Since April 2026, HubSpot has charged for its AI agents per completed task rather than by subscription. The predictable result: portals that had activated a prospecting agent here and a service agent there ended up with AI building blocks each running in isolation, with no overall view and no shared governance. Agent Hub is HubSpot's answer to this problem, launched in public beta on July 23, 2026: a single space to view, activate and manage all the Breeze agents in a portal, backed by Agent Builder to build new ones. Here is its actual scope, what changes compared with activating agents one by one, and what you need to know before devoting team time to it.
A prospecting agent running in Sales Hub. A service agent answering tickets in Service Hub. An enrichment agent running in the background in the CRM. Each has its own configuration interface, its own results, its own Analyze tab. Nobody in the organization has a cross-functional view.
This is the observation HubSpot highlights to justify Agent Hub: without a central registry, an organization that multiplies agents eventually no longer knows how many it has, which are actually active, and which have been running idle for three months. The term used by the trade press is “agent sprawl”, the uncontrolled spread of AI agents as they multiply within the same system.
This diagnosis is correct, but it needs qualifying. A hub that centralizes the view of poorly configured agents mostly centralizes the view of failures. Agent Hub solves a visibility problem, not a data quality or upstream configuration problem. The two topics are linked but distinct, and confusing them is the main source of disappointment we observe in portals that activate too many agents too quickly.
Agent Hub is found in HubSpot's navigation under Agents > Agent Hub. The interface is organized around a view called “Your agentic customer journey”, which maps where AI tools are active along the customer lifecycle, from first contact to post-sale support.
Three concrete functions are attached to it. First, a real-time view of the status and performance of each active agent in the portal: how many tasks handled, what completion rate, which alerts are open. Second, activating or deactivating an existing agent in one click, without going back through its original module. Finally, direct access to Agent Builder and the Agent Marketplace, the library of pre-built agents offered by HubSpot and its partners.
Results are grouped by business objective rather than by tool: generating demand, closing deals, satisfying customers, accelerating growth. This presentation choice has a real merit: it forces you to think in terms of objectives rather than features, which matches the way a sales or marketing director manages their business better than a technical list of agents.
A Context tab defines the business information all agents can use. This is where knowledge vaults are configured: spaces that store product documentation, support guides or any other resource intended to feed a specific use case.
This configuration point should not be underestimated. A service agent answering without an up-to-date knowledge base invents plausible but wrong answers. The quality of the content placed in the knowledge vaults directly determines the reliability of everything that runs on top of it.
Agent Builder is the creation tool that comes with Agent Hub. What sets it apart from a simple configuration form is a single canvas where you build an agent and the automation that triggers it at the same time.
An agent can be triggered by a recurring schedule, a property update on a contact record, an incoming webhook or a third-party integration. The desired behavior is described in natural language via Breeze Assistant, which translates the intention into executable logic. Another notable point: an agent built in Agent Builder natively uses the CRM's contextual database, with no additional configuration needed to give it access to the history of contacts, deals or tickets.
This is where the platform's real promise lies. A generic agent built outside HubSpot has to be connected to the CRM, authorized, and mapped field by field. An agent built in Agent Builder starts with this access already in place. The logical trade-off is that it only works inside the HubSpot ecosystem.
An example built at Markentive illustrates the mechanism well: an account handoff agent, designed for when an account changes owner. The user enters only two pieces of information: the company name and the new owner's role. The agent finds the company in the CRM, goes through all associated objects (contacts, deals, tickets, activities), and extracts customer verbatims and sentiment where they exist. The instructions require a precise balance: at least 80% of the report must come directly from CRM data, and the remaining 20% can include external news about the company to provide context.
The result is a structured handoff report (with a title such as “Handoff Report from [outgoing role] to [incoming role]”, history, opportunities, points of attention), written from the outgoing owner's point of view so that the new owner has no information gaps. Nothing in the instructions specifies where to find the contact or deal history: the agent already has native access to it, without prior mapping. This is exactly what Agent Builder promises.
Agent Hub does not create these agents; it makes them visible and manageable in one place. The current portfolio includes:
The Customer Agent, which handles first-level support conversations and the resolution of simple tickets. The Prospecting Agent, which researches target accounts, writes personalized prospecting messages and manages follow-ups. The Data Agent, responsible for enriching CRM data and maintaining its quality. An AEO-focused agent, which monitors the brand's visibility in answers generated by AI engines. A deal progression agent, which recommends the next step on an open opportunity. And a knowledge base agent, which powers self-service support.
Some of these agents have their own billing logic, such as the Prospecting Agent, billed per recommended lead since April 14, 2026 rather than through a fixed subscription. Agent Hub does not harmonize these pricing models; it displays them side by side.
