Anthropic has replaced much of its traditional sales-contact process with an AI buying agent, and the early results reported by the company are significant. Prospects passed from the agent to its sales team convert to opportunities at more than twice the rate of leads coming through the old contact form, deals reportedly close about five days faster, and the share of conversations requiring human help has been cut in half.
The interesting part is how the agent produces those results. It answers questions about pricing, security and data terms, asks follow-up questions, determines what the buyer needs, recommends a plan and can take some customers through checkout. When human involvement is appropriate, the salesperson receives the conversation and enters with substantially more context than a traditional contact form could provide.
This is a good example of where enterprise AI is headed. The agent is no longer sitting behind the employee helping draft an email or summarize a meeting. It is representing the company directly to the customer and influencing what happens next.
That changes the governance problem considerably.
The Agent Is Becoming Part of the Brand
A contact form has very little behavioral latitude. It collects the information it was designed to collect and passes it somewhere else. An AI agent can interpret questions, decide which information is relevant, make recommendations and determine when another person or system should become involved.
That discretion is precisely what makes the agent more useful, but it also means the customer’s experience with the brand increasingly depends on decisions made by probabilistic AI.
Anthropic’s example contains an especially interesting detail: Its agent sometimes recommends a lower-priced plan when that is the better fit for the customer. That is a sensible commercial strategy because maximizing the immediate transaction is different from maximizing the customer relationship. It also illustrates why the objective assigned to an agent matters so much.
Tell an agent to maximize sales and you may get very different behavior than if you tell it to help the customer find the appropriate solution. Even carefully written instructions, however, remain instructions interpreted by a probabilistic system. As agents gain more discretion, enterprises need a way to make important behavioral requirements enforceable rather than merely suggested.
The AI agent is becoming part of the brand experience, which means its behavior becomes a brand-governance issue.
The Customer Adds Another Dimension to Governance
Most agent-governance discussions concentrate on the machine side of the interaction: Which systems can the agent access? Which tools can it use? What transactions can it execute? When does it need authorization?
All of those controls matter in a sales environment. An agent recommending products, discussing contractual requirements or initiating purchases needs clearly defined authority. But the interaction contains another source of information that conventional agent controls generally do not understand: The person trying to buy something.
A prospective customer may be confused by pricing, frustrated by a requirement, concerned about security or increasingly dissatisfied with the conversation. Those signals can materially change what a well-governed organization wants its AI to do next. Continuing to optimize toward checkout when the customer is becoming frustrated may satisfy a narrow transactional objective while damaging the larger commercial relationship.
This is where emotional intelligence becomes relevant to governance rather than simply conversational polish.
VERN’s Emotion Recognition System is an independent, deterministic neurolinguistic model operating outside the LLM. It identifies emotional signals expressed during the interaction and makes them available to VERN OS as inputs into deterministic behavioral controls.
That gives an organization another dimension on which it can govern the agent. A meaningful change in the human interaction can trigger different behavioral requirements, escalation conditions or other controls defined by the enterprise rather than improvised by the underlying model.
Emotional Intelligence Should Not Depend on the Sales Agent Guessing
LLMs can infer emotion from language, but asking the sales agent itself to interpret the customer’s emotional state creates an important architectural dependency. The same probabilistic system is determining what the customer means, deciding how the customer feels, selecting what commercial objective to pursue and generating the response intended to accomplish it.
For a casual chatbot, that may be acceptable. For an AI increasingly responsible for representing a company and moving customers toward consequential decisions, enterprises should consider whether those responsibilities belong inside the same probabilistic loop.
VERN separates them.
The underlying AI remains free to reason about the customer’s question, search information and determine how best to accomplish its assigned task. VERN’s deterministic emotional model independently evaluates signals from the human interaction, while deterministic Behavioral Control Modules govern how the AI is permitted to behave when defined conditions occur.
This does not require pretending that software can know precisely what someone is thinking or diagnose an internal psychological state. The purpose is to consistently recognize signals expressed in language and allow the organization to determine what those signals should mean for the behavior of its AI.
That distinction becomes increasingly important as the AI assumes more responsibility for the customer relationship.
The Sales Agent Is Also Part of an Agent Network
Anthropic’s buying agent may look like a customer-facing application, but this category will increasingly become part of larger agentic systems. A sales agent could consult a pricing agent, query inventory, interact with a CRM agent, request contract information, initiate an approval workflow and then return to the customer.
The interaction can move from agent to agent to human and back again.
Governance needs to move with it.
VERN OS is designed to provide one behavioral governance architecture across those environments. During agent-to-agent interactions, deterministic controls can govern role, authority, delegation, escalation and consequential actions. When the workflow reaches a person, VERN’s deterministic emotional model adds information about what is happening on the human side of the interaction.
The enterprise therefore does not need separate philosophies for governing autonomous agents and customer-facing AI. The same control layer can govern the workflow across both, while incorporating additional human signals where they become relevant.
This is particularly important as enterprises adopt multiple underlying models. The customer-facing agent may use one LLM while specialized agents elsewhere in the workflow use others. Governance should remain consistent regardless of which probabilistic intelligence happens to be performing each task.
AI Agents Are Moving Into the Customer Relationship
Anthropic’s reported results make the business case for this transition easier to understand. If an AI agent can respond immediately, answer routine questions, qualify the buyer and give salespeople better context when human involvement becomes useful, the traditional contact form begins to look remarkably primitive.
Other companies will follow.
The larger consequence is that enterprises will increasingly put probabilistic AI directly between their brands and their customers. Those agents will make thousands or millions of small behavioral decisions that collectively determine how people experience the organization.
Permissions and cybersecurity controls will remain essential, but they address only part of that responsibility. Enterprises also need to govern how the agent behaves, when its behavior should change and how signals from the person on the other side should affect the interaction.
That is why we believe emotional intelligence belongs inside the AI governance architecture.
VERN combines deterministic behavioral controls with an independent deterministic emotional model, creating one governance layer capable of following AI from agent-to-agent execution through agent-to-human interaction.
Anthropic’s experiment demonstrates what can happen when an AI agent replaces a form.
The larger transformation begins when the AI agent starts representing the company.
At that point, governing the agent means governing part of the brand.
Source: https://www.inc.com/robert-gillespie/anthropic-replaced-contact-form-for-ai-agent/91415246
