, I see a clear connection to the global macro. The "digital labor" trend is about the arbitrage of labor costs, but instead of moving jobs to a different country, the jobs are being replaced by software.
This has implications for the global economy. If you can have an "AI agent" in a Salesforce cloud handle a customer service query for a company in the United States, the labor cost of the US is no longer a factor. This is a new form of globalization, and it could have a huge impact on the global labor market.
The "Data" is the New Oil
The key to the "AI agent" is the data it has access to. This is the "Data Cloud" that Salesforce has built. The data that's been collected by the CRM is the "oil" of the AI age. It is the data that makes AI agents actually useful.
This is the "new" moat. The models are all the same, but the data is unique. The "data flywheel" is the core competitive advantage.
The Contrarian Angle: The "Decoupling" Thesis
Let me put on my "Contrarian" hat and consider a scenario that the market might be getting wrong. The market is treating Agentforce as a "new growth engine." But what if it's just a "defensive move" that doesn't create a new market, but simply protects the existing one?
The "decoupling" thesis is about the decoupling of the AI value from the software value. In the long run, the "AI agent" might become a commodity. The models are available to everyone, and the "workflow" might be copied. The "data" is the only differentiator.
If the "data" is the moat, then the "data" is also the risk. If a customer decides to leave Salesforce, they will lose the "data" and the "AI agents" that have been built on top of it. This creates a "lock-in" effect, which is good for Salesforce but potentially a problem for the customers.
The "Sleeper" Threat: The "Composability" of the AI
The "composability" of the AI ecosystem is a double-edged sword. On the one hand, it allows for a "modular" approach to AI. On the other hand, it creates a system of "interconnected dependencies."
If Salesforce relies on OpenAI's models, and OpenAI has a problem, Salesforce has a problem. If the "Atomic Actions" are too complex, the "composability" is a liability.
Algorithms don't fail; models do.
The "Outcome-Based" Economy
The "2$ per conversation" model is a "result-oriented" pricing model. But the "result" is the key. What is the "result" of a conversation? Is it a "solved ticket" or a "customer satisfaction" or a "sale"?
If the "result" is not clear, the "pricing" will be a problem. This is the "new frontier" of the SaaS industry.
The Takeaway: Positioning for the "Digital Labor" Cycle
So, what is the takeaway for the market, the "macro" perspective?
The "200% growth" of Agentforce is not just a number. It is a signal. It is a signal that the enterprise AI is moving from a "tool" to a "worker." It is a signal that the SaaS business model is changing. It is a signal that the global labor market is about to change.
The "bubble" of "AI hype" has not burst, but the "lessons" are being learned. The "lessons" are that the "value" is in the "workflow" and the "data," not the "model."
The "AI agent" is not a "product" but a "service." The "service" is the "digital labor." The "price" of the "digital labor" is the "price" of the "outcome."
The "future" is not about "seats" but "sessions." The "future" is not about "people" but "processes." The "future" is not about "software" but "services."
The "positioning" for the next cycle is not about "buying" the "AI stocks" but "understanding" the "AI economics."
The "question" is not "how many people" but "how many conversations."
We're not watching a "software company" grow. We are watching the "labor market" being re-architected. The "stock price" of Salesforce is just the surface. The "underlying" is the "redefinition" of "work."
As we move into the next stage of this cycle, the "winners" will be those who understand the "new economics" of "digital labor." The "losers" will be those who are "stuck" in the "old world" of "per-seat" pricing and "human-only" labor.
The "signal" is clear. The "question" is: are you listening?