Access to Agent Hub and Agent Builder is reserved for Professional and Enterprise accounts, on Marketing Hub, Sales Hub, Service Hub, Data Hub, Content Hub or Smart CRM. No pricing specific to Agent Hub itself has been announced to date: access depends on the edition already subscribed to, and running agents consumes HubSpot credits, as was already the case before the hub was released.
In practice, a Starter portal does not see Agent Hub appear in its navigation. This is a point to check upfront for any organization considering an edition upgrade mainly for this feature: it is not enough on its own; it orchestrates agents which themselves must already bring individual value that justifies their cost in credits.
HubSpot is not alone in this space. Salesforce led the way with Agentforce, its own autonomous agent platform, and most major CRM vendors have since announced a comparable agentic layer. The vocabulary is converging from one vendor to the next: orchestration, agent marketplace, low-code building canvas. This is no coincidence. It shows that no vendor yet has a stable lead on the topic, and that organizations choosing HubSpot today for Agent Hub are doing so on the promise of a roadmap, not on a fully mature product.
What sets HubSpot apart in this race remains its CRM database, unified from the start. An agent built with Agent Builder inherits this history without prior connection work, a structural advantage that CRMs assembled through successive acquisitions cannot reproduce as easily. It is a real differentiator. It is not a guarantee of results: an agent well connected to a poorly maintained database is still an agent producing bad recommendations, only faster.
Two reservations should be stated before opening Agent Hub to a whole team.
The first concerns the order of the work. Activating an orchestration hub before having a single agent working properly on a clean database fixes nothing: it gives a consolidated view of poorly tuned agents, not better results. The order is the same as for Breeze in general: data structure and quality first, a well-tuned agent next, cross-functional orchestration last. An organization that activates four agents in one week because the hub allows it in one click gains nothing: it simply multiplies the possible points of failure.
The second concerns governance. One click to activate an agent is also one click to activate a bad one, or to leave one running that has been useless for months. Without regular reviews of active agents, a practice the hub makes easier but does not enforce, an organization reproduces at scale the dispersion problem it was trying to solve. The number of active agents is not an indicator of AI maturity. It is often the opposite.
Our position at Markentive: Agent Hub is a good monitoring tool for an organization that has already sorted out its data and validated one or two agents in real conditions. It is a bad starting point for an organization that has never activated Breeze. In that second case, it is better to scope a single agent, measure it over a full cycle, then consider the hub as a scaling step rather than an entry point.
“The way to launch an agent with HubSpot is to start by asking about use cases: which types of requests are the most frequent, and the easiest to cover first. With a client like Zeop, the test without customization was a gold mine for identifying gaps before investing in the next steps. You start small, add a first use case, a first data source, then keep analyzing what the agent cannot do yet to prioritize the next step,” explains Camille, senior consultant at Markentive, about the deployment of the Breeze Customer Agent at Zeop.
Zeop, a telecom operator on Réunion Island, has been supported by Markentive on HubSpot for nearly five years. The company runs Marketing Hub, Sales Hub and Service Hub at a volume that places it among the largest HubSpot instances in France: more than 1.5 million tickets handled since 2020, with around 25,000 calls and 4,000 written interactions per month today.
Rather than aiming for a complete agent from the outset, Zeop and Markentive launched in summer 2025 a first version of the Breeze Customer Agent that was deliberately not customized, fed only by the FAQ and website content. This one-and-a-half-month test was enough to demonstrate a satisfactory level of answers and to justify the next step: connecting the agent, renamed “Zoé”, to CRM data (contract, latest invoice, fiber technician appointment), then to Zeop's ERP through automated actions for more advanced use cases: breaking down the details of an invoice, tracking usage, confirming an ongoing network incident (frequent on the island during cyclone season).
Launched in December 2025, Zoé has an autonomy rate of 93% (the agent answers on its own, without handoff), a resolution rate judged satisfactory by the customer of 77%, and a satisfaction rate of 80%, comparable to that of phone customer service. The chat now handles about 4,000 conversations per month, the equivalent of the total volume of written requests handled before the project, across all channels. During a traffic peak (a commercial campaign that multiplied the daily volume of conversations more than tenfold in one day), the agent absorbed the load without the capacity limit of a call center, which mechanically reduced the impact of the peak on phone service.
Two methodological points to remember beyond the figures: the Markentive and Zeop teams meet every two weeks to review the knowledge gaps the agent flags natively when it cannot answer, which continuously feeds the backlog of new use cases. And the connection to the ERP goes through dedicated, filtered API endpoints, so that the agent only has access to the data strictly necessary for each use case rather than to the whole system.